Book knowledge toolsSearch and discuss the encyclopediaSearch this language edition or ask Chat V2 a question grounded in the approved book passages.

Search this bookSearch words or a source code

Discuss with Chat V2Book-only mode is the default. Every supported answer must cite an actual indexed passage.From this book only

THE LAND OF ALGORITHMS

Artificial Intelligence and the Making of the Agricultural Future

From the Field to the Algorithm... and Back

Thirty-six chapters examine in depth what artificial intelligence can contribute to agriculture, what it cannot yet prove, and who bears responsibility for the distance between promise and outcome.

A global book issued in ten carefully selected languages, whose communities of first- and second-language speakers represent an indicative reach of roughly 60% of the world's population. It is written to combine scientific depth with practical clarity: accessible to farmers, useful to specialists, and attentive to differences among agricultural environments and cultures.

Author: Dr Aladdin Ali PhD in Agricultural Sciences — Founder of Aladdin Pertanian Internasional Co.

Sponsoring body: PT Aladdin Pertanian Internasional Language of this edition: English Edition: Expanded working draft — 2026 Date: 18 September 2026

Rights and Documentation Methodology

This book was not built on a single document or a closed collection of files or references. It emerged from a process of research, documentation, and review conducted in several stages: beginning with the collection of scientific, technical, and agricultural material in different languages, then expanding during writing and editing to include studies and scientific reviews, reports from international institutions, official and regulatory documents, extension guides, technical materials, and implementation and test records associated with some of the systems and applications discussed in the book.

Throughout this process, merely locating a source was never sufficient grounds for adopting it. The identity, date, version, and context of each source were reviewed, and claims were compared with what the evidence actually allowed. Where the origin of a piece of information could not be traced, or the evidence was insufficient, the statement was qualified, the figure withheld, or the limits of confidence made explicit, rather than presenting it to the reader as an established fact.

Artificial-intelligence tools were used for some tasks involving organisation and comparison, duplicate detection, tracking differences between versions, language assistance, and proposing initial formulations. These tools, however, were not treated as scientific sources and were given no authority to issue an agricultural judgement or validate a fact or recommendation. Source verification, assessment of the strength of evidence, editing, choice of wording, and final approval are all human acts, and responsibility for them rests entirely with the author.

Names of companies, products, institutions, and programmes appear in these pages for purposes of identification, analysis, and comparison, not endorsement or advertising. Mention of a product or system does not constitute a recommendation to purchase or use it, and its omission does not represent a judgement on its quality or suitability. Names, trademarks, and the rights associated with them remain the property of their respective owners.

This book guides the reader to the evidence and helps them understand and examine its limits, but it does not replace the original source, the current approved product label, the regulatory authority, the laboratory, or a qualified agricultural or veterinary professional.

Copyright

Without the prior written permission of the copyright holder or a person authorised by the copyright holder in writing, this book may not be reproduced, in whole or in part, copied, printed, republished, distributed, translated, recorded, adapted, used to prepare a derivative version, converted into any format or medium, stored in any information-retrieval system, or transmitted by any electronic, mechanical, or other means, including photocopying and recording.

This restriction does not apply to limited quotation permitted under applicable law for purposes of criticism, review, research, or education, provided that the book is properly cited, the context of the quoted material is preserved, and its meaning is neither distorted nor presented in a manner that implies the author's or copyright holder's endorsement of a person, institution, product, or opinion where no such endorsement has been expressly given.

Quoting any figure, recommendation, or sensitive information from this book does not exempt the user from the duty to consult the original source or the competent official authority, nor does it transfer to the author responsibility for applying the information outside its context, after it has ceased to be current, or after its regulatory conditions or scientific basis have changed.

First edition — 2026

Statement of Sponsorship and Editorial Independence

PT Aladdin Pertanian Internasional sponsors this work, and its eighth part presents a case study of the company's platform. This is a relationship that can corrupt an entire book if it passes in silence, and it is therefore declared here at the beginning rather than in a closing footnote.

A potential conflict of interest is not treated by denial, nor by calm scientific language that conceals it. It is treated by written limits against which the reader can hold the book to account. These limits are three:

Nothing is attributed to the platform except what the repository or a dated engineering report establishes.
The existence of code does not mean the service runs in production — between the two lies a distance that Chapter Thirty-Two explains.
The platform is not used as evidence of scientific or commercial superiority over any other system.

The sponsor holds no authority over the findings of this book. It cannot make the absence of evidence into evidence of quality, nor delete a limitation because it is commercially inconvenient. And when a marketing description conflicts with a primary source or with the state of the repository, the word belongs to the higher authority and to the more conservative description — always and without exception.

This is what the author can promise. Judging whether the promise was kept is left to you as you read the eighth part.

Agricultural Safety Notice

This is a book that teaches you how to think about the agricultural decision, not a book that dictates the decision itself.

It explains the logic, analyses the evidence, and assists in designing systems. But it is not an official registration record, nor a fertilisation or irrigation prescription for your particular land, nor a veterinary diagnosis, nor a substitute for a laboratory, an agricultural engineer, a veterinarian, or a regulatory authority.

The confidence that rises in the tone of algorithms can turn into a dangerous illusion once it sets foot on the ground of the field. There are red lines in agricultural decision-making that exceed the capacity of a language model however fluent it may be: the chemical dose, the two safety intervals — the pre-harvest interval (PHI) and the restricted-entry interval (REI) — the legal registration of the pesticide, and the price of the commodity today.

Certain sensitive facts must never be guessed by an artificial-intelligence system or presented without a clear source: the permitted dose; the interval between the final pesticide application and harvest; the restricted-entry interval for workers entering a treated area; the legal registration status of the product for the crop and pest concerned; and the commodity's current market price.

The dose, the pre-harvest and restricted-entry intervals, and the registration status must be checked against the current approved product label and the instructions of the competent regulatory authority in the country of use. An error in any of these may damage the crop, endanger workers, or leave non-compliant residues that jeopardise acceptance of a shipment in an export market. A price, in turn, must be checked against a current and reliable market source that identifies the place, date, currency, and quality grade; an old or non-local price may lead to a bad decision about selling, buying, or production.

Artificial intelligence can gather observations, order probabilities, retrieve evidence, and reveal a pattern worth examining. That is a great deal. But it cannot transfer legal and professional responsibility from the accountable decision-maker to the algorithm. Responsibility does not migrate through technical delegation.

For this reason, any system that deserves trust in sensitive agricultural work knows three things about itself: when to abstain, what it lacks in order to answer, and to whom it escalates the decision when its consequence exceeds the system's limits.

Author's Preface

I did not begin my journey with artificial intelligence as an expert in the field, or as a developer who already understood how its models and tools worked. I began in mid-2023 as a novice user trying to understand what this new technology could truly do, and whether it could genuinely change how we learn, work, and solve problems.

At first, my curiosity was practical and uncomplicated. I tried the available tools as any new user might: asking questions, comparing answers, and watching the remarkable speed with which they produced text, images, and analysis. The initial wonder did not stand alone for long. I began to notice that a confident answer could still be incomplete, that eloquence did not guarantee truth, and that tools with similar interfaces could differ radically in their sources, capabilities, and limits.

By late 2024, casual use was no longer enough. I began experimenting with a wide range of artificial-intelligence tools, engines, and platforms, studying their differences and comparing how they worked. How did they understand a question? How did they retrieve knowledge? When did they answer from a source, and when did they fill a gap from what they had learned before? How did they handle different languages? And what happened when they were asked for specialised guidance in a consequential field such as agriculture?

Agriculture was both the most demanding test and the one closest to me. I do not see it merely as an economic sector, but as a science, a profession, and a responsibility touching water, food, soil, plant and animal health, and the livelihoods of millions of families. An error here does not remain inside text on a screen. It may become an unsuitable dose, irrigation at the wrong time, a misleading diagnosis, or a cost that a farmer cannot afford.

In September 2025, my journey moved from experimentation and comparison to deeper study and construction. I began work on my largest project: developing an artificial-intelligence system specialised in agriculture, designed not merely to produce general answers but to connect an agricultural question to its source, context, crop, location, language, and the consequence of the decision.

The founding ambition was as clear as it was demanding: agricultural students should have access to this knowledge and these tools entirely free of charge, while a farmer should be able to access an intelligent agricultural service at a target cost of no more than one US dollar a month.

That figure was not a marketing device. It was an engineering and ethical question placed before the project from the outset. Can advanced agricultural technology be built without remaining confined to major universities, wealthy companies, and farms able to afford expensive subscriptions? Can the same knowledge reach a student learning with limited resources and a smallholder unable to risk a substantial share of income on a digital tool not yet tested in that setting?

I record this target here as the purpose that guided design and development, not as a claim that every component has already been achieved or that the target price is now a commercial offer available in every country. This book itself rests on the distinction between what we aspire to do, what we have built, what we have tested, and what has demonstrated an effect in the real world. An ambition does not lose its value when we acknowledge that it remains unfinished; it becomes more honest and more testable.

The idea for this book grew out of that project.

The work began with an apparently simple intention: to catalogue agricultural artificial-intelligence applications around the world and compare their uses, levels of maturity, and value to farmers, researchers, and decision-makers. I first imagined an orderly collection of applications, studies, and technologies. As the files multiplied, however, it became clear that the real problem was not a shortage of information, but the difficulty of distinguishing among its kinds.

An editorial truth emerged that was no less important than the book's subject: an abundance of text does not mean that knowledge is complete.

I saw attractive figures travel from report to article, from article to sales presentation, and then into another language, shedding their original source, limits, and conditions along the way. I saw an image-classification success rate presented as evidence that a robot succeeded in the field, a global market forecast substituted for farm economics, a prototype presented as a mature service, the presence of code read as proof of operation, and local operation portrayed as demonstrated impact on farmers' lives.

I also saw how translation can carry an error farther and give it greater authority. A weak figure does not become stronger when written in ten languages, and an unsupported claim does not become true because it appears on dozens of pages. It may instead become more dangerous, because linguistic reach gives it a false appearance of consensus and stability.

Two easy paths lay before me, and both led to loss.

The first was to delete everything that raised doubt, producing a small and severely cautious text that appeared sound but sacrificed the book's usefulness, breadth, and memory. The second was to gather everything as I found it, preserving the details, names, and figures but also passing their errors to the reader in a more polished and imposing form than the one in which they reached me.

I chose a third path, harder and slower: to recover useful detail while separating the idea from the claim, the existence of a programme from its effectiveness, a model's success in a test from a system's success on a farm, and a result observed in one case from its capacity to travel to another crop, country, or community.

I chose to retain a name when it helped map the field, but to suspend a figure when I could not close the trail back to its source; to describe a product only to the extent supported by its evidence, not to the extent promised by its marketing; and to leave a question open for research when knowledge was incomplete, rather than closing it through deletion or assent.

The pages that follow therefore name systems, studies, projects, and fields of use. Beside them, they identify the type of evidence available, the level of maturity, what the source allows us to say, what may not be inferred, and what remains unknown or in need of independent verification.

You will find fewer figures than you might expect in a book about artificial intelligence and agriculture. This is not because digital agriculture lacks measurement. It is because a number detached from its task, context, environment, and comparator may mislead more than it explains. An accuracy of 95% means little until we ask: accuracy at what task, on which data, under what conditions, and at what cost of error? Was it compared with a human or with a simpler method? Did it succeed outside the laboratory?

A number without context is not information; it is decoration.

You will also find chapters on cybersecurity, data rights, labour and the just transition, vendor lock-in, intermittent connectivity, cascading failures, the protection of local knowledge, and the ability to export data and leave a service. These subjects may seem distant from the immediate question of how artificial intelligence might increase production or improve a decision. In fact, they lie at its centre.

These conditions determine whether a system can continue to function when the technology fails, the connection is interrupted, the provider changes, the company closes, or the farmer discovers that the data are locked inside a platform that cannot be left. Agricultural systems are tested not only on ordinary days, but at the moment when one part fails while land, plants, and animals still require a decision.

Because knowledge does not become accessible merely by existing, the choice of the book's languages was not a secondary publishing detail. This work has been prepared in ten languages selected to reach communities with widely differing cultures, agricultural experience, levels of urbanisation, and scientific, digital, and economic development. As first or second languages, they represent an indicative linguistic reach equivalent to roughly 60% of the world's population, with due allowance for the overlap among multilingual speakers.

The purpose, however, was not simply to pursue the largest percentage. These ten languages connect universities, laboratories, and cities with advanced digital infrastructure to villages and farming regions with limited connectivity; they span major industrial economies and communities where access to agricultural extension remains a daily challenge. They also carry scientific and cultural traditions, local terms, and inherited knowledge that must not be flattened into a literal translation or a single standardised vocabulary.

I wanted the book to speak to the agricultural student entering this field for the first time; the farmer seeking an intelligible answer rather than a technical display; the engineer who needs to understand the system's limits; the researcher asking about the evidence; the developer building the tool; and the decision-maker who will bear responsibility for approving or regulating it.

I therefore do not treat multilingualism as copying the same words into ten interfaces, but as a test of meaning and justice. A system that works well in one language may fail in another. A grammatically correct translation may be agriculturally wrong. The name of a plant, disease, or implement may preserve a distinction in local knowledge that the standard term does not. If technology does not respect these differences, it does not disseminate knowledge; it disseminates an incomplete version of it on a wider scale.

Nor does this book claim to close the debate or offer a final judgement on a field changing at such speed. It establishes a foundation open to criticism, correction, and revision, and asks one thing of its reader: always distinguish what has been established, what has appeared in an early study, what an institution has announced about a programme, what a vendor has said about a product, what a software test has demonstrated, and what I propose as a design or a better path.

The existence of an idea does not prove its success. The existence of a programme does not prove that it operates; operation does not prove safety; and technical safety does not prove that it improved a decision, saved water, raised income, or reduced loss. Each level requires its own evidence, and every claim has a boundary that language should not cross.

At its heart, this book is the record of a journey that began in mid-2023 with a novice user's question: what can artificial intelligence do? Through experimentation, study, and construction, that became a harder and more important question: what should it do, for whom, at what cost, under whose oversight, and who bears the consequence when it is wrong?

The project I began in September 2025 still carries its first purpose: to bring intelligent agricultural knowledge closer to those who need it, not only to those who can afford it; to give agricultural students free access to the tools of the future; and to give farmers a service they can afford, understand, scrutinise, and trust without surrendering control of either their decisions or their data.

Anyone who can distinguish evidence from promise, and knowledge from ornament, is less likely to be misled by a brilliant headline, a large number without a source, or a system that speaks with more confidence than its knowledge warrants. Put that capacity in the hands of students and farmers, and you give them more than a technical tool: you give them a fairer right to participate in shaping the agricultural future.

Foreword

The field is not a set of numbers on a screen. It is land waiting for rain, a plant seeking water through its roots, an animal whose health changes with feed and heat, and a family whose income may depend on a decision made in a single hour. Soil, weather, water, pests, labour, and money work together there; a change in one passes into the others, and a short delay in irrigation or an error in diagnosis can become the loss of an entire season.

An agricultural decision is therefore not simply a choice between two answers, nor can a clear image, an instrument reading, or a computer forecast make it sound on its own. A good decision understands the land on which it will be acted upon, brings measurement together with the farmer's experience, weighs cost and consequence, and remains open to revision if the weather changes or new information appears.

This is the reality from which our discussion of artificial intelligence begins: tools able to read large volumes of images, measurements, and records, and to search them for patterns that may help people understand and anticipate. Yet they do not see the field as those who live in it do, and they do not bear the outcome of what they propose. Their value should therefore be measured not by the elegance of the answer on the screen, but by what happens after it is acted upon. Did the water arrive at the right time? Was the diagnosis sound? Was loss reduced? Did the decision remain understandable, safe, and in human hands?

An agricultural decision may look simple when reduced to a short question: When should I irrigate? What is the disease? Which crop should I plant? Should I sell now or wait? Behind every question lies a network of conditions invisible in its words. Soil moisture cannot be read apart from texture, salinity, rooting depth, and growth stage. A clear image of an affected leaf cannot by itself reveal the cause of the symptom or its extent across the field. A rainfall forecast does not decide irrigation without knowledge of the soil's water-holding capacity. Nor does a forecast price increase erase storage costs, the risk of spoilage, or a family's need for cash this week.

The question itself may also change meaning from one community to another. The phrase “the soil is dry” may describe an instrument reading on a highly mechanised farm, a farmer's judgement after handling the soil, or a pastoralist's observation of animal and plant behaviour. The name of a crop, disease, or growth stage may differ between village and city, between the official language of extension and the language of the field, and between a scientific term and a name preserved across generations. Smart agriculture therefore begins not with the collection of numbers alone, but with an understanding of the context that gives those numbers meaning.

From this arises the dual nature of artificial intelligence in agriculture: immensely useful and immensely dangerous, for the same reason.

Its value lies in following what a person cannot gather and analyse manually in time: sequences of satellite, drone, and phone images; sensor readings; seasonal records; weather forecasts; soil analyses; price movements; and knowledge published in many languages. It can detect a pattern extending across thousands of fields, relate a small change in a plant to a long history of weather and irrigation, and return to the user information buried in a report, table, or extension bulletin.

Its danger arises from the same capacity to combine and condense. An algorithm can fuse signals of unequal quality into one elegant, persuasive answer whose fluency conceals missing data, a lost unit, an obsolete source, an image insufficient for diagnosis, or an experiment conducted on another crop in another region. The answer may sound most confident precisely where reality is most uncertain, because polished language does not always reveal the conditions and exceptions removed from it.

One distinction should guide every reader: linguistic fluency is not agricultural accuracy, and the confidence displayed by a model is not necessarily a scientifically calibrated probability. A sentence may be polished and persuasive yet wrong in substance. A forecast may be correct on average but unfit for the field in question. An algorithm may identify a symptom in an image and still fail to determine a safe course of action. Between naming a problem and knowing what should be done about it lies a space filled by measurement, experience, regulation, and responsibility.

This book therefore begins from a practical rather than ideological position. It does not ask whether we are for or against artificial intelligence, and it does not assume that everything new is better or everything traditional is backward. The more useful questions are these: Where does the technology add measurable value? What data does it need? What are the limits of its validity? Who reviews it? How does it remain useful, safe, fair, and durable when conditions change, connectivity fails, or the model is wrong?

The narrative is not built around the most dazzling figures or demonstrations that make a farm appear to run at the press of a button. It follows the complete decision chain, link by link. Where did the data come from? Who owns them? How were they measured and standardised? What was lost in transfer? What did the model predict, and with how much uncertainty? Who turned the prediction into action? Against what alternative was it compared? What then happened to water, soil, yield, returns, labour, and risk?

The book also asks questions that rarely receive equal space in technical presentations. Did the farmer understand the recommendation as the system intended it? Could the farmer contest it and correct its data? Was its source, date, and scope visible? Did authority remain with the farmer, or pass silently to a platform that could not be inspected? If the farmer chose to leave the service, could the data, records, and their meaning leave too, or would they remain captive to a closed format and a contract that offered no genuine exit?

Where any link in this chain breaks, model accuracy cannot rescue the system as a whole. A prediction may be correct but arrive too late; arrive on time but remain misunderstood; be understood yet prove economically impossible to act upon; or be acted upon without its outcome ever being measured. Smart agriculture must therefore be evaluated at the boundary of the decision and its real-world result, not merely at the boundary of the model.

This book extends across ten carefully selected languages: Arabic, English, Chinese, Spanish, French, Indonesian, Malay, Urdu, Persian, and Turkish. Aggregated estimates of their first- and second-language speakers suggest a potential linguistic reach equivalent to roughly 60% of the world's population. This is a statement about the possible breadth of the linguistic sphere, not a guaranteed readership, and it must be read with care because multilingual speakers overlap and cannot be counted more than once when estimating actual reach.

The importance of these languages does not lie in numbers alone. They cross continents, climate zones, and profoundly different agricultural systems. They connect rural and urban communities, advanced industrial and digital economies, rapidly transforming countries, and regions where direct agricultural extension and digital infrastructure remain limited. They are spoken by smallholders, pastoralists, agricultural workers, researchers, engineers, entrepreneurs, students, policymakers, and city residents whose food security depends on fields they may never see.

They also carry different traditions of knowledge: from research institutes, laboratories, and satellite networks to local experience that recognises soil by colour and smell, reads the season in wind and animal movement, and preserves the names of plants, diseases, and growth stages in words for which a standard dictionary may have no exact equivalent. Translation in this book is therefore not treated as the substitution of one set of words for another, but as a responsible movement of meaning among different scientific, cultural, and professional settings.

The choice of languages does not claim that these ten represent all humanity, nor does it grant them epistemic superiority over thousands of others. It acknowledges that any multilingual project remains incomplete unless it respects what lies beyond its scope. The aim is to build a broad and extensible bridge, not a new linguistic centre that asks everyone else to move towards it. Genuine multilingualism does not mean making the system repeat the same sentence in ten languages. It means understanding differences of context, preserving local terms, testing comprehension rather than translation alone, and allowing users to correct meanings the machine has failed to capture.

A platform does not become equitable merely because its interface has been translated. Its menus may appear in the user's language while its knowledge sources, training data, quality tests, and safety rules remain rooted in one language or environment. Speech recognition may work well in a high-resource language and then mishear the name of a pesticide, disease, or unit in a lower-resource language. A translation may be grammatically sound yet agriculturally wrong, replacing a precise local term with a general word that erases the distinction on which the decision depends.

Nor can differences among communities be reduced to a binary of “advanced” and “backward”. A farm may use satellites and self-guided machinery yet suffer poor interoperability or vendor lock-in. A holding may be small and intermittently connected yet possess precise local knowledge and an effective social system for sharing warnings and experience. A city may have advanced laboratories and networks while receiving its food through a fragile chain that cannot see the worker's or farmer's conditions at its beginning. Technical advancement is an important capability, but it is no substitute for fitness, equity, trust, or maintainability.

The book's thirty-six chapters range across computer vision, remote sensing, data systems, yield forecasting, land and crop selection, smart irrigation, soil and nutrient management, disease and pest diagnosis, robotics, protected cultivation, and crop rotation; then harvest and post-harvest operations, food safety, and value recovery; followed by livestock production, aquaculture, and the integrated farm; and on to economics, markets, supply chains, inclusion, agricultural labour, the just transition, and governance. Yet they do not separate technology from the society that uses it, the model from the institution that operates it, or efficiency from the ethical and legal questions that accompany it.

An application that succeeds on a farm connected to a laboratory and a stable network does not migrate of its own accord to a smallholding whose members share a phone and rely on intermittent connectivity. A system that predicts accurately in one country gains no automatic authority in another where varieties, soils, climate, practices, and law differ. Nor does a technology that saves water in a controlled trial necessarily deliver sustainability if it raises costs, marginalises workers' knowledge, or expands irrigated area until total consumption increases.

Through this scientific and linguistic breadth, the book seeks to serve two purposes: to give farmers, engineers, researchers, extension professionals, and decision-makers a clear map of what technology can do; and to give developers, investors, and service providers an explicit account of what may not be claimed. It does not promise one answer for every field. It offers a more rigorous way to frame the question, examine the evidence, define the system's limits, and connect innovation to its practical consequences.

Agriculture needs more than fewer promises. It needs more precise engineering, more trustworthy data, translation that preserves meaning, clearer evidence, institutions capable of review and correction, and responsibility that does not dissipate along the path between those who collected the data, built the model, sold the platform, issued the recommendation, and ultimately bear the outcome of the decision.

When technology is built on this foundation, the algorithm does not come to replace the land or the people who know it, but to help them see it more clearly. Scientific knowledge, local experience, and computational capacity can then take part in one conversation: beginning in the field, passing through data and the model, and returning to the field with greater humility, greater auditability, and a sharper awareness that the agricultural future will not be made by the algorithm alone, but by people who know how to use it, when to trust it, and when to ask it to stop.

The Executive View: From Technical Promise to Real Impact in the Field

The core question this book poses is not what artificial intelligence is capable of doing in agriculture; its capabilities are extensive, continually renewed, and expanding with each technical advance. The harder and more valuable question is this: what actually changes in the field when the system enters it? Between technical capability and agricultural impact stretches a decisive distance, traversed by data, decisions, institutions, and economic considerations. At every waypoint along this journey, value may erode or be lost entirely before it reaches the farmer.

It is precisely from this distance that the book's central argument emerges, expressed through six central conclusions:

First: value lies in the decision, not in the model alone. A model may achieve excellent statistical performance without changing a single decision at the right moment. Conversely, a simple model — if grounded in reliable measurement, tied to clear and comprehensible guidance, and backed by a manual fallback — may be of greater value than a complex system. The decisive criterion is not the degree of technical complexity, but the system's capacity to convert knowledge into an actionable decision at the moment when that decision makes a genuine difference.

Second: agricultural data are not neutral raw material. Behind every record lies an owner, a context, a unit, a date, and a method of measurement. Consequently, the loss of provenance, the conflation of zero with a missing value, or the treatment of two local names for a single crop as two distinct entities may lead to a wrong decision that appears — on its surface — precise.

The quality of a decision begins before the model is built; it begins with understanding the origin of the data, its meaning, its limits, and the context in which it was generated.

Third: laboratory results do not cross into the field on their own. Moving across seasons, regions, varieties, devices, and languages is not an automatic step but a process requiring external validation and continuous monitoring for drift. Image classification alone does not amount to a full diagnosis, simulation is not equivalent to live deployment, and model interpretation does not constitute causal proof.

Between experimental success and field adoption lies a more rigorous test: does the system remain reliable when the environment it was designed for changes?

Fourth: economic viability and equity are fundamental criteria for the success of a technology. A technical system's success is measured not only by its accuracy or its ability to complete tasks, but also by its economic value and its effects on the people who use it. Evaluation must therefore encompass all associated costs, including the costs of ownership, maintenance, connectivity, and training, and the time workers spend operating it, as well as the farmer's right to stop using the platform without becoming beholden to it.

Automation may raise productivity in certain tasks, but it may at the same time alter the distribution of roles, workloads, and risks among workers. Technological transformation should therefore not rest on the assumption that machines will replace humans, but on preparing and qualifying workers for a new agricultural economy in which skills evolve, responsibilities shift, and human judgement remains a pivotal element in decision-making.

For that reason, the book does not treat labour merely as a cost item in a feasibility study. Chapter Twenty-Four distinguishes a task from a job and a job from a livelihood. It examines the effects of automation on workers, families, and rural services before financing or expansion, and then sets out paid training, transitional protection, a right to contest decisions, and measures that bring production, income, safety, and equity together. A transformation that preserves produce and reduces its cost, yet makes the least protected groups pay through lost income without an alternative or a voice, has not achieved economic or social success.

Fifth: governance is not a secondary appendage that comes after the technology. Governance in this context means the set of rules, authorities, responsibilities, and controls that regulate the use of a technical system throughout its operating life cycle. It determines who owns the data, who has the right to view or use it, who approves the system's recommendations, when it must be halted or a decision referred to a human specialist, and who bears responsibility when an error occurs.

In this sense, governance is not an administrative procedure appended once the technology is complete; it constitutes an intrinsic part of its design and operation. Cybersecurity is not confined to protecting data and devices from intrusion; it also encompasses ensuring continuity of operation and the soundness of the decisions and procedures on which the farmer relies in managing the field.

Governance also extends to protecting the farmer's independence in relation to the technology platform. The farmer should be able to export data and transfer it to another system, or cease using the platform, without losing the data or becoming permanently dependent on a single technology provider. In this way, data portability and freedom of exit are transformed from formal technical features into practical guarantees that protect the farmer's right to choose.

The system should likewise know the limits of its own capability: refraining from issuing a recommendation when the available information is insufficient, referring the case to a human official when higher expertise or authority is required, and rolling back any action shown to be unsafe or inappropriate. These capabilities are not marginal add-ons but essential conditions for the system's reliability and sound performance.

An illustrative example: an intelligent irrigation management system

Suppose an intelligent system recommends to a farmer the timing and quantity of irrigation on the basis of soil moisture readings, temperature, and data related to field conditions. If soil moisture readings are missing, or if the measuring devices show conflicting results, the system should not issue a definitive recommendation as though the information were complete.

In such a case, governance specifies the course of action in advance: the system refrains from issuing a confident recommendation, makes clear to the user that the data are insufficient or inconsistent, and then refers the case to the person responsible for review before irrigation is carried out. Governance also determines who holds the authority to approve or halt the decision, how the intervention is documented, and who bears responsibility for the final action.

In the absence of such rules, the system may issue a recommendation built on incomplete information, without it being clear who ought to have reviewed it, who holds the authority to halt execution, or who bears responsibility for the outcomes.

Human oversight, therefore, is not achieved merely by announcing that there is a “human in the loop”. It requires identifying the person authorised to intervene, the role that person performs, the decisions that person is empowered to make, and the responsibility that person bears. Governance is the framework that makes technology safe, disciplined, and accountable, and that preserves the farmer's control over their data and decisions.

Sixth: there is no single regional pathway. Turkey, Indonesia, South America, the Arab world, the Gulf, Africa, and South Asia differ in water resources, landholding structure, cropping patterns, institutional infrastructure, and levels of connectivity. These environments cannot therefore be reduced to a single technical prescription or a uniform development pathway.

The function of a global framework is not to erase regional differences, but to organise the questions they reveal, and to turn diversity from an obstacle to implementation into a foundation for design that is more realistic, more equitable, and more effective.

Thus this book measures the success of artificial intelligence not by what a model can predict inside the laboratory, but by what an integrated system can change in the field: a better decision, fewer risks, clearer value, and a greater capacity to endure without sacrificing accuracy, equity, or the farmer's independence.

How to Use This Book

This book need not be read from beginning to end. Each reader has a pathway that begins with their own question and then leads back to method, safety, and evidence:

If you are new to the field, begin with Chapters One and Two; they contain the book's language and method, and are the best entry point for understanding the remaining chapters.
If you are a farmer or an agricultural engineer, begin with the Agricultural Safety Notice, then enter through the applied chapter that concerns your decision — land, irrigation, fertilisation, disease, harvest, or post-harvest handling. If your question concerns animal care, pond management, or the exchange of materials and nutrients within an integrated farm, begin with Chapters Eighteen and Nineteen. Read Chapter Twenty-Seven before converting any analysis into a sensitive decision; it shows where the system's function ends, who holds final approval authority, and when it is necessary to consult a specialist, the current approved product label, or a regulatory body.
If you are a researcher, begin with Chapter Three, then review the register of case studies in Appendix A, the verified references, and the provenance and scientific-quarantine registers in Appendices G and J. These sections offer you a methodological entry point for distinguishing a claim from its evidence, and for understanding what is comparable, what remains bound to its context, and what has not yet met the conditions for citation or generalisation.
If you are a buyer or an investor, read Chapters 20, 21, 28, and 30: the first for total cost and the investment decision, the second for market, financing, and viability, the third for data rights, export, and exit, and the fourth for maturity and operational readiness. If the investment will alter tasks, working days, or income in the region, add Chapter 24 before approving the business case. Then review the investor map in Chapter 35 and the reference directory of implemented programmes in Appendix O; inclusion there establishes the programme's identity and its implementation within the stated bounds, and does not constitute an endorsement or proof of impact.
If you are a policymaker, Chapters 23–30 and Chapter 35 present the elements of a programme for responsible digital agricultural transformation: access and extension, a just transition for agricultural labour, regional infrastructure, safety, data rights, verification, and institutional readiness. They do not constitute a complete national plan until they are tailored to context; purchasing equipment is a relatively easier step than building data, skills, social protection, governance, and maintenance, and it may remain limited in impact if detached from them.
If you are looking for a particular implemented programme or system, begin with Appendix O; it is confined to programmes and systems whose identity and implementation could be verified up to 31 August 2026. Then return to Appendix H and the corresponding applied chapter to understand the context and constraints. A programme's inclusion in the directory does not mean it is endorsed, nor does it prove its effectiveness, safety, or suitability for every context.
Read the Aladdin Agri AI study in Chapters 31–33 as a sponsor-linked case study that declares its conflict of interest, the nature of its internal evidence, and its limits — not as an independent evaluation, a purchase recommendation, or proof of agricultural impact. The opening of that part states what evidence the case lacks before presenting what the package does establish, and Appendix N completes the register of implementation evidence and the conditions for upgrading it.

Reference Map for Recurring Principles

The book develops each recurring principle fully in one home chapter. Later chapters recall that principle briefly and then explain only what changes in the new agricultural domain. This map helps readers follow those callbacks without mistaking a concise reminder for a new or weaker rule.

Recurring principle	Home chapter	How later chapters use it
Frame the decision before selecting technology	1	Recall the rule briefly, then apply it to the domain decision.
Transformation, distribution of benefit, and farmer agency	2	State who benefits and who decides without rebuilding the whole framework.
Evidence rank, claim limits, and accuracy versus impact	3	Name the evidence type and the case-specific limit.
AI method families and their outputs	4	Mention only the method needed to understand the decision.
Measurement, missingness, units, and provenance	5	State what was measured, calculated, or estimated and preserve the domain record.
Spatial validity and model transfer	6	State the scale, region, and relevant transfer gap.
Retrieval, grounding, and responsible abstention	7	Give the reason for abstention and the next evidence needed.
Land and crop suitability	8	Keep soil, land, field, and farm distinct without redefining them unless their meaning changes.
Monitoring and forecast revision	9	Preserve dated forecast versions and compare them with the measured outcome.
Command versus execution	10	Apply the distinction to pumps, machines, and facilities without re-explaining it.
Nutrient pathway and crop response	11	Refer here for fertigation and nutrient-decision logic.
Observation, diagnosis, confirmation, and authority	12	Use concise callbacks in later plant-health and action chapters.
Physical actuation and functional safety	13	Apply movement and energy limits when later systems act near crops or people.
Production-system taxonomy and levels of automation	14	Use this classification rather than rebuilding it elsewhere.
Multi-season planning, constraints, and trade-offs	15	Use this chapter as the temporal-planning reference in later parts.

The rule is practical: a later chapter may recall a home principle in one or two sentences, but it should spend its explanatory space on the new crop, system, consequence, or implementation condition.

Key Abbreviations

Abbreviation	English Term	Rendering Used in This English Edition
AI	Artificial Intelligence	artificial intelligence
ML	Machine Learning	machine learning
DL	Deep Learning	deep learning
CNN	Convolutional Neural Network	convolutional neural network
LLM	Large Language Model	large language model
RAG	Retrieval-Augmented Generation	retrieval-augmented generation
XAI	Explainable Artificial Intelligence	explainable artificial intelligence
IoT	Internet of Things	internet of things
FMIS	Farm Management Information System	farm management information system
GIS	Geographic Information System	geographic information system
UAV	Unmanned Aerial Vehicle	unmanned aerial vehicle
ET	Evapotranspiration	evapotranspiration
IPM	Integrated Pest Management	integrated pest management
PHI	Pre-Harvest Interval	pre-harvest interval
REI	Restricted-Entry Interval	restricted-entry interval
TCO	Total Cost of Ownership	total cost of ownership
MLOps	Machine Learning Operations	machine learning model operations
API	Application Programming Interface	application programming interface
R²	Coefficient of Determination	coefficient of determination — the proportion of variance in the target variable explained by the model within its own data
BLEU	Bilingual Evaluation Understudy	a measure of the similarity between a generated text and a reference text; it does not measure agricultural correctness
DQL	Deep Q-Learning	deep reinforcement learning for selecting sequential actions
AS-MO	Automated Scheduling with Manual Operation	automated scheduling with manual operation (hybrid irrigation interface)

Map of the Parts

One Argument in Ten Movements

The parts of this book are not arranged as ten volumes set side by side, nor do they follow one another as a list of separate AI applications. They build a single argument in ten interlinked movements: first defining the agricultural problem worth solving, then opening up the system from within and taking it down to the crop and the land; following the output through harvest and post-harvest; and then moving to the animal, the water, and the integrated farm. Once the production cycle is complete, the book places technology on the scales of economics and the value chain, asks whom it includes and excludes across different settings, tests its responsibility when error and failure occur, examines a declared applied case within the limits of its evidence, and returns in the tenth movement with what remains fit for decision, action, and research.

Part One — Why Does Agricultural Transformation Matter?

The governing question: what problem is artificial intelligence supposed to solve in agriculture? This part does not begin with the algorithm, but with agriculture operating under compound pressure: climate, water, soil, labour, prices, financing, and disparities in landholdings and access. It rejects the easy assumption that introducing an application or a sensor means, in itself, that a transformation has occurred; digitisation may change the tool of work without changing the decision, the outcome, or the distribution of benefit. Its three chapters lay the foundation through which the entire book will be read: what does agricultural transformation mean? What are the differences among productivity, efficiency, sustainability, and resilience? And how do we weigh evidence rather than being swayed by the force of a phrase? This part therefore offers a technical answer less than it calibrates the question that ought to precede the technology:

What decision do we want to improve?
And for whom?
At what cost and what risk?
And how do we know that the outcome has genuinely changed?

The governing conclusion: the value of artificial intelligence does not begin with what it is able to do, but with the problem for which it deserves to be used.

Part Two — Foundations of Artificial Intelligence and Data

Governing question: What happens inside the system before the answer reaches the farmer? This part opens the black box without turning the reader into a machine-learning engineer. It explains the families of methods — from vision, prediction, and optimisation to language models and retrieval — then moves to the material these methods learn from: data, its units, its sources, the rights of those who own it, its errors, and what it loses in the course of conversion. From data it moves to place and sensing, then to decision-support systems, knowledge retrieval, and extension. This ordering is deliberate; no answer is better than its sources, no model ultimately surpasses the quality of its data and the precision with which its task is defined, and retrieval cannot correct knowledge that is outdated or that lacks authority. The function of this part is to equip the reader with practical questions to put to any system:

What did it see?
What is the source of that knowledge?
How might it fail?
When should it refrain from answering or acting?

Governing conclusion: Every intelligent answer carries within it the history of its data and the limits of the task it was designed for.

Part Three — Crop and Land Systems

Governing question: Where does technology actually touch the plant, and where does it stop? Here the book descends from theoretical foundations to the decision cycle in the field. The chapters of this part follow the journey from land selection and crop suitability, to growth monitoring and yield forecasting, then irrigation and water management, nutrients and fertilisation, disease and pest diagnosis, weed detection and field robots, modern farming methods and artificial intelligence, and finally crop rotation and production planning across seasons. What binds these domains together is not merely an abundance of applications, but a single rule that recurs at every stage:

Perception is not diagnosis, diagnosis is not a prescription, and a prescription is not a decision.

An image may reveal a pattern without knowing its cause. A map may predict stress without knowing the farmer's capacity to act. A robot may target a weed without thereby establishing that the intervention is safe or worthwhile. This part examines the distance between what technology can measure and what still requires local agronomic judgement. It keeps the responsible human decision-maker at the centre — not as a fallback summoned when the machine fails, but as the source of the authority that gives measurement its meaning and action its legitimacy. Governing conclusion: Technology can widen the field of vision, but it does not eliminate the need for the agronomic judgement that turns vision into a responsible decision.

Part Four — Harvest and Post-Harvest

Governing question: How do we preserve what the land has produced as the crop moves from the moment of harvest into storage, processing, and handling? The agricultural decision does not end when fruit or grain is separated from the plant. Harvest timing, handling method, impact, temperature, humidity, delay, cleaning, drying, grading, packaging, and storage may preserve the season's value or dissipate part of it before the product reaches the market. Artificial intelligence adds the capacity to estimate maturity, coordinate machinery and labour, monitor lot conditions, predict spoilage and remaining shelf life, and detect defects and hazards; but it does not turn every reading into a food-safety judgement, nor does it make a digital record correct when the input, sample, or sensor is wrong. The harvest chapter follows the chain from vision to contact, detachment, and delivery. The post-harvest chapter then follows the product through quantitative and qualitative loss, quality and safety, the cold chain, traceability, and the recovery of value from by-products and residues. Residues are managed according to an order that first prevents them, then recovers edible food, and then tests what may be suitable for feed, fertiliser, energy, or materials, so that “circularity” does not become a route by which contaminants return to the farm. Governing conclusion: Harvest is not the end of production; it is the beginning of a new test of value. Post-harvest is not a logistical corridor, but a decision system that may preserve the season or lose it.

Part Five — Livestock and Aquatic Production and the Integrated Farm

Governing question: How can technology extend care and regulate the rhythm of exchange among crop, animal, and water without reducing a living being to a data point or exceeding competent authority? Here the book moves from the harvested product to the living being that needs water, air, feed, observation, and response. The chapters address livestock production and animal welfare, then aquaculture, and place the integrated farm at the heart of their relationship with the crop. An identified, treated, and tested by-product may enter a safe and authorised feed pathway, while organic nutrients or water from an aquatic system may return to the crop under a plan, measurement, and regulatory limits. Yet every link may also carry a pathogen, contaminant, salt, or accumulated error. Reuse therefore does not mean safety, and a model has no authority to approve feed, water, treatment, or fertilisation. The role of artificial intelligence is to maintain the balance of material, water, nutrients, and energy; synchronise harvest, storage, feeding, irrigation, and ventilation; detect deviation; compare scenarios; and show how a decision in one link affects the rest of the cycle. Care, veterinary medicine, and the safety of feed, water, and the environment remain under named human, professional, and regulatory authority. Governing conclusion: An integrated farm is not a collection of adjacent activities. It is a cycle of benefits and responsibilities that becomes intelligent only when every material remains traceable, every risk can be stopped, and every sensitive decision remains with the competent decision-maker.

Part Six — Economics and the Value Chain

Governing question: Who pays for the technology, and who reaps its value? A system may work technically and fail economically. Productivity may rise while the gain flows to the supplier or the buyer, leaving the farmer with the subscription, the debt, the maintenance, and the risks of downtime. This part therefore places technology on a different scale; it asks not only: Does it work? but also: Is it worth it? And worth it for whom? Its chapters address farm economics and the investment decision; markets, finance, credit, insurance, and feasibility studies; then logistics, traceability, and supply-chain resilience. The analysis moves:

from purchase price to total cost of ownership;
from average return to cash flow and risk;
from the sum of benefits to how they are distributed among the parties.

This part also tests the value of information itself. A forecast acquires its value not by accuracy alone, but when it arrives while the user can still change the decision, when workable alternatives are available, and when the price of that knowledge is not the loss of one's data or of one's ability to exit should the supplier change.

Governing conclusion: Value is realised not merely when a system produces more accurate information, but when that information gives its user a genuine and equitable capacity to make a better decision.

Part Seven — Inclusion and Regional Realities

Governing question: For whom are these systems designed, and how do work and income change when they enter the field and the rural community? This part challenges the presumed agricultural user embedded in many products: a large farm, a modern phone, a stable connection, a standard language, and resources that permit trial and error. Agricultural reality, however, includes smallholdings, shared phones, intermittent networks, local knowledge, and risks whose bearers cannot survive a single failed season. Its chapters address smallholders and agricultural extension; then agricultural labour and the just transition; design for constrained resources and intermittent connectivity; and regional pathways for transformation in contexts that differ in water, energy, institutions, language, and holding size. It does not assume that efficiency gains distribute themselves. It asks who gained time and income, who lost working days, who gained access to training and a new job, and who bore the cost of transition without a voice or protection. The governing idea is clear: inclusion is not a diluted version of a product designed for a different user. Translation alone does not achieve equity, offline operation is not the same as stripping out context, and what succeeds on a large farm does not transfer automatically to a small cooperative, an archipelago, or a severely water-scarce region. When the region changes, it is not only the interface that changes; the problem itself changes, and with it the resources, the constraints, the labour market, and the meaning of success. Governing conclusion: Inclusive design does not begin by widening access to a predefined solution; it begins by redesigning that solution around the conditions in which its intended users actually work.

Part Eight — Responsible Agricultural Artificial Intelligence

Governing question: What happens when the system errs, and who bears the outcome? This part assumes from the outset that things will not always go as planned. The network may go down, the data may be corrupted, a disease may appear outside the training categories, the law may change, a control system may be breached, or the model may deliver a confident answer where it should have refrained. Its chapters examine agricultural safety and human authority, farmer data rights and governance, bias, uncertainty, explanation and validation, and then the transition from prototype to a dependable service with monitoring, support, rollback, security, and a responsible owner. This part does not present ethics as an addition appended to engineering once engineering is complete; it holds instead that governance is a performance property:

A system that does not know when to refrain is less fit for use.
A platform that cannot be left costs more than it advertises.
A decision that no one can contest lacks an orderly path for learning from its own error.

Governing conclusion: A responsible system is not one that never errs, but one that anticipates error, limits its effects, discloses it, allows it to be contested, and specifies who bears the duty to remedy it.

Part Nine — Who Gets the Algorithms? Aladdin Agri AI’s Bet on Equitable Access

Governing question: Can agricultural AI become a practical right for small-scale farmers and agricultural students, rather than a capability reserved for organisations with money, data, infrastructure, and specialist staff? This part begins before model accuracy, at the field gate. Many agricultural AI systems are priced, worded, and supported for large enterprises; a farmer with an intermittent connection, a low-cost phone, and a question in locally used agricultural language may never reach the intelligence inside them. Agricultural students face a related barrier when useful tools and knowledge remain behind subscriptions or presume technical resources they have not yet been given. Exclusion at this point is not merely a software inconvenience: it shapes who can make a safer decision today and who can learn to build and challenge agricultural technology tomorrow.

Aladdin Agri AI is examined here because it was founded around three declared access commitments: a target price of no more than one US dollar per month for a farmer, entirely free access for agricultural students, and support for ten languages intended collectively to reach communities representing about 60 per cent of the world’s population. These are design and test constraints, not findings of affordability, reach, completed localisation, or field benefit. A low headline price does not establish a low total cost; a language selector does not establish comprehension; and free registration does not establish that a student can complete meaningful work.

The case is connected directly to the book’s author and sponsor. That proximity opens design records and failures that a product page would not reveal, but it cannot make the sponsor’s testimony independent. The chapters therefore separate purpose, architecture, implementation, release readiness, completion of a user journey, and measured agricultural or educational outcome. Accessible external references support the governing principles—transparency, safety, accountability, provenance, localisation, data quality, and pesticide governance—while claims about Aladdin itself remain bounded by the private implementation evidence recorded as IMP01–IMP14 and by the limits stated for each item.

Chapter Thirty-One turns the promise of equitable access into questions that can be tested and declares who is speaking. Chapter Thirty-Two follows a field question through the service, asking whether a farmer or student can understand, act, recover, and contest rather than merely open a screen. Chapter Thirty-Three treats failures as evidence for repair: it orders incidents by their consequence for the crop, worker, environment, knowledge, and user rights, then returns each proposed remedy to a verifiable test.

Governing conclusion: Equitable access is achieved only when the intended user can reach the service, understand it, complete a safe and useful task, and obtain support or redress at a sustainable total cost. The existence of a component is evidence at one level; independent user and field outcomes belong to another.

Part Ten — Conclusions and Action

Governing question: After all of the above, what do we really know, and what do we do with it tomorrow? Here the book returns to the promise with which it began: not to settle for displaying possibilities, but to distinguish what the evidence supports, what it supports within a limited context, and what it does not support however often it is repeated in marketing or public discourse. Chapter Thirty-Four assembles the book's conclusions and confidence levels, and states explicitly where the evidence is weak and what may not be generalised. Chapter Thirty-Five then converts principles into roadmaps for farmers, cooperatives, businesses, developers, extension agents, policymakers, researchers, investors, and suppliers. Chapter Thirty-Six closes the circle with a research agenda for what remains unsettled, and with conditions for learning, for stopping, and for not investing when no real value accumulates. This part is thus not a mere summary of what preceded it; it is the point of passage from knowledge to responsibility:

What decision can be taken now?
What needs to be trialed?
What needs regulation or an institution?
What must not be purchased or deployed before stronger evidence appears?

Governing conclusion: Knowledge is not complete when it describes what is possible, but when it determines what may be done, what should be tested, and what must be deferred or refused.

The Path of the Complete Argument

This map is not an alternative table of contents, nor a page-by-page summary of the chapters. It reveals the intellectual movement that binds the book's parts together:

An agricultural problem worth understanding ↓ A system that requires explanation ↓ A crop and land that test the value of measurement in the field ↓ Harvest and post-harvest that preserve the output or lose it ↓ An animal, water, and an integrated farm that regulate the exchange of benefit and risk ↓ A value that must be calculated and distributed ↓ A social and regional reality, and a labour market, that change the meaning of the solution and the distribution of its benefits ↓ Responsibility and governance that appear when error and failure occur ↓ An applied case examined within the limits of its evidence ↓ Knowledge that turns into a decision, a roadmap, and a research question Whoever reads the parts in order will see the argument built layer upon layer:

technology does not precede the problem;
performance does not precede evidence;
automation does not precede authority;
purchase does not precede economics;
scaling does not precede verification.

Whoever enters the book at a particular part will find its question declared at the outset, and can then return to this map to see where that question sits within the complete argument. At its core, this book asks more than how far artificial intelligence can be brought into agriculture. It poses the harder and more useful question:

When does agricultural artificial intelligence become a better decision, for whom, under what evidence, and at what cost and responsibility?

Controlled References

The references below follow the source-authority register used in the build. A source number does not mean that every part of the document was examined to the same depth; the description specifies the source type and the limits of its use.

Every reference in this list is a published external document that the reader can access through the digital object identifier (DOI) or the stable link recorded in its entry. This is the condition for inclusion here, not a general description. The number of these references is two hundred and eighty.

What does not appear in this list: implementation evidence specific to the sponsoring platform. In earlier drafts, the first nine items of evidence carried SRC identifiers and were counted among the references, a methodological confusion corrected in this edition: an implementation trace that the reader cannot access is not a reference, however faithful its description may be. That evidence was moved to a separate register in Appendix N, and five further internal traces with identifiers IMP10–IMP14 were later added to it without counting any of them among the references.

Note on numbering: identifiers are fixed and are not recycled. The range SRC016–SRC024 carried the implementation evidence moved to Appendix N, and the range SRC025–SRC029 is reserved for sources nominated during the build that did not pass the verification gate. SRC235 remained unassigned to a verified source when the Chapter Twenty-Four additions were numbered, while SRC251 was withheld from the list because the document bearing that identifier in the Chapter Twenty-Nine draft is the same as SRC033 and is not a new reference. These positions are deliberately left empty so that an old or draft citation cannot shift to a different source. The absence of a number here is not an editorial oversight, but the deliberate trace of a documented transfer, exclusion, or prevention of duplication.

\[SRC001\] Nautiyal, M., et al. (2025). “Revolutionizing agriculture: A comprehensive review on artificial intelligence applications in enhancing properties of agricultural produce.” Food Chemistry: X. https://doi.org/10.1016/j.fochx.2025.102748 — Peer-reviewed review mapping the application field.

\[SRC002\] Wu, K., et al. (2025). “A Comprehensive Review of AI Methods in Agri-Food Engineering: Applications, Challenges, and Future Directions.” Electronics, 14(20), 3994\. https://doi.org/10.3390/electronics14203994 — Peer-reviewed review of methods, applications, and challenges.

\[SRC003\] Özoğul, G. (2025). “Applications of artificial intelligence technologies in agriculture: advantages, challenges, risks, prospects, and recommendations.” Cogent Food & Agriculture. https://doi.org/10.1080/23311932.2025.2568199 — Review of benefits, risks, and prospects.

\[SRC004\] Mohammed, S. P., et al. (2025). “A systematic literature review on artificial intelligence in transforming precision agriculture for sustainable farming: Current status and future directions.” Plant Science Today. https://doi.org/10.14719/pst.6175 — External systematic review; that status does not transfer to this book.

\[SRC005\] Wei, M. C. F., Longchamps, L., Colaço, A. F., & Molin, J. P. (2026). “Integrating stability zones and machine learning for enhanced crop management.” Precision Agriculture. https://doi.org/10.1007/s11119-026-10318-9 — Study identity and scope verified; do not import unexamined findings.

\[SRC006\] Mohan, R. N. V. J., Rayanoothala, P. S., & Sree, R. P. (2025). “Next-gen agriculture: integrating AI and XAI for precision crop yield predictions.” Frontiers in Plant Science. https://doi.org/10.3389/fpls.2024.1451607 — Abstract verified; it permits reporting an R² of up to 0.92 only within the study setting.

\[SRC007\] Askale, G. T., Yibel, A. B., Taye, B. M., & Wubneh, G. D. (2025). “Mobile based deep CNN model for maize leaf disease detection and classification.” Plant Methods. https://doi.org/10.1186/s13007-025-01386-5 — Identity and task verified; do not use the 95% figure until the results table has been checked.

\[SRC008\] Zhang, Y., et al. (2026). “CropGPT: A large multimodal model for precise and explainable diagnosis of crop pests and diseases.” Journal of Industrial Information Integration. https://doi.org/10.1016/j.jii.2026.101107 — Identity and scope verified; do not use the inherited accuracy, BLEU, or class-count figures.

\[SRC009\] dos Santos Ferreira, A., Freitas, D. M., da Silva, G. G., Pistori, H., & Folhes, M. T. (2017). “Weed detection in soybean crops using ConvNets.” Computers and Electronics in Agriculture, 143, 314–324. https://doi.org/10.1016/j.compag.2017.10.027 — Image classification; not Q-learning or robotic removal.

\[SRC010\] Öz, M., & Üstüner, E. (2026). “Artificial Intelligence for Blue Transformation: A Review of Predictive Modeling and Decision Support Systems in Sustainable Aquaculture.” Sustainable Development. https://doi.org/10.1002/sd.71290 — Abstract verified; it does not support a 30% cost-reduction figure.

\[SRC011\] Bai, S., Liu, Z., Wu, W., Xu, N., & Jiang, M. (2025). “GenAI \+ blockchain to coordinate agricultural supply chains to improve quality trust: an agent-based simulation study.” Frontiers in Sustainable Food Systems. https://doi.org/10.3389/fsufs.2025.1591350 — Agent-based simulation, not live deployment.

\[SRC012\] Khanna, A., et al. (2025). “Generative AI and Blockchain-Integrated Multi-Agent Framework for Resilient and Sustainable Fruit Cold-Chain Logistics.” Foods, 14(17), 3004\. https://doi.org/10.3390/foods14173004 — The abstract mentions a 30% reduction in transit time in experiments; do not generalise it.

\[SRC013\] Aman Mohammadi, M., Tajdar-oranj, B., & Esmaeili, E. (2026). “Spectral-AI technologies for food safety: Advances, challenges, and future directions.” Food Control. https://doi.org/10.1016/j.foodcont.2026.112144 — Scope review; it does not support the inherited 99.85% or 96% figures.

\[SRC014\] Food and Agriculture Organization of the United Nations. (2009). How to Feed the World in 2050. https://www.fao.org/fileadmin/templates/wsfs/docs/expert\_paper/How\_to\_Feed\_the\_World\_in\_2050.pdf — A historical foresight document that estimated world population at 9.1 billion people and the need to increase global food production by about 70% by 2050, based on data and assumptions available in 2009; it does not represent a current estimate or a figure valid for use outside its temporal and methodological context.

\[SRC015\] Panagi, P., et al. (2025). “Poultry Farm Intelligence: An Integrated Multi-Sensor AI Platform for Enhanced Welfare and Productivity.” arXiv:2510.15757. https://arxiv.org/abs/2510.15757 — Preprint/emerging evidence.

\[SRC030\] Campoverde-Molina, M., & Luján-Mora, S. (2025). “Cybersecurity in smart agriculture: A systematic literature review.” Computers & Security, 150, 104284\. https://doi.org/10.1016/j.cose.2024.104284 — Reference for agricultural cybersecurity; do not carry over unexamined incident counts from it.

\[SRC031\] Food and Agriculture Organization of the United Nations. (2022). The State of Food and Agriculture 2022: Leveraging automation in agriculture for transforming agrifood systems. FAO. https://doi.org/10.4060/cb9479en — Official report on broad, responsible automation.

\[SRC032\] Jouanjean, M.-A., Casalini, F., Wiseman, L., & Gray, E. (2020). Issues around data governance in the digital transformation of agriculture: The farmers’ perspective. OECD Food, Agriculture and Fisheries Papers, No. 146\. https://doi.org/10.1787/53ecf2ab-en — Data rights, access, transfer, and trust.

\[SRC033\] Tabassi, E. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). NIST. https://doi.org/10.6028/NIST.AI.100-1 — General framework requiring agricultural adaptation.

\[SRC034\] Organisation for Economic Co-operation and Development. (2025). OECD Supply Chain Resilience Review: Navigating risks. OECD Publishing. https://doi.org/10.1787/94e3a8ea-en — Qualitative resilience framework; no imported performance metric.

\[SRC035\] Shamshiri, R. R., et al. (2024). “Digitalization of agriculture for sustainable crop production: a use-case review.” Frontiers in Environmental Science, 12, 1375193\. https://doi.org/10.3389/fenvs.2024.1375193 — Review of the sensing, data, decision, and operations chain.

\[SRC036\] Sanyaolu, M., & Sadowski, A. (2024). “The Role of Precision Agriculture Technologies in Enhancing Sustainable Agriculture.” Sustainability, 16(15), 6668\. https://doi.org/10.3390/su16156668 — Contextual economic and sustainability analysis; do not import unexamined percentages.

\[SRC037\] Lowder, S. K., Sánchez, M. V., & Bertini, R. (2021). “Which farms feed the world and has farmland become more concentrated?” World Development, 142, 105455\. study identifier; open-access copy; FAO presentation of the findings — A peer-reviewed study of farm structure, land concentration, and the contribution of farm-size groups to the value of food production, based on censuses and data from different years; it is not a global census for 2026 and does not make “small farm” and “family farm” synonymous.

\[SRC038\] Van de Zande, G. D., Grant, F., Sheline, C., Amrose, S., Costello, J., Ghodgaonkar, A., & Winter V, A. G. (2024). “Design and Evaluation of a Precision Irrigation Tool’s Human–Machine Interaction to Bring Water- and Energy-Efficient Irrigation to Resource-Constrained Farmers.” Sustainability, 16(19), 8402\. https://doi.org/10.3390/su16198402 — AS-MO design and user evaluation in Kenya, Jordan, and Morocco; no general effect on water or yield.

\[SRC039\] Digital Green. (No date). Farmer.Chat: Your Free Farming Assistant. https://www.farmer.chat/ — Official product page for text, voice, and image query functions and location use; a changing source, accessed 18 September 2026; the page does not establish diagnostic accuracy or agricultural impact.

\[SRC040\] Digital Green. (No date). FarmerChat: Farming Answers in Seconds. https://www.digitalgreen.org/farmerchat — Official programme page for service scope, research materials, and open data; accessed 18 September 2026; the reach and cost figures on it are developer claims unless supported by an independent study.

\[SRC041\] Digital Green. (2026, 9 July). “FarmerChat V2 Is Here, and Farmers Are Asking More Than Ever.” https://www.digitalgreen.org/insights/v2-launch-performance — Product analysis issued by the developer on service spread, activation, retention, and question volume; it permits the use of usage indicators with their definitions and dates, but does not convert them into measures of agricultural accuracy or causal impact.

\[SRC042\] 60 Decibels. (2025, 11 November; updated 24 February 2026). “Smallholder Farmers Are Following AI Recommendations.” https://60decibels.com/insights/farmers-ai-recommendations/ — An evaluation conducted by an organisation outside the product team in partnership with Digital Green and covering 450 farmers in Kenya; it supports findings on recall, use, trust, and self-reported action, but does not establish recommendation correctness, causal effects on yield and income, or generalisability beyond the sample.

\[SRC043\] Digital Green. (2026). FarmerChat Agricultural Q\&A Dataset (Large) \[dataset\]. Hugging Face. https://huggingface.co/datasets/DigiGreen/farmerchat-queries-large — Data card for a dataset containing 1,446,866 question–answer pairs from August 2024 to June 2026 under a CC BY 4.0 licence; the answers are machine-generated, and the selection criterion concerns users who prefer English, so the dataset must not be treated as a truth standard or as evidence of equivalence across languages.

\[SRC044\] Digital Green. (2026). FarmerChat Prompts \[source code\]. GitHub. https://github.com/digitalgreenorg/farmerchat-prompts — Public repository under an MIT licence for managing multi-vendor agricultural templates and evaluation templates for privacy, compliance, conflict, and relevance; it establishes the existence of this open component but does not disclose the full production code or prove that all its functions operate in the live service.

\[SRC045\] Chandrashekar, M. S., Singh, V., & Pedapudi, L. (2026). “Benchmarking Automatic Speech Recognition for Indian Languages in Agricultural Contexts.” arXiv, 2602.03868v2. https://doi.org/10.48550/arXiv.2602.03868 — A preliminary draft comparing speech recognition across 10,934 agricultural recordings in Hindi, Telugu, and Odia; it supports variation in error rates between languages in the study setting, and does not establish end-to-end voice performance for Farmer.Chat, nor does it represent a peer-reviewed paper.

\[SRC046\] Digital Green Foundation. (2025, December). Privacy Policy. https://www.digitalgreen.org/privacy — An official policy covering the scope of collection of images, recordings, location data, chatbot interactions, data sharing, retention, and rights; it describes the stated policy and does not establish operational compliance in every country or case.

\[SRC047\] International Food Policy Research Institute. (undated). Generative AI for Agriculture (GAIA) — Phase I & II. https://www.ifpri.org/project/generative-ai-for-agriculture-gaia/ — An official project page from the lead organisation; it establishes the programme’s identity, direction, and partnerships, and should not be treated as an independent evaluation of a final tool’s performance. Accessed 18 September 2026\.

\[SRC048\] CABI. (undated). Generative AI for Agriculture Advisory. https://www.cabi.org/projects/generative-ai-for-agriculture-advisory/ — An official project page that identifies the duration, partners, countries, roles, Phase I results, and the Phase II plan; the reported willingness-to-use percentage is a project-reported acceptance result, not a measure of accuracy or field impact. Accessed 18 September 2026\.

\[SRC049\] King, E. (2025, 6 January). “Piloting a CABI Crop Health Advisory Chatbot.” PlantwisePlus Blog. https://blog.plantwise.org/2025/01/06/piloting-a-cabi-crop-health-advisory-chatbot/ — An institutional description of workshops and prototype tests in Kenya and India, the crops, participants, and the method of collecting feedback; it is not a peer-reviewed study nor an independent measure of agricultural impact.

\[SRC050\] Msengezi, C. (2025, 14 May). “Lessons in AI Governance from the GAIA Project.” CABI Blog. https://blog.cabi.org/2025/05/14/lessons-in-ai-governance-from-the-gaia-project/ — An institutional source on licensing, attribution, decision-making, and the cost of content preparation in GAIA; it supports the governance description but does not establish its completeness or the system’s effectiveness.

\[SRC051\] CABI. (2026). Model Content License Agreement for AI — Special Terms and General Terms, version 1\. Gates Open Research. https://doi.org/10.21955/gatesopenres.1117269.1 — An open licensing instrument for the fair and sustainable use of content in AI development; this version is not peer-reviewed and does not constitute binding law in every jurisdiction.

\[SRC052\] CABI. (2025). Risk and Ethics Assessment and the GAIA Ethics Toolkit. CGIAR Research Repository. https://hdl.handle.net/10568/181439 — An open report classified as grey literature and internally reviewed; it frames risks relating to data quality, representation, power, transparency, privacy, and traditional knowledge, and does not constitute external verification that GAIA tools actually addressed these risks.

\[SRC053\] Cropin Technology Solutions. (undated). Cropin: SaaS-based AgTech and Agricultural Intelligence Platform. https://www.cropin.com/ — The vendor’s official page for the Cropin platform and its management, analytics, and crop-condition and yield-estimation functions; it establishes only the stated functions and does not establish local accuracy or economic impact. Accessed 25 September 2026\.

\[SRC054\] EOS Data Analytics. (undated). EOSDA Crop Monitoring: Crop Monitoring Software for Remote Farm Analytics. https://eos.com/products/crop-monitoring/ — The official product page for satellite monitoring, vegetation health, and field risks; it does not establish diagnosis of the cause of change or the accuracy of yield forecasting. Accessed 25 September 2026\.

\[SRC055\] Aerobotics. (undated). Leading Yield Forecast Technology for Fruit Companies. https://aerobotics.com/ — The vendor’s official page for orchard functions including yield forecasting and the measurement of some fruit traits; it does not represent independent verification of accuracy or coverage across all crops and varieties. Accessed 25 September 2026\.

\[SRC056\] Outfield. (undated). Orchard Management Platform. https://outfield.xyz/ — The vendor’s official page for orchard monitoring, flowering mapping, crop-load tracking, and yield forecasting; it does not establish the validity of the forecast for every orchard or training system. Accessed 25 September 2026\.

\[SRC057\] Source.ag. (undated). AI Solutions for Greenhouse Growers. https://www.source.ag/ — The vendor’s official page for AI solutions in protected agriculture, including harvest forecasting and operational planning; it does not establish weekly forecast accuracy or the definition of marketable yield in every application. Accessed 25 September 2026\.

\[SRC058\] IUNU. (undated). LUNA: Greenhouse Crop Monitoring and Forecasting; How It Works. official website; equipment and function overview — The vendor’s two pages for a crop-imaging, monitoring, and analysis platform for greenhouses; they describe an advertised commercial function and do not establish independent performance or suitability for every crop and location. Reverified 19 October 2026\.

\[SRC059\] Allen, R. G., Pereira, L. S., Raes, D., & Smith, M. (1998). Crop Evapotranspiration: Guidelines for Computing Crop Water Requirements. FAO Irrigation and Drainage Paper 56\. Food and Agriculture Organization of the United Nations. https://www.fao.org/4/x0490e/x0490e00.htm — An official reference for calculating reference evapotranspiration, crop evapotranspiration, and crop coefficients; it is not an AI system nor a direct instruction to operate irrigation.

\[SRC060\] Brouwer, C., Prins, K., & Heibloem, M. (1989). Irrigation Water Management: Irrigation Scheduling. Training Manual No. 4\. Food and Agriculture Organization of the United Nations. https://www.fao.org/4/t7202e/t7202e00.htm — An official training manual on the principles of irrigation scheduling and of relating water requirements to crop, soil, and climate; it requires local adaptation and does not represent a trained model.

\[SRC061\] Brouwer, C., Prins, K., Kay, M., & Heibloem, M. (undated). Irrigation Water Management: Irrigation Methods. Training Manual No. 5, provisional edition. Food and Agriculture Organization of the United Nations. https://www.fao.org/4/s8684e/s8684e00.htm — An official guide to surface, sprinkler, and drip irrigation methods, wetting patterns, and operation; it does not remove the need to assess system design and local conditions. Accessed 25 September 2026\.

\[SRC062\] Ayers, R. S., & Westcot, D. W. (1985; reprinted 1989 and 1994). Water Quality for Agriculture. FAO Irrigation and Drainage Paper 29 Rev. 1\. Food and Agriculture Organization of the United Nations. https://www.fao.org/4/t0234e/t0234e00.htm — An official reference for assessing irrigation-water salinity, infiltration problems, toxicity, and operational risks in the context of soil, crop, and management; it does not turn a single indicator into a final local determination.

\[SRC063\] Food and Agriculture Organization of the United Nations. (undated). AquaCrop. https://www.fao.org/aquacrop/en/ — An official page for a crop response-to-water and water-productivity simulation model; the model is an agricultural representation of processes and is not automatically an AI system or an irrigation controller. Accessed 25 September 2026\.

\[SRC064\] CropX Technologies. (undated). CropX Agronomic Farm Management System; Crop Nutrition Monitoring. https://cropx.com/; https://cropx.com/cropx-system/nutrition-monitoring/ — The vendor’s two official pages for farm-management and irrigation functions, the integration of soil and weather data, and the monitoring of crop nutrition, salinity, nitrogen leaching events, and application records; they establish only the stated functions and do not establish local accuracy, a savings rate, or an effect on yield. Accessed 25 September 2026 for the general page and 26 September 2026 for the nutrition page\.

\[SRC065\] Phytech. (undated). Connecting Growers to Their Plants. https://www.phytech.com/ — The vendor’s official page for plant, irrigation, pressure, and flow monitoring, and recommendation and execution functions; it does not represent an independent evaluation of the safety of recommendations or their suitability for every crop and location. Accessed 25 September 2026\.

\[SRC066\] Arable Labs. (undated). Arable Connect: Know the Worth of Water. https://www.arable.com/ — The vendor’s official page for field monitoring, rainfall indicators, crop water use, and irrigation decision support; it does not establish that every function relies on AI or that the platform controls irrigation. Accessed 25 September 2026\.

\[SRC067\] Lumo. (undated). Precision Irrigation Platform. https://lumo.ag/ — The vendor’s official page for irrigation scheduling and valve, flow, and pressure monitoring; it establishes the stated functions and does not make automation into AI or establish the stated savings and return figures. Accessed 25 September 2026\.

\[SRC068\] Food and Agriculture Organization of the United Nations. (undated). WaPOR Data: Remote Sensing for Water Productivity. https://www.fao.org/in-action/remote-sensing-for-water-productivity/wapor-data/en/ — An official page for a remote-sensing data service on land and water productivity; it does not represent a direct measurement of what reached the roots of a specific plant, nor a valve-control system. Accessed 25 September 2026\.

\[SRC069\] OpenET. (undated). Open-Source Transparent Water Management Data. https://etdata.org/ — The official service page for satellite-derived evapotranspiration data and tools; it does not replace soil and irrigation-system measurements, and the validity of its use is constrained by place and scale. Accessed 25 September 2026\.

\[SRC070\] Navarro-Hellín, H., Martínez-del-Rincon, J., Domingo-Miguel, R., Soto-Valles, F., & Torres-Sánchez, R. (2016). “A decision support system for managing irrigation in agriculture.” Computers and Electronics in Agriculture, 124, 121–131. https://doi.org/10.1016/j.compag.2016.04.003 — A study of a decision-support system for irrigation management that integrates data and forecasting; it establishes what was tested in its setting, not the existence of a general product covering all crops, soils, and constraints.

\[SRC071\] Goap, A., Sharma, D., Shukla, A. K., & Rama Krishna, C. (2018). “An IoT based smart irrigation management system using Machine learning and open source technologies.” Computers and Electronics in Agriculture, 155, 41–49. https://doi.org/10.1016/j.compag.2018.09.040 — A research prototype integrating field measurements, weather forecasting, and machine learning; it does not establish transferability to all crops and production systems.

\[SRC072\] Huber, L., & Gillespie, T. J. (1992). “Modeling Leaf Wetness in Relation to Plant Disease Epidemiology.” Annual Review of Phytopathology, 30(1), 553–577. https://doi.org/10.1146/annurev.phyto.30.1.553 — A specialist review of the relation between leaf wetness and plant-disease epidemics; it supports risk framing and does not establish the presence of a pathogen in a specific field.

\[SRC073\] Gleason, M. L., Duttweiler, K. B., Batzer, J. C., Taylor, S. E., Sentelhas, P. C., Monteiro, J. E. B. A., & Gillespie, T. J. (2008). “Obtaining weather data for input to crop disease-warning systems: leaf wetness duration as a case study.” Scientia Agricola, 65(spe), 76–87. https://doi.org/10.1590/S0103-90162008000700013 — A methodological study of weather-data sources and leaf wetness duration in crop disease-warning systems; it does not turn a climatic indicator into a diagnosis.

\[SRC074\] Rowlandson, T., Gleason, M., Sentelhas, P., Gillespie, T., Thomas, C., & Hornbuckle, B. (2015). “Reconsidering Leaf Wetness Duration Determination for Plant Disease Management.” Plant Disease, 99(3), 310–319. https://doi.org/10.1094/PDIS-05-14-0529-FE — A review of the measurement and estimation of leaf wetness duration in plant disease management; it does not represent an integrated irrigation-and-disease platform.

\[SRC075\] Mashonjowa, E., Ronsse, F., Mubvuma, M., Milford, J. R., & Pieters, J. G. (2013). “Estimation of leaf wetness duration for greenhouse roses using a dynamic greenhouse climate model in Zimbabwe.” Computers and Electronics in Agriculture, 95, 70–81. https://doi.org/10.1016/j.compag.2013.04.007 — A study estimating leaf wetness duration for rose leaves inside a greenhouse under specific conditions; it does not generalise to every crop or facility and does not establish that changing irrigation alone prevents disease.

\[SRC076\] Food and Agriculture Organization of the United Nations. (2006). Plant Nutrition for Food Security: A Guide for Integrated Nutrient Management. Fertilizer and Plant Nutrition Bulletin 16\. https://www.fao.org/4/a0443e/a0443e00.htm — An institutional technical guide to the principles of plant nutrition, soil fertility, nutrient sources, integrated management, and its environmental and economic aspects; it does not provide a local recipe or establish the performance of a digital tool or AI model.

\[SRC077\] Food and Agriculture Organization of the United Nations. (2013). Good Agricultural Practices for Greenhouse Vegetable Crops: Principles for Mediterranean Climate Areas. https://www.fao.org/4/i3284e/i3284e.pdf — An institutional technical guide to protected agriculture explaining how root-zone conditions, water, nutrients, and management differ in greenhouse crops within its scope; it does not justify transferring the settings of a specific facility to every greenhouse or tunnel, nor does it establish the effectiveness of a commercial platform or autonomous control.

\[SRC078\] United States Department of Agriculture, Natural Resources Conservation Service. (undated). Nutrient Management (Code 590) Conservation Practice Standard. https://www.nrcs.usda.gov/resources/guides-and-instructions/nutrient-management-ac-590-conservation-practice-standard — A government practice standard supporting planning, documentation, and the reduction of nutrient risks under specified standards and constraints; it does not turn US regulatory limits into a global legal rule or a dose for an unspecified crop. Accessed 26 September 2026\.

\[SRC079\] The Fertilizer Institute. (undated). 4R Nutrient Stewardship. https://www.tfi.org/insights/nutrient-stewardship/ — Documentation issued by a fertiliser industry association to define the framework of the right source, rate, time, and place and link it to economic, environmental, and social dimensions; it is an industry source, not an independent evaluation of the effect of a specific practice or evidence of its validity for every country and crop. Accessed 26 September 2026\.

\[SRC080\] Pampolino, M. F., Witt, C., Pasuquin, J. M., Johnston, A., & Fisher, M. J. (2012). “Development approach and evaluation of the Nutrient Expert software for nutrient management in cereal crops.” Computers and Electronics in Agriculture, 88, 103–110. https://doi.org/10.1016/j.compag.2012.07.007 — A peer-reviewed study describing the development of Nutrient Expert and its evaluation for cereal crops in the study contexts; it does not establish its current availability, its coverage of all crops and countries, or the validity of transferring its results without local verification.

\[SRC081\] Chlingaryan, A., Sukkarieh, S., & Whelan, B. (2018). “Machine learning approaches for crop yield prediction and nitrogen status estimation in precision agriculture: A review.” Computers and Electronics in Agriculture, 151, 61–69. https://doi.org/10.1016/j.compag.2018.05.012 — A peer-reviewed scientific review of machine-learning applications in crop-yield prediction and nitrogen-status estimation in precision agriculture; it does not establish diagnosis of a deficiency from a single image or transferable accuracy for every crop and sensor.

\[SRC082\] Grzebisz, W. (2021). “Site-Specific Nutrient Management.” Agronomy, 11(4), 752. https://doi.org/10.3390/agronomy11040752 — A peer-reviewed review article on site-specific nutrient management and the linking of spatial variability to management decisions; it does not mean that every map justifies variable-rate application, nor does it establish the result of any particular platform.

\[SRC083\] Tröster, M. F., & Sauer, J. (2021). “IoFarm: A novel decision support system to reduce fertilizer expenditures at the farm level.” Computers and Electronics in Agriculture, 188, 106322. https://doi.org/10.1016/j.compag.2021.106322 — A peer-reviewed study of a decision-support system and a research-optimisation approach that compares fertilisation decisions within an economic framework; it does not establish the existence of a general commercial product or the transferability of the result to all crops, regions, and objectives.

\[SRC084\] Li, D., Yang, S., Du, Z., Xu, X., Zhang, P., Yu, K., Zhang, J., & Shu, M. (2024). “Application of unmanned aerial vehicle optical remote sensing in crop nitrogen diagnosis: A systematic literature review.” Computers and Electronics in Agriculture, 228, 109565. https://doi.org/10.1016/j.compag.2024.109565 — A peer-reviewed systematic review of the use of UAV optical remote sensing in assessing crop nitrogen status; it does not make the image alone proof of nitrogen deficiency, nor does it make a single model valid for every crop, lighting condition, and season.

\[SRC085\] Savvas, D., & Gruda, N. (2018). “Application of soilless culture technologies in the modern greenhouse industry – A review.” European Journal of Horticultural Science, 83(5), 280–293. https://doi.org/10.17660/ejhs.2018/83.5.2 — A peer-reviewed scientific review of soilless culture in the greenhouse industry, supporting the distinction between systems and the importance of the medium, the nutrient solution, and the root environment; it does not provide a universal nutrient-solution recipe, nor does it establish that automatic control is artificial intelligence.

\[SRC086\] Benke, K., & Tomkins, B. (2017). “Future food-production systems: vertical farming and controlled-environment agriculture.” Sustainability: Science, Practice and Policy, 13(1), 13–26. https://doi.org/10.1080/15487733.2017.1394054 — A peer-reviewed scientific article on vertical farming and controlled-environment agriculture, discussing their general potential and limitations; it does not guarantee profit, energy efficiency, or sustainability for every crop and design, nor does it establish the performance of an AI system.

\[SRC087\] Lakhiar, I. A., Gao, J., Syed, T. N., Chandio, F. A., & Buttar, N. A. (2018). “Modern plant cultivation technologies in agriculture under controlled environment: a review on aeroponics.” Journal of Plant Interactions, 13(1), 338–352. https://doi.org/10.1080/17429145.2018.1472308 — A peer-reviewed scientific review defining aeroponics and the characteristics of delivering water and nutrients to suspended roots, along with its operational dependencies; it does not provide general spray or solution settings, nor does it establish guaranteed superiority or the performance of any specific algorithm.

\[SRC088\] Yara International. (undated). Atfarm. https://at.farm/ — The vendor’s official page for commercial software providing field and crop insights linked to nitrogen-use management; it establishes only the advertised function, and does not establish recommendation accuracy, savings, yield, or that every function relies on machine learning. Accessed 26 September 2026\.

\[SRC089\] Climate LLC. (undated). Climate FieldView. https://climate.com/en-us.html — The vendor’s official page for a platform that brings together field data, scouting, prescription building, and yield-data analysis; it does not establish that every fertiliser prescription is AI-generated, nor does it establish its agronomic validity or any independently demonstrated return. Accessed 26 September 2026\.

\[SRC090\] Priva. (undated). Priva Nutri-Line. https://www.priva.com/horticulture/solutions/greenhouse-water-systems/priva-nutri-line — The vendor’s official page for a family of nutrient-dosing and fertigation systems in protected horticulture; it establishes the advertised automation and control functions, but does not make them artificial intelligence, nor does it establish independent agricultural impact or suitability for every crop, water source, and growing medium. Accessed 26 September 2026\.

\[SRC091\] DSSAT Foundation. (undated). Decision Support System for Agrotechnology Transfer (DSSAT). https://dssat.net/ — Official documentation for a framework for modelling and simulating crop, soil, weather, and management interactions within the available models and data; it is not a local fertiliser recipe, nor an automatic controller or necessarily artificial intelligence. Accessed 26 September 2026\.

\[SRC092\] APSIM Initiative. (undated). Agricultural Production Systems sIMulator (APSIM). https://www.apsim.info/ — Official documentation for a framework for modelling and simulating agricultural systems, including plant, animal, soil, climate, and management modules; it does not establish the validity of every module for every site, nor does it make simulation machine learning or an operational instruction. Accessed 26 September 2026\.

\[SRC093\] PEAT GmbH. (undated). Plantix. https://plantix.net/en/ — The vendor’s official page for an image-based inspection and advisory application that advertises photographing crop problems and providing a diagnosis, suggestions, and agricultural knowledge; it does not establish independent performance, nor does it justify diagnosing a nutrient deficiency or prescribing from the image alone. Accessed 26 September 2026\.

\[SRC094\] Anderson, H. C., Rogers, M. A., & Hoover, E. E. (2019). “Low Tunnel Covering and Microclimate, Fruit Yield, and Quality in an Organic Strawberry Production System.” HortTechnology, 29(5), 590–598. https://doi.org/10.21273/HORTTECH04319-19 — A peer-reviewed study on the effect of low-tunnel covering on microclimate, yield, and quality within an organic strawberry system; it does not generalise the result to every crop or cover, nor does it establish the effectiveness of an AI system or a commercial sensor.

\[SRC095\] Lamont, W. J. (2005). “Plastics: Modifying the Microclimate for the Production of Vegetable Crops.” HortTechnology, 15(3), 477–481. https://doi.org/10.21273/HORTTECH.15.3.0477 — A peer-reviewed scientific article on the use of plastics to modify the microclimate in vegetable production; it supports the general basis for the effect of covers and tunnels, and does not provide a climate setting, fertilisation programme, or sensor design valid for every site.

\[SRC096\] Aranet. (undated). Horticulture Solutions; Wireless Monitoring for a Greenhouse. https://aranet.com/en/home/solutions/horticulture; https://aranet.com/downloads/wireless-monitoring-for-a-greenhouse-eng/ — Vendor pages and documents for solutions that collect environmental and agricultural readings wirelessly and display them for monitoring; they do not establish that the solution is designed specifically for low tunnels, that every analysis relies on artificial intelligence, or that connectivity succeeds under every cover. Accessed 26 September 2026\.

\[SRC097\] Pessl Instruments. (undated). METOS; FieldClimate; iSCOUT. https://metos.global/en/; https://metos.global/en/fieldclimate/; https://metos.global/en/iscout/ — The vendor’s official pages for weather, soil, and water stations, loggers, and sensors, and for forecasting services, water management, and risk models, and for the iSCOUT trap with camera, sticky board, image analysis, and display of counted targets and their trend over time. They establish the advertised functions, but do not establish independent accuracy for identifying every species, nor that the trap count equals damage or justifies intervention, nor that a single sensor represents all parts of the field or facility. Accessed 27 September 2026\.

\[SRC098\] Doktar Technologies. (undated). Smart Digital Agriculture Solutions for Farmers; Product Portfolio; Contact. https://www.doktar.com/solutions/farmers; https://www.doktar.com/products; https://www.doktar.com/contact — The vendor’s official pages for a platform with an advertised presence in Turkey that combines satellite imagery, field-sensor data, AI functions, and tools for irrigation planning and flow monitoring; they do not establish a package dedicated to low tunnels, crop visibility through the cover, or independent field accuracy. Accessed 26 September 2026\.

\[SRC099\] Lea-Cox, J. D., Ross, D. S., & Teffeau, K. M. (2004). “Developing Water and Nutrient Management Plans for Container Nursery and Greenhouse Production Systems.” Acta Horticulturae, 633, 373–379. https://doi.org/10.17660/ActaHortic.2004.633.45 — A published scientific paper on planning water and nutrients in container nursery and greenhouse production systems, and on the complexity of multiple species and irrigation and fertilisation strategies; it does not provide a general recipe for ornamental plants, nor does it establish the accuracy of an AI model or its commercial impact.

\[SRC100\] Mrisho, L. M., et al. (2020). “Accuracy of a Smartphone-Based Object Detection Model, PlantVillage Nuru, in Identifying the Foliar Symptoms of the Viral Diseases of Cassava–CMD and CBSD.” Frontiers in Plant Science, 11, 590889. https://doi.org/10.3389/fpls.2020.590889; https://pmc.ncbi.nlm.nih.gov/articles/PMC7775399/ — A peer-reviewed study with a specific field scope in East Africa, covering recognition of cassava mosaic disease symptoms, cassava brown streak disease, and green-mite damage; some authors are affiliated with the system, and the study does not establish global accuracy or the current availability of every function and crop.

\[SRC101\] FarmQA. (undated). FarmQA Scouting. https://farmqa.com/scouting/ — The vendor’s official page for field-scouting templates, images, reports, and visit management for row and specialty crops; it establishes the advertised work-organisation functions, but does not establish that the platform diagnoses the cause, uses artificial intelligence in every function, or achieves independent agricultural impact. Accessed 27 September 2026\.

\[SRC102\] Network for Environment and Weather Applications (NEWA). (undated). NEWA. https://newa.cornell.edu/ — A university and extension service that provides weather data, risk models, and integrated management tools for listed crops, pests, and diseases; an alert issued by a risk model is not a diagnosis, and it does not provide a universal intervention threshold outside the stations, environments, and models it covers. Accessed 27 September 2026\.

\[SRC103\] United States Department of Agriculture, Animal and Plant Health Inspection Service. (undated). Identification Technology Program — IDtools. https://idtools.org/aboutITP.cfm — A government resource providing keys and digital tools to assist in the identification of invasive invertebrates, plant diseases, and weeds, and guides for sorting trap contents; not every tool in it is necessarily AI-based, and digital identification in itself does not confer authority to take quarantine action. Accessed 27 September 2026\.

\[SRC104\] National Plant Diagnostic Network. (undated). National Plant Diagnostic Network. https://www.npdn.org/ — An institutional website for a network of plant-health laboratories, diagnostic and referral resources, regulatory linkages, and a data repository in the United States; it establishes the existence of a network and referral structure, but does not represent a single test or confer validity outside its institutional and geographical scope. Accessed 27 September 2026\.

\[SRC105\] European and Mediterranean Plant Protection Organization. (undated). EPPO Global Database. https://gd.eppo.int/ — An institutional database for plant-health information, taxonomy, distribution, standards, and measures related to quarantine, diagnosis, and the health of planting material; it is not an AI system, nor a substitute for the national authority or local law. Accessed 27 September 2026\.

\[SRC106\] Early Detection & Distribution Mapping System. (undated). EDDMapS. https://www.eddmaps.org/ — A US platform for invasive-species reports, mobile applications, and maps, including referral of reports to a network of reviewers; it establishes the advertised collection and verification function within the network, but does not make a report a global official declaration or establish the absence of the condition from areas with no points. Accessed 27 September 2026\.

\[SRC107\] IPM Decisions. (undated). IPM Decisions. https://ipmdecisions.net/ — A European research and institutional platform providing access to multiple systems for integrated management decision support and weather data, together with information on the suitability and reliability of those systems; their collection on one platform does not mean that they use one method, or that all of them are artificial intelligence or valid for every country and crop. Accessed 27 September 2026\.

\[SRC108\] CABI. (undated). CABI BioProtection Portal. https://bioprotectionportal.com/ — An institutional portal for searching registered biological protection products by country, crop, and pest, with guidance resources; it establishes the search function within registration information, but does not establish the effectiveness of every product on a particular farm, nor does it create an application rate or independent permission for use. Accessed 27 September 2026\.

\[SRC109\] United States Environmental Protection Agency. (undated). Pesticide Product and Label System. https://ordspub.epa.gov/ords/pesticides/f?p=PPLS:1 — An official regulatory service for product labels accepted by the agency under US law, clarifying the role of the label in handling information and safe use; its scope is the United States and it does not transfer registration to another country, and it is a regulatory source for the label rather than an independent evaluation of the effectiveness of every product. Accessed 27 September 2026\.

\[SRC110\] PlantVillage, Pennsylvania State University. (undated). PlantVillage AI. https://plantvillage.psu.edu/ — An institutional page presenting an assistant in pilot testing that combines external models and tools with weather and soil data and evaluates multiple models; it does not establish general accuracy or coverage, does not confer authority to diagnose or treat, and does not make PlantVillage AI and PlantVillage Nuru a single product. Accessed 27 September 2026\.

\[SRC111\] van Delden, S. H., et al. (2021). “Current status and future challenges in implementing and upscaling vertical farming systems.” Nature Food, 2(12), 944–956. https://doi.org/10.1038/s43016-021-00402-w — A peer-reviewed scientific review of the challenges of implementing and scaling up vertical-farming systems and their relationships with crop, technology, and resources; it does not establish guaranteed profitability or energy efficiency for every facility or crop.

\[SRC112\] Beacham, A. M., Vickers, L. H., & Monaghan, J. M. (2019). “Vertical farming: a summary of approaches to growing skywards.” The Journal of Horticultural Science and Biotechnology, 94(3), 277–283. https://doi.org/10.1080/14620316.2019.1574214 — A peer-reviewed scientific review of the diversity of vertical-farming designs and approaches; it does not provide a single result for energy, production, or economics, nor does it make stacked layers evidence of sustainability.

\[SRC113\] Shamshiri, R. R., et al. (2018). “Advances in greenhouse automation and controlled environment agriculture: A transition to plant factories and urban agriculture.” International Journal of Agricultural and Biological Engineering, 11(1), 1–22. https://doi.org/10.25165/j.ijabe.20181101.3210 — A peer-reviewed scientific review of developments in automation and control in greenhouses and controlled-environment agriculture; it does not establish complete operational autonomy or the suitability of every technology for every facility.

\[SRC114\] Gebbers, R., & Adamchuk, V. I. (2010). “Precision Agriculture and Food Security.” Science, 327(5967), 828–831. https://doi.org/10.1126/science.1183899 — A peer-reviewed scientific article framing precision agriculture as the management of variability using information and technology; it does not establish the accuracy or feasibility of any particular commercial product, nor does it make the map alone an executable prescription.

\[SRC115\] Wolfert, S., Ge, L., Verdouw, C., & Bogaardt, M.-J. (2017). “Big Data in Smart Farming – A review.” Agricultural Systems, 153, 69–80. https://doi.org/10.1016/j.agsy.2017.01.023 — A peer-reviewed scientific review of data architectures, actors, and relationships in smart agriculture; it does not establish that more data or greater connectivity automatically improve decision-making, nor does it treat automation and artificial intelligence as synonymous.

\[SRC116\] Blue Radix. (n.d.). Crop Controller / Autonomous Climate & Irrigation Control. https://blue-radix.com/solutions/ — The supplier’s official page for an advertised AI-based product for greenhouse climate and irrigation control as part of an operating strategy; it establishes the description of the commercial function, but not independent figures for yield, resource use, or labour, nor validity for every crop and facility. Content checked on 26 September 2026, and link availability rechecked on 29 September 2026\.

\[SRC117\] Source.ag. (n.d.). Harvest Forecast; Irrigation Control; Source Intelligence. harvest forecasting; irrigation control; platform — The vendor’s pages for advertised functions in greenhouse harvest forecasting, irrigation control, and facility information management; they do not establish commercially presented accuracy or improvement rates as an independent assessment, nor the transferability of performance or safety of integration to every facility. Reverified 19 October 2026\.

\[SRC118\] Priva. (n.d.). Greenhouse Technology and Horticulture Automation. https://www.priva.com/horticulture — The supplier’s official page for advertised solutions including climate computers, sensors, irrigation, water management, and horticultural automation; it does not establish that all functions use machine learning, nor any general independent effect on yield, efficiency, or profit. Content checked on 26 September 2026, and link availability rechecked on 29 September 2026\.

\[SRC119\] John Deere. (n.d.). See & Spray Ultimate. https://www.deere.com/en-us/products-solutions/sprayers-and-applicators/see-spray-ultimate — The supplier’s official page for an advertised use of computer vision and machine learning to target weeds in specific crops and equipment configurations; it does not establish independent savings, accuracy, or impact, nor suitability for use outside the scope of the product, the label, and local law. Content checked on 26 September 2026, and link availability rechecked on 29 September 2026\.

\[SRC120\] DJI Agriculture. (n.d.). Agricultural Drone and Smart Farm Ecosystem. https://ag.dji.com/ — The supplier’s official page for advertised drones and platforms for planning, monitoring, and spraying or spreading; it does not establish that every agricultural decision function is AI-based, nor the safety, legality, or performance of operation in every country and crop. Content checked on 26 September 2026, and link availability rechecked on 29 September 2026\.

\[SRC121\] Slaughter, D. C., Giles, D. K., & Downey, D. (2008). “Autonomous robotic weed control systems: A review.” Computers and Electronics in Agriculture, 61(1), 63–78. https://doi.org/10.1016/j.compag.2007.05.008 — A peer-reviewed scientific review of the architecture of robotic weed-control systems and the challenges of perception, guidance, and field execution; it maps the field, but does not establish the performance of a current product or its suitability for a particular farm.

\[SRC122\] Norsworthy, J. K., Ward, S. M., Shaw, D. R., Llewellyn, R. S., Nichols, R. L., Webster, T. M., Bradley, K. W., Frisvold, G., Powles, S. B., Burgos, N. R., Witt, W. W., & Barrett, M. (2012). “Reducing the Risks of Herbicide Resistance: Best Management Practices and Recommendations.” Weed Science, 60(SP1), 31–62. https://doi.org/10.1614/WS-D-11-00155.1 — A scientific reference on managing the risk of herbicide resistance through diversifying methods, monitoring survivors, and preventing additions to the seed bank; it does not provide a single prescription for every crop or country, nor does it replace the label and local recommendation.

\[SRC123\] Hasan, A. S. M. M., Sohel, F., Diepeveen, D., Laga, H., & Jones, M. G. K. (2021). “A survey of deep learning techniques for weed detection from images.” Computers and Electronics in Agriculture, 184, 106067. https://doi.org/10.1016/j.compag.2021.106067 — A peer-reviewed scientific review of deep-learning methods for weed detection from images and the diversity of their data and tasks; it supports interpretation of the perception layer and the limits of comparison, but does not by itself establish a control decision or robotic field performance.

\[SRC124\] Mohler, C. L., & Johnson, S. E. (Eds.). (2009; reprinted 2020). Crop Rotation on Organic Farms: A Planning Manual. Sustainable Agriculture Research and Education (SARE), NRAES-177. ISBN 978-1-933395-21-0. resource page; full text — An open practical planning guide that brings together the experience of farmers and researchers in maps, management units, records, crop sequences, cover crops, and contingency plans. Its principles are broadly useful, but the crop and pest schedules and return intervals are aimed chiefly at the north-eastern United States and eastern Canada and should not be transferred literally to other regions.

\[SRC125\] Al-Ansari, Majid Mohsen; Al-Younis, Abdul Hamid Ahmad; et al. (1980). «Crop Rotation» in Principles of Field Crops, pp. 181–204. electronic text; bibliographic record of the book — A historical Arabic academic source linking rotation design to soil, climate, irrigation, labour, and marketing, and setting out how to calculate rotation length and the number of its sections. Its experimental figures, climatic limits, and classificatory terminology should be treated as historical, requiring verification from the primary source and local updating before advisory use.

\[SRC126\] Ma, B.-L. (Ed.). (2016). Crop Rotations: Farming Practices, Monitoring and Environmental Benefits. Nova Science Publishers, 213 pp. ISBN 978-1-63484-496-3. publisher page; open record — A multi-author synthesis covering agriculture, soil, the environment, and economics, including nitrogen fixation, soil health, disease and insects, diversity, and the carbon footprint. The publisher’s description and the record establish identity and scope, but not a uniform effect size for every system.

\[SRC127\] Su, Q., Wu, Y., Dong, Y., Ding, Z., Zhang, W., Ye, T., Lu, X., & Tang, Z. (2026). “Crop Rotation for Sustainable Agriculture: Mechanisms, Technologies, and Regional Recommendations.” Applied Sciences, 16(13), 6511. https://doi.org/10.3390/app16136511 — A peer-reviewed scientific review linking plant–soil–microbe interactions to the role of remote sensing, variable-rate fertilisation, and regional design. It supports the framework of mechanisms and regional adaptation, but does not turn a technical result in a specific system into a universal prescription.

\[SRC128\] World Organisation for Animal Health (WOAH). (the electronic edition in force at the time of consultation). Terrestrial Animal Health Code, Chapter 7.1: Introduction to the Recommendations for Animal Welfare. official text — An official reference for the general principles of animal welfare, the Five Freedoms, and the use of animal- and outcome-based measures; it does not turn a single indicator into a complete assessment, nor does it replace national requirements or veterinary assessment. Content and availability checked on 5 October 2026.

\[SRC129\] Halachmi, I., Guarino, M., Bewley, J., & Pastell, M. (2019). “Smart Animal Agriculture: Application of Real-Time Sensors to Improve Animal Well-Being and Production.” Annual Review of Animal Biosciences, 7, 403–425. https://doi.org/10.1146/annurev-animal-020518-114851 — A peer-reviewed review of the scope of real-time sensing in animal systems, the distinction between individual and group measurement, and the need to turn data into action; it does not establish the validity of a particular device or a general effect on welfare.

\[SRC130\] Rutten, C. J., Velthuis, A. G. J., Steeneveld, W., & Hogeveen, H. (2013). “Invited review: Sensors to support health management on dairy farms.” Journal of Dairy Science, 96(4), 1928–1952. https://doi.org/10.3168/jds.2012-6107 — A peer-reviewed review distinguishing levels of measurement, interpretation, information integration, and decision-making in dairy sensors, and showing the effect of the verification reference, test size, and number of farms; its surveyed results are historical and limited by the search period, and do not describe the current market.

\[SRC131\] Stygar, A. H., Gómez, Y., Berteselli, G. V., Dalla Costa, E., Canali, E., Niemi, J. K., Llonch, P., & Pastell, M. (2021). “A Systematic Review on Commercially Available and Validated Sensor Technologies for Welfare Assessment of Dairy Cattle.” Frontiers in Veterinary Science, 8, 634338. https://doi.org/10.3389/fvets.2021.634338 — An open-access systematic review of the external validation of sensor technologies for dairy-cattle welfare assessment. The numbers of commercial tools discussed in the chapter are limited by the market search conducted between March and May 2020 and by the study criteria, and do not represent a fixed industry rate or approval of every tool.

\[SRC132\] Li, N., Ren, Z., Li, D., & Zeng, L. (2020). “Review: Automated techniques for monitoring the behaviour and welfare of broilers and laying hens: towards the goal of precision livestock farming.” Animal, 14(3), 617–625. https://doi.org/10.1017/S1751731119002155 — A peer-reviewed review of sound, wearables, and image processing in monitoring broilers and laying hens; it establishes the scope of possibilities, the logistical constraints, and the need for further development, but does not establish a generally validated system.

\[SRC133\] Schillings, J., Bennett, R., & Rose, D. C. (2021). “Animal welfare and other ethical implications of Precision Livestock Farming technology.” CABI Agriculture and Bioscience, 2, 17. https://doi.org/10.1186/s43170-021-00037-8 — An open-access scientific commentary exploring ethical risks and opportunities on the basis of an online workshop involving international stakeholders; it is useful for framing questions of welfare, the human relationship, and responsibility, but does not represent a causal experiment or regulatory consensus.

\[SRC134\] World Organisation for Animal Health (WOAH). (the electronic edition in force at the time of consultation). Terrestrial Animal Health Code, Chapter 6.10: Responsible and Prudent Use of Antimicrobial Agents in Veterinary Medicine. official text — An official reference for the responsibilities of responsible and prudent veterinary antimicrobial use, including clinical examination, prescription, dosage regimens, treatment duration, and withdrawal periods within the relevant frameworks; it does not replace national law, current registration, or the veterinarian’s judgement. Content and availability checked on 5 October 2026.

\[SRC135\] Food and Agriculture Organization of the United Nations (FAO). (2019). The State of Food and Agriculture 2019: Moving Forward on Food Loss and Waste Reduction. https://doi.org/10.4060/CA6030EN — An official reference explaining the framework for measuring food loss, chain boundaries, and the need to define the denominator and the point of measurement; it does not establish a general local loss rate, nor does it remove the need for field measurement by product and pathway.

\[SRC136\] Codex Alimentarius Commission. (2004). Code of Practice on Good Animal Feeding (CXC 54-2004). official codes of practice page — An international code on feed safety, good practices, traceability, and their connection to the safety of food of animal origin; it does not replace national law, approval of the feed material, or veterinary and nutritional assessment of the batch and intended use.

\[SRC137\] Food and Agriculture Organization of the United Nations (FAO), International Institute of Rural Reconstruction (IIRR), & WorldFish Center. (2001). Integrated Agriculture-Aquaculture: A Primer. https://www.fao.org/4/y1187e/y1187e00.htm — An official introductory reference to integrated agriculture–aquaculture systems, their contexts, and ways of linking their components; it does not provide a general recipe, nor does it authorise the transfer of water, sediment, waste, or feed materials without local health, environmental, and regulatory verification.

\[SRC138\] Bac, C. W., van Henten, E. J., Hemming, J., & Edan, Y. (2014). “Harvesting Robots for High-value Crops: State-of-the-art Review and Challenges Ahead.” Journal of Field Robotics, 31(6), 888–911. https://doi.org/10.1002/rob.21525 — A scientific review of the state of harvesting robots for high-value crops and the challenges of perception, handling, and operation; it does not establish the maturity of a current product or its commercial performance.

\[SRC139\] Suttie, J. M. (2000). Hay and Straw Conservation: For Small-Scale Farming and Pastoral Conditions. FAO Plant Production and Protection Series No. 29. https://www.fao.org/4/x7660e/x7660e00.htm — An official reference for the sequence of cutting, drying, handling, conservation, and the diversity of forage systems; it does not provide timing or automatic control that is valid for every crop and climate.

\[SRC140\] Food and Agriculture Organization of the United Nations. (2009). International Year of the Potato 2008: New Light on a Hidden Treasure. https://www.fao.org/4/i0500e/i0500e00.htm — A general official reference on potatoes, their production, and their chain; it does not establish a performance figure or machine setting of any particular kind.

\[SRC141\] John Deere. Operations Center. https://www.deere.com/en-us/products-solutions/technology-solutions/precision-ag-technology/operations-center (accessed on 7 October 2026) — The supplier’s official page for collecting field, machine, and operator data and supporting operations management; it does not establish that every function is artificial intelligence, nor the validity of an agricultural decision or an independent effect.

\[SRC142\] John Deere. Combines. https://www.deere.com/en-us/products-solutions/harvesting/combines (accessed on 7 October 2026) — The supplier’s official page for combines and integrated technologies; it does not establish local performance, automatic optimum calibration, or an independent effect.

\[SRC143\] Tevel Aerobotics Technologies. Autonomous Flying Robots for Orchard Harvesting. https://www.tevel-tech.com/ (accessed on 7 October 2026) — The supplier’s official description of flying robots that rely on vision and machine learning to pick orchard fruit and collect data on the fruit; it does not establish general commercial maturity, independent performance, or local service availability.

\[SRC144\] Croptracker. Harvest Crop Yield Records. https://www.croptracker.com/product/farm-management-software/harvest-crop-yield-records.html (accessed on 7 October 2026) — The supplier’s official page for recording harvest, inventory, yield, and bins, and linking them to location, workers, quality, and traceability; it does not establish that every function is artificial intelligence, nor does it warrant a safety or quality decision.

\[SRC145\] Clarifresh. AI-powered Fresh Produce Quality Control. https://clarifresh.com/ (accessed on 7 October 2026) — The supplier’s official page for a platform for fresh-produce quality inspection and AI-supported reporting and analytics; it does not establish food safety or independently verified savings or productivity.

\[SRC146\] AgroFresh. FreshCloud. https://www.agrofresh.com/solutions/freshcloud/ (accessed on 7 October 2026) — The supplier’s official page for a digital platform linking data from the orchard, inspection, storage, and transport; it does not establish that all its functions are artificial intelligence, nor does it confer a safety or quality judgement, nor establish impact independently.

\[SRC147\] World Bank. Employment in agriculture (% of total employment) (modeled ILO estimate), Indicator SL.AGR.EMPL.ZS. Arab World, Egypt, Morocco, and Jordan, 2025; Sub-Saharan Africa, South Asia, and Latin America and the Caribbean, 2025; indicator definition — Official data based on International Labour Organization modelled estimates. The values are shares of total employment in the specified aggregates, not direct counts of farmers or measurements of all household and logistics work associated with agriculture. Reviewed 19 October 2026; latest update shown in the data response: 8 October 2026.

\[SRC148\] International Labour Organization (ILO). (2015). Guidelines for a Just Transition towards Environmentally Sustainable Economies and Societies for All. official page — A guiding framework for social dialogue, social protection, skills development, and labour-market policies during economic transitions; it was used to build the principles of a just transition, but does not provide a prescription specific to harvest automation or an alternative to labour law and national policies.

\[SRC149\] Food and Agriculture Organization of the United Nations (FAO). (2023). The Status of Women in Agrifood Systems. report identifier; official report file — A global report on women’s work, opportunities, and constraints in agrifood systems. The wage-gap analysis used in the development of Chapter Twenty-Four appears on pages 58–59 and in Table 2.4 and concerns a sample of ten countries; it does not establish that automation or the wage gap follows the same pattern in every country, crop, or group.

Additional references for chapter nineteen — revision of 15 October 2026

The additions below comprise seventy-two external link entries used in developing the chapter, with SRC117 reused for irrigation in integrated settings, SRC010 for general review, and SRC137 for agricultural–water integration. The identifier order is a fixed continuation of the earlier register. The revision and access date is 15 October 2026; the publication year of the research or guide is given where verified, whereas “n.d.” refers to a page for which no stable publication year could be established. These additions preserve the historical inventory in Appendix C up to 31 August 2026 as is.

Each entry identifies the source type and the limit of its use. A commercial reference describes a function claimed by the vendor; a technical document explains the conditions of a tool or method; a professional guide sets out inspection and work requirements; and a study is read within the limits of its type, data, and source context. Inclusion of a source does not imply endorsement of a product, availability in Arabic or locally, or the success of a trial in the educational examples. Entries reviewed only through a record or abstract state this explicitly; providing a link for the reader does not mean the full text was successfully retrieved in every verification.

Feeding, water, and visible welfare

\[SRC150\] AKVA group. (n.d.). AKVA observe. Original text or page (accessed 15 October 2026) — Commercial description: The vendor’s official page on video analysis, feeding data and sensor data, monitoring fish behaviour and visible feed pellets, and alerting the operator; it does not establish independent performance, suitability for every species and system, or support in Arabic or local service.

\[SRC151\] UMITRON. (n.d.). Services: CELL and Fish Appetite Index (FAI). Original text or page (accessed 15 October 2026) — Commercial description: The vendor’s presentation of an AI- and Internet of Things-based feeding system, and of an index that analyses group behaviour using machine learning; it does not establish complete intake or satiation, local savings, or transferability of the system across all species.

\[SRC152\] Innovasea. (n.d.). Environmental Monitoring. Original text or page (accessed 15 October 2026) — Commercial description: Official description of measuring oxygen and temperature and displaying data, charts, and configurable alerts through Realfish Pro; measurement, display, and rule-based thresholds are not all AI, and the page does not establish local operational benefit.

\[SRC153\] JALA. (n.d.). JALA App. Original text or page (accessed 15 October 2026) — Commercial description: App page for recording shrimp pond cycles, water, feed, and growth, and for offline entry of field data; it does not make every function an AI prediction, nor does it establish continued alerts or control during service interruption.

\[SRC154\] Innovasea. (7 November 2024). Innovasea Partners with BiOceanOr to Deliver Oxygen Forecasts. Original text or page (accessed 15 October 2026) — Commercial description: Partnership announcement for forecasting oxygen within cages using sensor data, open data, weather and marine models, and AI algorithms; it does not establish activation of the feature in every contract, a two-year forecast horizon and accuracy, or suitability for ponds and recirculating aquaculture systems.

\[SRC155\] Aquabyte. (n.d.). AI and Software. Original text or page (accessed 15 October 2026) — Commercial description: The vendor’s explanation of computer vision and machine learning in analysing images and data, estimating weight, and tracking lice indicators, behaviour, and welfare; it does not establish independent complete counts of the group, clinical diagnosis, or general transferability.

\[SRC156\] OptoScale. (n.d.). Weight. Original text or page (accessed 15 October 2026) — Commercial description: Page for a unit estimating weight, size distribution, and trend from fish images; it does not describe direct weighing of all individuals, nor does it establish live headcount, representativeness of the imaged sample, or performance under local conditions.

\[SRC157\] OptoScale. (n.d.). Welfare. Original text or page (accessed 15 October 2026) — Commercial description: Official description of analysis of visible indicators such as wounds, scale loss, and fin condition, and of guiding follow-up; it does not rule out internal disease, or establish diagnosis, treatment, or performance across every species.

Interpreting evidence, local operation, production, and testing

\[SRC158\] Elicit. (n.d.). Getting Started with Elicit: Which Workflow Should I Use?. Original text or page (accessed 15 October 2026) — Technical document: Official guide to workflows for search, extraction, analysis of papers and reports, systematic review, and the limits of the plan; it establishes the functions described, while the reader must still verify the source and its fit for the decision.

\[SRC159\] Consensus. (n.d.). How Consensus Works. Original text or page (accessed 15 October 2026) — Technical document: Official explanation of literature search and synthesis with citations, the use of full texts where available, and the consensus indicator; the indicator does not provide an independent judgement of study quality or local adoption.

\[SRC160\] Consensus. (n.d.). Responsible AI and Limitations. Original text or page (accessed 15 October 2026) — Technical document: Document on coverage limits and the possibility of misinterpreting the source; it supports returning to the original research and verifying citations, and does not turn automated summarization into a complete systematic review.

\[SRC161\] AKVA group. (n.d.). AKVA Equipment App FAQ: Is It Possible to Use the App in Offline Mode?. Original text or page (accessed 15 October 2026) — Technical document: The vendor’s answer on recording data, making it available offline, and sending it when service returns; it does not establish local model operation, emergency protection, or equipment control independent of the network.

\[SRC162\] AQ1 Systems. (n.d.). Sonic Feeding System. Original text or page (accessed 15 October 2026) — Commercial description: Official description of acoustic detection of shrimp feeding, behavioural algorithms, control and measurement units, and AQ1 Analytics; it does not establish every function under network-independent operation, nor does it classify every rule as machine learning.

\[SRC163\] Edge Impulse. (n.d.). Run Docker: Model Deployment. Original text or page (accessed 15 October 2026) — Technical document: Documentation for running a local model prepared for hardware; the default set-up for retrieving the model requires internet access, and inference can run locally if the necessary components and a suitable file are available. It is a general engineering tool that still requires a model, measurement, and on-farm testing.

\[SRC164\] Food and Agriculture Organization of the United Nations (FAO), Coordinating Working Party on Fishery Statistics (CWP). (n.d.). Handbook: Conversion Factors. Original text or page (accessed 15 October 2026) — Professional reference: Official statistical reference for converting product weight to live-weight equivalent, including the effects of species, size, processing, and conversion conditions; it does not make a single factor valid for every product, nor turn an estimate into a directly measured weight.

\[SRC165\] aquaManager. (n.d.). Platform. Original text or page (accessed 15 October 2026) — Commercial description: The vendor’s presentation of management of production, feeding, inventory, biomass, cost, reporting, data collection, and camera applications for biomass and pellets; it does not make every function AI, nor establish independent accuracy or local availability.

\[SRC166\] Wittaya Aqua. (n.d.). Farm Management Tool. Original text or page (accessed 15 October 2026) — Commercial description: Official description of records for production, feed, samples, comparison, growth, biomass, economic conversion, and bioeconomic modelling; it does not establish that all calculations are machine learning, or that forecasts held true on another farm.

\[SRC167\] Edge Impulse. (n.d.). Model Testing. Original text or page (accessed 15 October 2026) — Technical document: Documentation on test sets, metrics, the confusion matrix, model variables, and thresholds, including conditions for excluding some categories; use of the tool does not guarantee independent testing or the validity of the farm reference used.

\[SRC168\] MLflow. (n.d.). Experiment Tracking. Original text or page (accessed 15 October 2026) — Technical document: Documentation for storing parameters, versions, metrics, files, and links to models and data, and for comparing experiments; an orderly test record does not establish correct inputs, independence, or verified benefit.

\[SRC169\] Evidently. (n.d.). Data Drift. Original text or page (accessed 15 October 2026) — Technical document: Documentation on comparing the distributions of reference and current data; distribution change is an indicator that requires interpretation and measured reference outcomes, and missing data require independent checking. Default settings do not represent farm safety thresholds.

Cages, imaging, emergencies, and cost of use

\[SRC170\] Cardia, F., & Lovatelli, A. (2015). Aquaculture Operations in Floating HDPE Cages: A Field Handbook. FAO Fisheries and Aquaculture Technical Paper No. 593. Rome: FAO. Original text or page (accessed 15 October 2026) — Professional reference: Field handbook on site selection, exposure, currents, nets, feed distribution, moorings, biofouling, and maintenance; used to understand the marine operating context, without transferring design or operating limits to an unexamined site.

\[SRC171\] CageEye. (n.d.). Discovery. Original text or page (accessed 15 October 2026) — Commercial description: The vendor’s description of underwater acoustic monitoring of fish behaviour and distribution, and the use of machine learning; it does not establish complete counting of the group, diagnosis, or verification of feed intake.

\[SRC172\] CageEye. (n.d.). Feeding. Original text or page (accessed 15 October 2026) — Commercial description: The vendor’s presentation of meal planning, configured rules, feedback, meal review, and operating modes; automated mode does not confer local operating authority or remove the conditions for authorizing an action.

\[SRC173\] Georgopoulou, D. G., Vouidaskis, C., & Papandroulakis, N. (2024). “Swimming Behavior as a Potential Metric to Detect Satiation Levels of European Seabass in Marine Cages.” Frontiers in Marine Science, 11, 1350385. DOI: 10.3389/fmars.2024.1350385. Original text or page (accessed 15 October 2026) — Scientific research: Research on European seabass feeding behaviour using computer vision and tracking; its results should be read in light of stocking density, lighting, and the experimental scope, and do not amount to a locally adopted measure of satiation or transferability to every species and system. Publication date: 4 March 2024.

\[SRC174\] AKVA group. (n.d.). Artificial Intelligence Is Shaping the Future of Aquaculture. Original text or page (accessed 15 October 2026) — Commercial description: Corporate explanation of camera use and feeding support in the context of salmon farms; used to describe the development context, without turning installation numbers or future promises into an inventory of current availability or an independent outcome.

\[SRC175\] Beddow, T. A., Ross, L., & Marchant, J. A. (1996). “Predicting Salmon Biomass Remotely Using a Digital Stereo-imaging Technique.” Aquaculture, 146(3–4), 189–203. DOI: 10.1016/S0044-8486(96)01384-1. Original text or page (accessed 15 October 2026) — Scientific research: University of Stirling record and abstract for research on estimating weight and biomass from stereo-image dimensions and regression relationships in Atlantic salmon; only the record and abstract were reviewed, and no claim is made of reading the full text, of current local benefit, or of transferring accuracy rates to another farm.

\[SRC176\] Aquabyte. (n.d.). WEIGHT. Original text or page (accessed 15 October 2026) — Commercial description: Page on weight estimation, growth, size distribution, variation, and harvest reports; the vendor’s description does not establish representation of all sizes, the number of individuals, return on investment, or independent performance.

\[SRC177\] Ace Aquatec. (n.d.). A-BIOMASS. Original text or page (accessed 15 October 2026) — Commercial description: Official description of synchronized stereo cameras and machine learning for body landmarks, dimensions, weight estimation, and tracking; it clarifies the estimation chain and does not make imaging a direct weight measurement or a complete count of the group.

\[SRC178\] Aquabyte. (n.d.). Species: Atlantic Salmon. Original text or page (accessed 15 October 2026) — Commercial description: Species page describing the Atlantic salmon context for which the company presents its models; it specifies Atlantic salmon and does not establish suitability for tilapia, seabass, seabream, or a local Arabic-language service.

\[SRC179\] Masser, M. P., Rakocy, J., & Losordo, T. M. (1999, revised March). Recirculating Aquaculture Tank Production Systems: Management of Recirculating Systems. Southern Regional Aquaculture Center, Publication No. 452. Original text or page (accessed 15 October 2026) — Professional reference: Extension guide on continuity of power, water, and oxygen, the sensitivity of biological filtration, and service interruption; the original copy is hosted by the University of Florida and was used for readiness principles without transferring endurance times, limits, or engineering specifications.

\[SRC180\] OxyGuard. (n.d.). Commander Pacific. Original text or page (accessed 15 October 2026) — Commercial description: Page for a process monitoring and control system; it does not establish that the system is AI, that a particular configuration delivers local protection during interruption, or that it is available with service suited to the site.

\[SRC181\] Innovasea. (n.d.). Realfish Pro: Precision Aquaculture Platform. Original text or page (accessed 15 October 2026) — Commercial description: Description of a platform for measurement, cameras, analysis, alerts, and operating tools; it does not establish continuity of critical functions under local conditions, support in Arabic, or general suitability for ponds and recirculating aquaculture systems.

\[SRC182\] Innovasea. (n.d.). Farm360: Farm Management Software. Original text or page (accessed 15 October 2026) — Commercial description: Page for cloud software covering fish and shrimp farm records, feeding, health, environment, inventory, planning, and traceability; it does not make every function machine learning, nor establish savings, pricing, or local support.

\[SRC183\] Aquabyte. (n.d.). FEEDING. Original text or page (accessed 15 October 2026) — Commercial description: The vendor’s description of pellet tracking, behaviour analysis, appetite estimation, alerts, and equipment integration; it does not guarantee feed savings, forecast horizon, complete intake, or suitability for tilapia ponds.

Simulation, integration, and capture fisheries

\[SRC184\] Bluefarm Environment. (n.d.). DiGiTrout. Original text or page (accessed 15 October 2026) — Commercial description: Presentation of a simulation platform based on scientific models for trout facilities; it does not establish live linkage to sensors, independent performance, or AI in every function.

\[SRC185\] DHI. (n.d.). MIKE ECO Lab. Original text or page (accessed 15 October 2026) — Technical document: Description of software for modelling the aquatic environment, water quality, nutrients, their effects on oxygen, and planning; it requires suitable set-up, limits, and data, and does not amount to ready-made control of a tank or diagnosis of the organism.

\[SRC186\] SINTEF Ocean. (n.d.). FhSim. Original text or page (accessed 15 October 2026) — Technical document: Presentation of a maritime mathematical simulation platform including cages, moorings, and digital twin examples; component availability and licensing vary, and the engineering model does not establish biological suitability for all farm functions.

\[SRC187\] Føre, M., Alver, M. O., Alfredsen, J. A., Rasheed, A., Hukkelås, T., Bjelland, H. V., Su, B., Ohrem, S. J., Kelasidi, E., Norton, T., & Papandroulakis, N. (2024). “Digital Twins in Intensive Aquaculture — Challenges, Opportunities and Future Prospects.” Computers and Electronics in Agriculture, 218, 108676. DOI: 10.1016/j.compag.2024.108676. Original text or page (accessed 15 October 2026) — Scientific research: Scientific review of the definition of the digital twin, its components, use cases, and prospects; the SINTEF record and abstract were reviewed, and the maturity reading applies to the publication date, without turning it into a comprehensive 2026 inventory or local field verification.

\[SRC188\] Katsidoniotaki, E., Su, B., Kelasidi, E., & Sapsis, T. P. (2025). “Multifidelity Digital Twin for Real-time Monitoring of Structural Dynamics in Aquaculture Net Cages.” Scientific Reports, 15, 44281. DOI: 10.1038/s41598-025-27812-1. Original text or page; SINTEF record and reviewed abstract (accessed 15 October 2026) — Scientific research: research combining simulation, machine learning, and field measurements of cage-structure dynamics at SINTEF ACE; its details were completed from the SINTEF record and reviewed abstract after the publisher page could not be opened, without claiming a fresh full-text reading or a ready digital twin for every farm function.

\[SRC189\] Somerville, C., Cohen, M., Pantanella, E., Stankus, A., & Lovatelli, A. (2014). Small-scale Aquaponic Food Production: Integrated Fish and Plant Farming. FAO Fisheries and Aquaculture Technical Paper No. 589. Rome: FAO. Original text or page; Alternative official entry and author details; Document in the FAO repository (accessed 15 October 2026) — Professional reference: an institutional guide to plant and fish production, processing, and differing requirements; its details and official description were checked, and it does not prescribe a specific design, water use, material, or local recipe. The entry link cited in the main text was retained, with an alternative official link because reopening the first was not possible.

\[SRC190\] Bluelab. (n.d.). Edenic. Original text or page (accessed 15 October 2026) — Commercial description: The supplier’s description of monitoring, alerts, data export, and control via compatible devices; it does not turn measurement and rule-based functions into artificial intelligence, nor does it authorize water transfer to another unit.

\[SRC191\] Son, S., & Jeong, Y. (2026, volume year). “Resource-Efficient Nutrient Dosing for Sustainable Aquaponics: Analysis System for Nutrient Requirements in Hydroponics (ASNRH) Using Aquaculture Byproducts and Neural Networks.” Sustainability, 18(1), 247. DOI: 10.3390/su18010247. Original text or page (accessed 15 October 2026) — Scientific research: a framework study for estimating nutrient requirements with neural networks; the electronic publication date was 25 December 2025, and the volume year is 2026. The indexed publisher abstract and citation data were used; the full text could not be retrieved (HTTP 429), without carrying over quantitative findings or claiming independent field validation or approval of material transfer.

\[SRC192\] NOAA Fisheries. (n.d.). Fish Stock Assessment 101 Series, Part 1: Data Required for Assessing U.S. Fish Stocks. Original text or page (accessed 15 October 2026) — Professional reference: an official explanation of catch, abundance, biology, and fishery-independent survey data; it clarifies the requirements for stock assessment and does not determine local stock status or equate the number of observations with total abundance.

\[SRC193\] Hoyle, S. D., Campbell, R. A., Ducharme-Barth, N. D., Gruss, A., Moore, B. R., Thorson, J. T., Tremblay-Boyer, L., Winker, H., Zhou, S., & Maunder, M. N. (2024). “Catch per Unit Effort Modelling for Stock Assessment: A Summary of Good Practices.” Fisheries Research, 269, 106860. DOI: 10.1016/j.fishres.2023.106860. Original text or page (accessed 15 October 2026) — Scientific research: the NOAA page and research abstract on practices that account for catch definition, effort, catchability, environment, location, reporting, and uncertainty; it does not make catch per unit effort a direct measure of stock biomass, and no full-text reading is claimed.

\[SRC194\] NOAA Fisheries. (n.d.). Increasing Efficiency of Video Surveys with Artificial Intelligence. Original text or page (accessed 15 October 2026) — Professional reference: An institutional overview of the use of VIAME for organism detection, motion tracking, video and image annotation, and accuracy review; it does not establish stock abundance, general performance under local conditions, or support in Arabic.

\[SRC195\] Archipelago. (n.d.). Technology: FishVue AI and Interpret. Original text or page (accessed 15 October 2026) — Commercial description: the supplier’s description of detecting clips within video, ranking key moments, and referring them to Interpret for human analyst verification; it does not establish accuracy for every species, direct weight measurement, or general legal acceptance.

\[SRC196\] Global Fishing Watch. (n.d.). Datasets and Code: Apparent Fishing Effort. Original text or page (accessed 15 October 2026) — Technical document: a methodology for inferring apparent fishing activity from vessel automatic identification system (AIS) data using machine learning, with limits related to small-vessel coverage, reception, and use; it is not a measurement of catch or stock abundance.

\[SRC197\] Global Fishing Watch. (n.d.). API Documentation: Data Caveats. Original text or page (accessed 15 October 2026) — Technical document: documentation of the limits of events, inferred classification, context, and uncertainty in AIS data; it supports separating inference from proof of compliance, a legal ruling, or a complete inventory of activity.

\[SRC198\] Food and Agriculture Organization of the United Nations (FAO). (n.d.). Small-scale Fisheries Governance. Original text or page (accessed 15 October 2026) — Professional reference: the voluntary-guidelines page on small-scale fisheries governance, responsible use, co-management, and the inclusion of fishers, women, and marginalised groups; it does not grant a local exploitation right or replace the competent authority.

Biosecurity and technology localization

\[SRC199\] World Organisation for Animal Health (WOAH). (electronic edition at the time of access). Aquatic Animal Health Code: Biosecurity for Aquaculture Establishments. Original text or page (accessed 15 October 2026) — Professional reference: an official reference for risk assessment and establishment-specific administrative and physical measures to prevent the introduction, spread, and release of pathogens; it does not approve a site-specific procedure or replace national requirements and specialists.

\[SRC200\] World Organisation for Animal Health (WOAH). (electronic edition at the time of access). Aquatic Animal Health Code: Disinfection of Aquaculture Establishments and Equipment. Original text or page (accessed 15 October 2026) — Professional reference: an official reference on how disinfection relates to organic-matter removal, surfaces, water, and appropriate procedures; it was used for principles only, without transferring chemical doses, a standardized protocol, or any certificate of success.

\[SRC201\] World Organisation for Animal Health (WOAH). (electronic edition at the time of access). Manual of Diagnostic Tests for Aquatic Animals: General Introduction. Original text or page (accessed 15 October 2026) — Professional reference: an official introduction to the link between sampling and the purpose of diagnosis and testing; it distinguishes targeted detection from population representation and does not turn a limited sample into a judgement on all individuals in the establishment.

\[SRC202\] World Organisation for Animal Health (WOAH). (electronic edition at the time of access). Aquatic Animal Health Code: Aquatic Animal Health Surveillance. Original text or page (accessed 15 October 2026) — Professional reference: an official reference for population definition, surveillance design, test characteristics, and uncertainty; it does not turn a single result into a comprehensive health status or an automatic movement or release decision.

\[SRC203\] aquaManager. (n.d.). aquaManager: Core Platform. Original text or page (accessed 15 October 2026) — Commercial description: a page on organizing production, movement, feeding, and mortality events and linking hatchery and grow-out; data management alone is not an intelligent model, a biosecurity plan, or proof that all pathways are complete.

\[SRC204\] Aquabyte. (n.d.). LICE. Original text or page (accessed 15 October 2026) — Commercial description: the supplier’s description of AI-based sea-lice counting and categorization from images, with images available for review; it does not establish diagnosis of all pathogens, authorization for movement or treatment, or performance for local species.

\[SRC205\] Manolin. (n.d.). Aquaculture Data Intelligence. Original text or page (accessed 15 October 2026) — Commercial description: an official description of linking group history, movement, splitting, and production cycles; this page alone does not establish the presence of a machine-learning model in every function or the completeness of the establishment record.

\[SRC206\] Manolin. (n.d.). Watershed: Aquaculture Fish Health Software. Original text or page (accessed 15 October 2026) — Commercial description: the supplier’s presentation of disease-risk prediction models and site and group indicators; each model’s scope, reference basis, validation, and contractual activation require examination, and it does not authorize automatic diagnosis or treatment.

\[SRC207\] Aquabyte. (n.d.). Species: Trout. Original text or page (accessed 15 October 2026) — Commercial description: a page on models specialized for rainbow trout in estimating weight, growth, welfare, and sea lice; it does not establish performance in every freshwater system, a local service in Arabic, or suitability for another species.

\[SRC208\] Innovasea. (n.d.). BiomassPro: Biomass Estimation. Original text or page (accessed 15 October 2026) — Commercial description: the supplier’s presentation of estimating size and weight with stereo cameras, artificial intelligence, and size distribution, for the species and installation positions it describes; it does not establish complete counting, guaranteed total biomass, or local verification.

\[SRC209\] AQ1 Systems. (n.d.). Our Innovation. Original text or page (accessed 15 October 2026) — Commercial description: the supplier’s explanation of using artificial intelligence in developing behavioural detection and control; it does not establish that every acoustic algorithm, calculation, or rule is machine learning, nor does it generalize the system’s suitability.

\[SRC210\] Alfred Wegener Institute (AWI). (2023). MonitorShrimp: Mit künstlicher Intelligenz zu artgerechter und nachhaltiger Garnelenzucht. Original text or page (accessed 15 October 2026) — Research project: the institute’s report on a prototype tested in a research and development facility in Kiel in 2022, using phone images and a local server to count shrimp, estimate their length, and study visible indicators; it is not evidence of current general commercial availability or local adoption.

\[SRC211\] Zhao, T., Shen, Z., Zou, H., Zhong, P., & Chen, Y. (2022). “Unsupervised Adversarial Domain Adaptation Based on Interpolation Image for Fish Detection in Aquaculture.” Computers and Electronics in Agriculture, 198, 107004. DOI: 10.1016/j.compag.2022.107004. Original text or page; AGRIS record and reviewed abstract (accessed 15 October 2026) — Scientific research: research on adapting detection across different aquaculture scenes; the indexed abstract was checked and the author names and details were completed from AGRIS after the publisher was inaccessible (HTTP 403). No claim is made to have read the full text or to have established transfer of the model to every species and system.

Traceability, operator experience, and learning

\[SRC212\] Global Dialogue on Seafood Traceability (GDST). (n.d.). What Is the Objective of the Standard?. Original text or page (accessed 15 October 2026) — Professional reference: an official explanation of key data elements, critical tracking events, and the format for exchanging them digitally; it does not specify here the latest version number or a compliance date, nor does it establish the safety or correctness of every entry.

\[SRC213\] Trace Register. (n.d.). ORIGIN: Platform & Benefits. Original text or page (accessed 15 October 2026) — Commercial description: a page on linking lots and documents, exchanging data, integrating with CSV, Excel, and API, and supporting traceability and recall; it does not make all functions AI, nor does it establish chain completeness or local compliance.

\[SRC214\] ThisFish. (n.d.). Tally Features. Original text or page (accessed 15 October 2026) — Commercial description: the supplier’s presentation of production flows, inventory, quality, traceability from material to product, splitting and merging lots, and linking the reader, printer, and scale; representation of the local process requires testing.

\[SRC215\] ThisFish. (n.d.). TallyBot. Original text or page (accessed 15 October 2026) — Commercial description: a description of assistance with automation, analysis, prediction, alerting, checking external databases, specification inspection, and visual classification through TallyVision; calculations, rules, and search are distinguished from learned prediction and classification, without guaranteeing identity or product safety.

\[SRC216\] Microsoft. (n.d.). Azure Document Intelligence in Foundry Tools. Original text or page (accessed 15 October 2026) — Technical document: a page on AI-based extraction of text, fields, tables, and structure from documents, and on the service’s inclusion within Azure Content Understanding; it is a component that requires integration and testing on local documents, not a complete traceability platform or a guaranteed reader of identity documents in Arabic.

\[SRC217\] aquaManager. (n.d.). Mobile Applications: aquaMobile. Original text or page (accessed 15 October 2026) — Commercial description: a description of recording feeding, mortality, and environmental data in the field, with online and offline operation and automatic synchronization; it does not establish that recording and synchronization are machine learning, or that a critical alert was received during an outage.

\[SRC218\] Aquabyte. (n.d.). WELFARE. Original text or page (accessed 15 October 2026) — Commercial description: a page on image analysis for visible indicators, trends, and alerts for units needing follow-up; it does not validate a diagnosis, establish a causal relationship, or replace the operator, nor does it establish local verification.

\[SRC219\] Aquabyte. (n.d.). Species. Original text or page (accessed 15 October 2026) — Commercial description: the company page identifying models and species such as Atlantic salmon, coho, and trout; it does not grant general support for tilapia, seabream, or shrimp, or validate transfer of the tool without verification.

\[SRC220\] FAO e-learning Academy. (n.d.). Occupational Safety and Health in Inland Aquaculture and Fisheries. Original text or page (accessed 15 October 2026) — Professional reference: an official digital course for identifying, assessing, and managing occupational hazards; digital learning is not in itself artificial intelligence, and completing the course does not grant field authorization or establish that its content is available in Arabic.

\[SRC221\] FAO e-learning Academy. (n.d.). Pathway to Aquaculture Biosecurity. Original text or page (accessed 15 October 2026) — Professional reference: an official course on stepwise management and risk analysis across the aquaculture value chain; it is a learning resource and does not certify the readiness of a site plan or delegate authority for a health or operational action.

\[SRC222\] Farmforce. (n.d.). Farmforce Origin (IMS). Original text or page (accessed 15 October 2026) — Commercial description: the supplier’s page for recording farms and fields, linking product units to their origin, and tracing aggregation and weights; it does not establish the correctness of entries, the system’s suitability for every crop, or the availability of local service in the language required by the farm, nor do record functions by themselves amount to AI.

\[SRC223\] RELEX Solutions. (n.d.). RELEX for Fresh Food Producers. Original text or page (accessed 15 October 2026) — Commercial description: an official page on machine-learning demand forecasting and production planning in fresh foods; evidence of an enterprise capability being offered, without independent proof of accuracy, return, or readiness for deployment on a small farm or in a specific local market.

\[SRC224\] Google for Developers. (2025, updated 22 January). Vehicle Routing — OR-Tools. Original text or page (accessed 15 October 2026) — Technical document: documentation for optimization and vehicle routing with capacities, time windows, resources, and limits on reaching an optimal solution; the tools require modelling, data, and integration, and do not by themselves amount to machine-learning prediction or approval of a route and loading conditions.

\[SRC225\] Tive. (n.d.). Solutions. Original text or page (accessed 15 October 2026) — Commercial description: a page on devices and a platform for monitoring shipment location, some conditions, alerts, and reports; functions vary by device and service. It does not establish coverage of every location and package or local performance, nor does it turn a sensor reading into a safety judgement or product release.

\[SRC226\] GS1. (2023, March). EPCIS and CBV Implementation Guideline, Release 2.0. Original text or page (accessed 15 October 2026) — Technical document: an official guide for implementing traceability events, unit relationships, and sensor information; specific passages from the guide were used without claiming to have read all its pages. The reference relates to Release 2.0 and does not claim that it is the latest available release; the standard does not establish the truth of an event or the success of integration between two systems without testing.

\[SRC227\] anyLogistix. (n.d.). Supply Chain Software Features. Original text or page (accessed 15 October 2026) — Commercial description: a page of functions for network optimization, dynamic scenario simulation, disruptions, and resources; it does not establish validation of a local agricultural model or the readiness of an on-site alternative, nor does it make simulation artificial intelligence merely because it is presented digitally.

\[SRC228\] National Institute of Standards and Technology (NIST). (2024). The NIST Cybersecurity Framework (CSF) 2.0. NIST CSWP 29. document; official resource centre (reverified 19 October 2026) — Professional reference: an official framework for structuring cybersecurity risk management, review, and responsibilities; it does not attest to the implementation of controls or the readiness of recovery in any particular facility.

Additions for developing Chapters Twenty-Three to Thirty

\[SRC229\] Charlton, D., Hill, A. E., & Taylor, J. E. (2022). Automation and Social Impacts: Winners and Losers. FAO Agricultural Development Economics Working Paper 22-09. Official publication; DOI — Research working paper: analyses variation in the social impacts of automation and in the distribution of its gains and losses; it does not produce a single estimate of labour displacement for every crop and country.

\[SRC230\] Li, R. T., Yuan, F. H., Ali, S., Yin, X., He, Y., & Liu, Y. F. (2026). “Method for the Estimation of the Cutting Points in Tomato Seedling Grafting Based on Improved YOLO11n.” International Journal of Agricultural and Biological Engineering, 19(1), 179–186. Article page; DOI — Peer-reviewed study: establishes the results of two computer-vision tasks in the study setting; it does not establish grafting success, income, safety, or direct transfer to another nursery.

\[SRC231\] University of California, Davis, Horticulture Innovation Lab. (2011). Tomato Grafting Guide. University guide page; Official PDF — University guide: explains cleft grafting and the associated care; cut positions and instructions should not be transferred to other grafting methods without specialist review.

\[SRC232\] Gyalai, I. M., Kedves, O., Kedves, A., Kónya, Z., & Kovács, F. (2026). “Challenges and Solutions in Hand Pollination for Hybrid Pepper (Capsicum annuum L.) Seed Production: A Review.” Planta, 263, 138. Publisher page; Open-access text — Scientific review: discusses the challenges of hand pollination for hybrid pepper seed production; it does not document the operation of a labour-estimation model on the teaching farm.

\[SRC233\] Croptracker. (n.d.). Farm Labor Tracking / Punch Clock; Farm Management Apps. Labour tracking; Apps (both accessed on 19 October 2026) — Commercial description: establishes advertised functions for time and task tracking; it does not establish record accuracy, the fairness of a wage rule, or legal and operational suitability at a particular site.

\[SRC234\] International Labour Organization (ILO). (2025). Revolutionizing Health and Safety: The Role of AI and Digitalization at Work. Official report; Algorithmic management materials — Professional reference: discusses the risks and opportunities of digitalization and algorithmic management at work; it does not set a work pace or a safety procedure for a particular farm.

\[SRC236\] International Federation of Robotics (IFR). (2026). World Robotics 2026: Service Robots — Executive Summary. Official summary — Industry report: presents 2025 sales for the agricultural robots category on printed page 13; the sample does not represent a complete census of robots or of the farms using them.

\[SRC237\] International Federation of Robotics (IFR). (2026). World Robotics 2026: Service Robots — Sources & Methods. Official methodology — Methodology document: sets out the agriculture classification and the limits of the supplier sample, and warns against constructing a time series from editions with differing samples; it does not permit converting annual sales into an adoption rate among farms.

\[SRC238\] GSMA Intelligence; GSMA. (2026). The State of Mobile Internet Connectivity 2026. GSMA Intelligence findings; Report portal; Barriers to adoption and use; Press release — Industry-body report: provides an estimate of the population connected to mobile internet at the end of 2025; its definition and methodology are not the same as internet use in general, and its results do not measure farmers or agricultural land area.

\[SRC239\] Hello Tractor. (n.d.). Growing Together; Book a Tractor; Fleet Management; Our Global Footprint. Official website (accessed on 19 October 2026) — Commercial description: describes booking, demand aggregation, operators, technicians, and geographic footprint; it does not establish tractor autonomy or the service’s net effect on jobs or income.

\[SRC240\] Neo, G. H., & Rao, K. T. (2023, 15 June). How to Make Digital Transformation of Agriculture Work: Lessons from Telangana. World Economic Forum. Article — Project description from a participating party: describes the pilot phase of the Saagu Baagu project and its services; 2023 figures do not represent a current count and do not measure a causal effect on labour.

\[SRC241\] Solinftec. (2023; 2026). Sustainability Report 2023; Solinftec Expands U.S. Footprint 243%, Deploys 100+ Autonomous Robots…. Company report; 2026 announcement — Commercial description: establishes what the company states about operating Solix in Brazil and the United States; it does not establish general adoption, a farm without workers, or an independent effect on labour.

\[SRC242\] Federal Bureau of Investigation, Cyber Division. (2022, 20 April). Ransomware Attacks on Agricultural Cooperatives Potentially Timed to Critical Seasons. Private Industry Notification 20220420-001. Official bulletin — Official alert: documents a threat pattern reported by the agency; it does not measure the probability of compromise of a particular platform or farm and does not replace a security assessment for that platform or farm.

\[SRC243\] International Telecommunication Union (ITU). (2024, 10 September). Global Cybersecurity Index 2024: Countries Strengthening Cybersecurity Efforts, but Increased Action Needed on Evolving Threats. Official statement — Institutional source: presents country-level index results; it does not establish the security of an agricultural service or the effectiveness of a particular technical control.

\[SRC244\] United Nations Conference on Trade and Development (UNCTAD). (2025). Implementing World Summit on the Information Society Outcomes: A Twenty-Year Review. TCS/DTL/INF/2025/2. Official document — UN report: provides a review of policies, connectivity, and digital transformation; it does not measure the readiness of a particular farm or local platform.

\[SRC245\] Organisation for Economic Co-operation and Development (OECD). (2022). Declaration on Government Access to Personal Data Held by Private Sector Entities. OECD/LEGAL/0487. Official text — International legal instrument: sets out stated principles and safeguards among adhering countries; it is not a uniform global law or legal opinion for a particular jurisdiction.

\[SRC246\] European Data Protection Board (EDPB). (2024). Guidelines 01/2022 on Data Subject Rights — Right of Access, Version 2.1. Official guidelines — European regulatory guidance: interprets the right of access within the scope of the European regulation; it does not apply automatically outside that scope and does not determine ownership of non-personal agricultural data.

\[SRC247\] Article 29 Data Protection Working Party; endorsed by the European Data Protection Board. (2017). Guidelines on the Right to Data Portability, WP242 rev.01. Official page — European regulatory guidance: explains the portability of personal data within its legal scope; it does not by itself require portability of all farm data or of inferences derived from them.

\[SRC248\] Information Commissioner’s Office (ICO). (n.d.). How Do We Ensure Individual Rights in Our AI Systems? Official guidance (accessed on 19 October 2026) — Regulatory guidance: sets out considerations for individuals’ rights in AI systems; it does not replace determining the applicable law, the contract, and the rights of the parties in each service.

\[SRC249\] Mohanty, S. P., Hughes, D. P., & Salathé, M. (2016). “Using Deep Learning for Image-Based Plant Disease Detection.” Frontiers in Plant Science, 7, 1419. DOI — Peer-reviewed study: compares model performance on curated PlantVillage images and small external image sets; it illustrates a transfer gap and does not estimate global in-field accuracy.

\[SRC250\] Guo, C., Pleiss, G., Sun, Y., & Weinberger, K. Q. (2017). “On Calibration of Modern Neural Networks.” Proceedings of the 34th International Conference on Machine Learning, PMLR 70, 1321–1330. Published text — Methodological study: explains the calibration of classification probabilities in the study settings; it does not make a score shown in an agricultural product interface a valid probability without matched local testing.

\[SRC252\] Schwartz, R., Vassilev, A., Greene, K. K., Perine, L., Burt, A., & Hall, P. (2022). Towards a Standard for Identifying and Managing Bias in Artificial Intelligence. NIST Special Publication 1270. Document — Professional reference: addresses sources of bias in the sociotechnical system; it does not establish that a particular model or institution is free of bias.

\[SRC253\] Food and Agriculture Organization of the United Nations (FAO). (2018). World Programme for the Census of Agriculture 2020: Volume 2 — Operational Guidelines. FAO Statistical Development Series 16. Official document — Statistical guide: sets out agricultural census concepts and operations; it does not make data from a private platform a representative sample or an official census.

\[SRC254\] Digital Green Foundation. (n.d.). FarmerChat – Agri Assistant. Google Play app listing (accessed on 18 October 2026) — App description from the developer: sets out the advertised functions, countries, and availability; it does not establish crop or dialect coverage or recommendation quality at a particular site.

\[SRC255\] University of California Agriculture and Natural Resources. (n.d.). Figs. Guidance resource — University guidance reference: explains aspects of fig fruiting and management; transferring timings and pruning severity requires the local cultivar, climate, and training system to be taken into account.

\[SRC256\] University of California Statewide Integrated Pest Management Program. (n.d.). Figs: Pruning and Thinning. Guidance resource — University guidance reference: explains principles for pruning figs and removing broken and crossing branches; it does not specify a cutting plan for a tree that has not been examined.

\[SRC257\] University of Georgia Cooperative Extension. (n.d.). Basic Principles of Pruning Woody Plants. Guidance resource — University guidance reference: explains the effects of heading back, removal, and apical dominance; these are general principles and do not replace knowledge of the species, cultivar, tree condition, and production objective.

\[SRC258\] Agriculture and Horticulture Development Board (AHDB). (n.d.). Key Development Phases and Growth Stages in Wheat. Guide — Professional guidance reference: explains wheat development and the factors affecting it; its UK seasonal calendar should not be transferred to another environment without localization.

\[SRC259\] Agriculture and Horticulture Development Board (AHDB). (n.d.). Leaf Emergence and Tillering Growth Stages in Winter Wheat (GS1–GS2). Guide — Professional guidance reference: explains leaf emergence and tillering; it does not turn early tiller number into a promise of ear number or yield.

\[SRC260\] Agriculture and Horticulture Development Board (AHDB). (n.d.). The Growth Stages of Cereals. Guide — Professional guidance reference: defines cereal stages and their indicators, including GS31; naming the stage does not by itself determine the dose or timing of an agricultural input.

\[SRC261\] Food and Agriculture Organization of the United Nations, FAO e-learning Academy. (n.d.). Introduction to the Farmer Field School Approach. Learning resource — Institutional educational reference: explains learning through observation, analysis, and participation; it does not establish the effect of a programme or AI add-on that has not been tested in the community concerned.

\[SRC262\] Lewis, P., et al. (2020). “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks.” Advances in Neural Information Processing Systems 33. Published paper — Technical paper: establishes a method for retrieval-augmented generation in the study settings; it does not establish the correctness of an agricultural recommendation or the suitability of a source for a particular field.

\[SRC263\] Fabregas, R., Kremer, M., & Schilbach, F. (2019). “Realizing the Potential of Digital Development: The Case of Agricultural Advice.” Science, 366(6471), eaay3038. DOI — Scientific review: summarizes evidence on digital advisory services and the conditions for customization, interaction, and financing; it does not prove the effect of a new service or language model in every field.

\[SRC264\] Cole, S. A., & Fernando, A. N. (2020). “‘Mobile’izing Agricultural Advice: Technology Adoption, Diffusion and Sustainability.” The Economic Journal, 131(633), 192–219. DOI — Experimental study: evaluates a voice telephone service in Gujarat within its study setting; its findings do not transfer automatically to another country, channel, or AI service.

\[SRC265\] Waddington, H., Snilstveit, B., Hombrados, J., Vojtkova, M., Phillips, D., Davies, P., & White, H. (2014). “Farmer Field Schools for Improving Farming Practices and Farmer Outcomes: A Systematic Review.” Campbell Systematic Reviews, 10(1). DOI — Systematic review: synthesizes evaluations of farmer field schools and shows the limits of transfer to non-participants; it does not establish the effect of a new programme in a specific location.

\[SRC266\] Agency for Healthcare Research and Quality (AHRQ). (n.d.). Use the Teach-Back Method: Tool #5. Health Literacy Universal Precautions Toolkit. official tool (accessed 19 October 2026) — Health communication tool: explains how to check understanding through teach-back; its principle is borrowed for agricultural communication and is not presented as having tested impact in agricultural extension.

\[SRC267\] International Telecommunication Union (ITU). (2025). Facts and Figures 2025: Internet Use in Urban and Rural Areas. official page; statistics portal — Official statistic: estimates internet use among the population in urban and rural areas; it does not measure farmland area or connectivity for a specific field task.

\[SRC268\] International Telecommunication Union (ITU). (2026). Facts and Figures 2026: Mobile Network Coverage. official results; methodology — Official statistic: coverage is the share of the population within an estimated signal footprint; it does not establish service quality, use, or coverage of a specific agricultural area.

\[SRC269\] World Bank, International Energy Agency, International Renewable Energy Agency, United Nations Statistics Division, & World Health Organization. (2026). Accelerating Universal Energy Access. official statement — Joint institutional source: presents 2024 estimates of access to electricity; basic access does not establish continuity of supply, voltage quality, or adequacy of power for farm equipment.

\[SRC270\] International Telecommunication Union (ITU). (2025). Innovative Solutions for Connectivity in Rural and Remote Areas. report page; last-mile solutions resources — Institutional technical reference: presents connectivity alternatives for rural and remote areas; it does not select a solution for a site before examining terrain, power, spectrum, finance, and maintenance.

\[SRC271\] Food and Agriculture Organization of the United Nations (FAO). (2022). Leveraging Automation and Digitalization for Precision Agriculture: Evidence from the Case Studies. FAO Agricultural Development Economics Technical Study No. 24. official page — Technical study: draws on 22 case studies and sets out infrastructure and enabling conditions; it does not establish a uniform effect for every setting or tool.

\[SRC272\] ABALOBI. (n.d.). Monitor. service provider page (accessed 19 October 2026) — Commercial/institutional description: sets out stated functions for recording catch data, synchronization, and working in multiple languages; it does not establish an independent effect on income or stock status.

\[SRC273\] Gandhi, R., Veeraraghavan, R., Toyama, K., & Ramprasad, V. (2009). “Digital Green: Participatory Video and Mediated Instruction for Agricultural Extension.” Information Technologies & International Development. article page — Research case study: describes the participatory video model and mediation in extension; it does not establish results for a newer service or generalize impact to every community.

\[SRC274\] Baul, T., Karlan, D., Toyama, K., & Vasilaky, K. (2024). “Improving Smallholder Agriculture via Video-Based Group Extension.” Journal of Development Economics, 169, 103267. DOI; open-access full text — Experimental study: tests adding video to face-to-face extension with women farmers in Bihar within its study setting; it does not establish the effect of video or mediation itself in every crop and community.

\[SRC275\] World Bank. (n.d.). Case Study: Huaral Valley Agrarian Information System, Peru; Community Telecommunications Cooperative: The SIA Project in the Chancay–Huaral Valley, Peru. case page; World Bank document — Historical case: describes an information and communications network linked to a water users’ association; it does not establish continued operation, AI use, or a contemporary economic effect.

\[SRC276\] International Telecommunication Union (ITU). (2006, January). Agricultural Information Project for Farmers of the Chancay-Huaral Valley. official case document — Historical case: documents the project components and local connectivity; it does not establish its current status or an independent causal effect.

\[SRC277\] Internet Society. (n.d.). RedINC: A Community Network in Buenos Aires, Cauca, Colombia. case study — Historical institutional case: shows the roles of participation, spectrum, power, terrain, finance, and maintenance; it does not measure a causal agricultural effect or guarantee continuity of service.

\[SRC278\] Food and Agriculture Organization of the United Nations (FAO). (2020). How to Overcome Water Challenges in Agriculture. official page — Institutional source: presents global estimates for rainfed agriculture and irrigated land under water stress in the context of the published report; it does not describe 2026 or a specific field.

\[SRC279\] Food and Agriculture Organization of the United Nations (FAO). (n.d.). WaPOR Data; WaPOR Portal. data documentation; portal (both accessed 19 October 2026) — Institutional data platform: documents the stated layers, spatial resolution, and update frequency; remote-sensing data are not a direct measurement of a particular plant’s root zone or an irrigation prescription.

\[SRC280\] PlantVillage. (n.d.). PlantVillage+. Google Play app listing (accessed 19 October 2026) — Developer app description: sets out the stated functions, updates, and availability, which vary by crop, country, language, device, and connectivity; it does not establish global accuracy or that all functions work offline.

\[SRC281\] Fujiuchi, N., Inaba, K., Oh, S., Okajima, S., Asai, Y., Nishina, H., & Takayama, K. (2024). “Using a Real-Time Photosynthesis and Transpiration Monitoring System to Develop Random Forests Models for Predicting Cherry Tomato Yield in a Commercial Greenhouse.” Environment Control in Biology, 62(2), 29–39. article page; DOI — Peer-reviewed research: the prediction results are specific to the facility, crop, inputs, and windows studied; they do not establish the performance of a commercial product or transferability to other facilities.

\[SRC282\] Organisation for Economic Co-operation and Development (OECD). (n.d.). Transparency and Explainability — OECD AI Principle 1.3. official principle page (accessed 22 October 2026) — Intergovernmental AI principle: supports disclosure appropriate to context, meaningful information about an AI system, and the ability of affected people to challenge an output; it does not verify that any named platform provides those safeguards in operation.

\[SRC283\] Organisation for Economic Co-operation and Development (OECD). (n.d.). Robustness, Security and Safety — OECD AI Principle 1.4. official principle page (accessed 22 October 2026) — Intergovernmental AI principle: supports life-cycle risk management, traceability, and the ability to override, repair, or decommission unsafe systems; it is not an operational safety assessment of Aladdin or any other product.

\[SRC284\] Organisation for Economic Co-operation and Development (OECD). (n.d.). Accountability — OECD AI Principle 1.5. official principle page (accessed 22 October 2026) — Intergovernmental AI principle: establishes that AI actors should be accountable for proper functioning and adherence to the OECD principles; it does not identify the accountable actor or prove compliance in a particular deployment.

\[SRC285\] Food and Agriculture Organization of the United Nations (FAO) & World Health Organization (WHO). (2014). International Code of Conduct on Pesticide Management. official FAO page — International pesticide-management code: provides a life-cycle framework for pesticide governance and risk reduction; it does not replace current national law, product authorisation, the approved label or directions for use, or case-specific agronomic judgement.

\[SRC286\] Food and Agriculture Organization of the United Nations (FAO). (n.d.). Pesticide Registration — Registration Toolkit and Guidelines. official FAO page (accessed 22 October 2026) — Institutional registration guidance: explains the role and practice of pesticide registration and points to assessment resources; it does not establish that a product is currently authorised for a particular crop, pest, use, or jurisdiction.

\[SRC287\] Food and Agriculture Organization of the United Nations (FAO) & World Health Organization (WHO). (2022). International Code of Conduct on Pesticide Management: Guidance on Good Labelling Practice for Pesticides (second revision). official WHO publication page — International technical guidance: supports clear, accurate, and context-appropriate pesticide labels as a core risk-management instrument; it does not substitute for the current label approved by the competent authority in the country of use.

\[SRC288\] World Wide Web Consortium (W3C). (2013). PROV-Overview: An Overview of the PROV Family of Documents. W3C Working Group Note. official specification — Technical provenance standard overview: supplies a common model for describing entities, activities, agents, and their provenance; recording provenance under this model does not by itself establish that the source data are accurate or fit for an agricultural decision.

\[SRC289\] World Wide Web Consortium (W3C). (2016). Data on the Web Best Practices: Data Quality Vocabulary. W3C Note. official specification — Technical data-quality vocabulary: supports explicit quality metadata, measurements, annotations, and policies; it provides a way to describe and exchange assessments, not proof that a dataset has passed a particular quality threshold.

\[SRC290\] World Wide Web Consortium (W3C), Internationalization Working Group. (2026). Internationalization Best Practices for Spec Developers. W3C Group Note, 7 August 2026. official note — Internationalisation guidance: supports designing specifications for language, script, direction, locale, and culturally variable data from the outset; it does not demonstrate the completeness or usability of a product’s localisation.

\[SRC291\] World Wide Web Consortium (W3C), Internationalization Working Group. (2026). Strings on the Web: Language and Direction Metadata. First Public Working Draft, 16 July 2026. official working draft — Draft technical specification: explains why natural-language strings need explicit language and direction metadata and proposes implementation patterns; as a working draft, it may change and cannot be treated as a final standard or as evidence that a platform applied it correctly.

\[SRC292\] World Wide Web Consortium (W3C), Internationalization Working Group. (2014). Authoring HTML: Handling Right-to-Left Scripts. W3C Working Group Note. official note — Web-authoring guidance: supports correct handling of bidirectional text and right-to-left interfaces; it does not demonstrate that a product handles them correctly without testing the actual interface, content, device, and user journey.

\[SRC293\] European Commission, Directorate-General for Translation. (n.d.). Translation Quality. official page (accessed 22 October 2026) — Institutional quality framework: supports meaning-and-effect equivalence, risk-based quality assurance, and revision or review appropriate to the text’s consequences; it does not certify any Aladdin translation or establish user comprehension in an agricultural setting.

\[SRC294\] Autio, C., Schwartz, R., Dunietz, J., Jain, S., Stanley, M., Tabassi, E., Hall, P., & Roberts, K. (2024). Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile. NIST AI 600-1. official NIST publication page; DOI — Government technical profile: identifies generative-AI risks and actions for governing, mapping, measuring, and managing them across the life cycle; it is voluntary guidance, not a certification or an evaluation of a particular agricultural system.

\[SRC295\] Unicode Consortium. (2026). Unicode Locale Data Markup Language (LDML), Part 3: Numbers. Unicode Common Locale Data Repository (CLDR), Version 49, 21 October 2026. official specification — Locale-data technical specification: supports locale-sensitive number symbols, patterns, formatting, and related conventions; use of CLDR data does not establish that units, magnitudes, or agronomic meanings survived a translation or interface transformation.

\[SRC296\] Unicode Consortium, Common Locale Data Repository (CLDR). (n.d.). Plural Rules. official specification page (accessed 22 October 2026) — Locale-data specification: defines grammatical plural categories and the numeric examples used to determine them across languages; it does not supply translated agricultural terminology or prove that a message is understandable or safe.

Relationship of the Reference Additions to the Development of Chapters Thirty-One to Thirty-Three

References SRC282–SRC296 were registered in the controlled list on 22 October 2026 to document the accessible external sources used for the general principles governing the revised Part Nine. They strengthen the public basis for transparency, safety, accountability, pesticide governance, provenance, data quality, internationalisation, translation quality, locale handling, and generative-AI risk management. They do not convert the private IMP artefacts into public references, independently verify an Aladdin capability or deployment state, or establish localisation success, affordability, agricultural benefit, or field impact.

Chapter	New references	Supporting function and evidentiary limit
31	SRC282–SRC284, SRC294	Transparency, robustness, accountability, and risk-based evaluation; no independent proof of platform performance or user benefit
32	SRC282, SRC285–SRC293, SRC295, SRC296	User protection, pesticide governance, provenance, data quality, and technical and semantic localisation; no proof that a local user completed the intended task safely
33	SRC282–SRC296, as matched to individual claims	Provenance, quarantine, abstention, pesticide safeguards, localisation controls, and risk ordering; no proof that Aladdin implemented a proposed policy or produced an agricultural outcome

Relationship of the Reference Additions to the Development of Chapters Twenty-Three to Thirty

The following additions were logged during development of the working filters for Chapters 23–30, after verification of the source identity, link, and evidentiary boundary. Recording them centrally preserves the provenance of that development and does not mean that every paragraph in those filters has in fact been integrated into the body of the adopted Arabic manuscript; therefore, the “Evidence density by chapter” table below remains an inventory of the adopted text alone.

Chapter	New references logged	Existing references reused or expanded
23	SRC254–SRC266	SRC031, SRC037, SRC039
24	SRC229–SRC234, SRC236–SRC241	SRC031, SRC037, SRC058, SRC147–SRC149
25	SRC267–SRC277	SRC100, SRC110, SRC238
26	SRC278–SRC281	SRC038, SRC100, SRC117
27	No new additions	SRC030, SRC033
28	SRC242–SRC248	SRC032, SRC228
29	SRC249, SRC250, SRC252, SRC253	SRC006, SRC033
30	No new additions	SRC030; IMP cross-references remain separate implementation evidence and are not included in the reference count

No SRC251 entry was created because its material was a verbatim duplicate, in bibliographic identity and digital identifier, of the NIST AI RMF reference already registered under SRC033, and the citation in the Chapter Twenty-Nine candidate was corrected to the original identifier. SRC235 also remained blank because it was not assigned to a verified source in the numbering sequence of the Chapter Twenty-Four candidate. Neither number is included in the total count.

How the references relate to the sections of chapter nineteen

This index links the descriptive references in the sections to their identifiers in the list, without changing the text of those sections. The general review SRC010, integration SRC137, and irrigation SRC117 appear under their listed identifiers; interpretive topics that do not carry an independent link still require professional claim-level matching, and this index does not close the local verification gap.

Item	Usage focus	Relevant references
1	Why does fish condition differ from one pond to another?	SRC010
2	Feeding management: How does feed turn into growth?	SRC150, SRC151
3	Water quality: How do we detect a problem and act at the right time?	SRC152, SRC153, SRC154
4	Aquatic animal health and growth: Monitoring condition and estimating live biomass	SRC155, SRC156, SRC157
5	From research to farm: Evaluating the evidence before adopting the technology	SRC010, SRC158, SRC159, SRC160
6	When connectivity is lost: How does the farm continue monitoring and response?	SRC161, SRC162, SRC163
7	Harvest, growth, and survival: How do we evaluate farm production?	SRC164, SRC165, SRC166
8	Before adoption: How do we test the system under farm conditions?	SRC167, SRC168, SRC169
9	Marine cages: How do we care for fish as the current and sea state change?	SRC150, SRC170, SRC171, SRC172, SRC173, SRC174
10	The camera and the sample: How do we estimate fish weight and size distribution?	SRC155, SRC175, SRC176, SRC177, SRC178
11	Emergency preparedness: Protecting aquatic animals during power outages and equipment failure	SRC154, SRC179, SRC180, SRC181
12	Technology cost and value: What the farm pays, and what it gains from using it	SRC155, SRC182, SRC183
13	Simulating aquaculture systems: How do we study alternatives before applying them on the farm?	SRC184, SRC185, SRC186, SRC187, SRC188
14	From the fish tank to the crop: How do we use water and nutrients safely?	SRC137, SRC189, SRC190, SRC117, SRC191
15	Capture fisheries: How do we assess fish stocks and manage a shared resource?	SRC192, SRC193, SRC194, SRC195, SRC196, SRC197, SRC198
16	Data validity: Keeping underwater measurements and images reliable	SRC010
17	Biosecurity on the farm: How do we protect organisms and limit the spread of infection?	SRC199, SRC200, SRC201, SRC202, SRC203, SRC204, SRC205, SRC206
18	Localizing technology: How do we verify its suitability for the species and production system?	SRC162, SRC207, SRC208, SRC209, SRC210, SRC211
19	Tracing seedstock and feed through to delivery: A verifiable chain of origin	SRC203, SRC212, SRC213, SRC214, SRC215, SRC216
20	Environmental impact: What reaches the water and land outside the farm	SRC010
21	The operator on the smart farm: Inspection expertise, authority to act, and continuous learning	SRC181, SRC217, SRC218, SRC219, SRC220, SRC221

SRC010's connection to items 16 and 20 provides scoping background for the general review, not stand-alone primary evidence for each monitoring method or environmental impact. The professional references on water, cages, biosecurity, integration, and fisheries help readers understand the limits of the work; their number does not turn them into validated limits, prescriptions, or unexamined local outcomes.

Relationship of references to the items in Chapter twenty-two

This index maps functions and approaches to their sources; a commercial reference points to the vendor description, a technical reference to a function or standard, and a research paper to its study set-up. SRC216 has been reused for the documents without repeating its entry, and the chapter's four earlier sources have been retained. Comparisons and worked numerical examples remain educational, and the effectiveness of tools, alternatives, and local performance requires suitable independent verification.

Item	Usage focus	Relevant references
1	Dependency network and shared resources	SRC034
2	Case continuity, synchronization, and duplicate prevention	SRC226, SRC228
3	Demand forecasting and fresh-food planning	SRC223
4	Transport optimization and shipment-condition monitoring	SRC224, SRC225
5	Agents, blockchain, simulation, and the limits of experimentation	SRC012, SRC011, SRC227
6	Traceability, identity, aggregation, and event exchange	SRC222, SRC226
7	Risk scenarios and alternatives	SRC034, SRC227
8	Cybersecurity and risk management governance	SRC030, SRC228
9	Scenario and resilience metrics	SRC034, SRC227
10	Document processing and its connection to evidence	SRC216
11	Cross-sector dependency and shock transmission	SRC034
12	Exchange, data rights, and service review	SRC226, SRC228
13	Learning from scenarios and repairing functions	SRC034, SRC227

The references linked to items 2 and 12, and to the learning metrics, provide technical and professional background; they do not establish that any specific vendor implements all synchronization and exit controls, or that a general framework settles the contract. Investigation, recall, safety, acceptance, and recovery steps, however, are procedures that require approval and verification by specialists and duly authorized parties on site. This index is not an independent literature review of every sentence.

Internal sources for editorial governance

Corpus inventory: 00\_Audit/00\_Corpus\_Inventory.csv.
Claim verification register: 00\_Audit/06\_Claim\_Verification\_Register.csv.
Statistics update register: 00\_Audit/07\_Statistics\_Update\_Register.csv.
Visual asset register: 00\_Audit/09\_Visual\_Asset\_Register.csv.
Consolidation ledger: 00\_Audit/13\_Master\_Consolidation\_Ledger.csv.
Legacy edition useful additions register: Master\_Build/01\_Evidence/Legacy\_Edition\_Useful\_Additions\_Register.csv.
For Review content register: Master\_Build/01\_Evidence/For\_Review\_Content\_Register.csv.

These registers are editorial evidence of decisions and process trails; they are not general scientific references.

Separate register: implementation evidence

Evidence of implementation on the sponsoring entity’s platform does not appear in the reference list above. Its place is Appendix N, under identifiers IMP01–IMP14. It is separated because it is not accessible to the reader and does not carry a permanent public identifier such as a DOI, although its traces are internally tagged with dates and fingerprints declared in Appendix N. Reviewing Appendix N is a prerequisite for understanding the limits of the Aladdin case study in Part Nine and the AgroGenie evidence layers presented in Chapters Thirty-Two and Thirty-Three.

Appendices, indexes, and registers

These pages do not re-explain the chapters, nor do they turn their examples into ready-made recipes. Their function is to give the reader a path for review and action: where to find the case or term, what rank its evidence carries, and what is required before moving it to another farm, institution, or region. Accordingly, each register is read with its cut-off date and source limits, and each work template is read only after the crop, animal, or agricultural material; the place and season; the user, language, and unit; the competent authority; and the locally available service and alternative have all been specified.

Localisation here does not mean replacing foreign words with Arabic ones. A phrase may be linguistically correct yet impossible to implement because the unit is different, the local term refers to another variety, the named body has no authority in that country, or spare parts, service, and connectivity are unavailable. For that reason, the appearance of a program, study, or scenario in the appendices does not grant permission to transfer, purchase, or operate; rather, it identifies the starting point and the questions that must be resolved locally.

Appendix A: Register of case studies and applications

Identifier	Case/product	Domain	Current rank	Where cited	Use decision
CS001	Mohan et al. Yield/XAI	Yield prediction	Early study; verified abstract	9, 29, 33	Permits R² up to 0.92 within the study set-up
CS002	Stability Zones \+ ML	Spatial management	Early study; descriptive data	8, 9, 33	Concept scope only
CS003	Mobile CNN for Maize	Maize disease	Early study; full text not retrieved	12, 33	Task and method; not 95%
CS004	CropGPT	Multimodal diagnosis	Early study; full text not retrieved	12, 33	Existence and scope; not 93.1%/BLEU
CS005	Soybean ConvNet	Weeds	Early study	13, 33	Image classification; not DQL/removal
CS006	PoultryFI	Poultry	Preliminary draft	18, 33	Platform described; not a production standard
CS007	Aquaculture review	Aquaculture	Peer-reviewed review	19, 33	Domain map; not 30% cost
CS008	Bai agent simulation	Supply chain	Simulation	22, 33	Co-ordination analysis; not live deployment
CS009	Khanna cold-chain	Logistics	Computational experiment	17, 22, 33	30% travel time in experiments only
CS010	Spectral-AI review	Food safety	Peer-reviewed review	17, 33	Scope; not 99.85%/96%
CS011	AS-MO irrigation	Irrigation/design	Design and user study	10, 23, 26, 33	Differences in automation preferences; no general savings
CS012	Farmer.Chat	Multilingual, multimodal advisory	Institutionally operational service \+ geographically limited user evaluation	7, 26	The sources permit description of the functions, the declared rollout, and the results from a sample of 450 farmers in Kenya; they do not establish general agricultural accuracy or a causal effect on yield or income
CS013	GAIA	Generative advisory and knowledge governance	Research and development programme \+ prototypes \+ governance documents	7	The sources permit description of the partnership, preliminary testing, licensing, and governance; they do not establish a finished product or broad field impact
CS014	AgroGenie	Governed integration/search	Three time layers: August IMP \[IMP06\], development package \[IMP13\], and official distribution package \[IMP14\]	7, 32	Each layer establishes only what was examined within it: the first preserves the historical judgement, the second establishes isolated local tests of the development package, and the third establishes the presence of specific components, contracts, and checks within the official package; none of them establishes live WordPress operation, activation in every installation, local suitability, or field impact, and no runtime result is inherited between packages without retesting
CS015	PlantVillage Nuru	Limited connectivity/disease	Lead source	25, 26	Version and performance must be documented
CS016	Carbon Robotics	Weed removal	Market description/lead source	13	Technology map; no impact figures
CS017	Tomato harvesting	Robotics	Case family not closed	16	Maturity/productivity question; no commercial claim
CS018	Strawberry harvesting	Robotics	Case family not closed	16	Fragile handling; primary source required
CS019	Cucumber harvesting	Robotics	Case family not closed	16	Occlusion and access; primary source required
CS020	Apple harvesting	Robotics/quality	Case family not closed	16	Harvest and quality; verification required
CS021	Aladdin AiBridge / Chat V2	AI governance and retrieval	August IMP with limited local live verification, and official distribution package \[IMP14\]	31–33	The August IMP establishes a governed architecture in the source and limited local live verification of chat and retrieval paths, with release decision `INCOMPLETE / NO-GO`. \[IMP14\] establishes the presence of specific files, contracts, and checks in the official package and reveals constraints in the AiBridge test set; it does not carry over the earlier runtime result without retesting, and it does not establish provider connectivity, answer quality, local suitability, or agricultural impact
CS022	Aladdin Mobile Gateway	Mobile — server gateway	August-dated IMP evidence	25, 32	The August IMP establishes an implemented server gateway with contract, routes, signature, and sessions; the client is not included, distribution is not verified, and bidirectional synchronisation is not supported by the contract declaration. These judgements are not carried over to the September package as a runtime result unless they are retested on it
CS023	Aladdin Desktop	Desktop	August-dated IMP evidence \+ local build	32	Test build, packaging, and signing succeeded locally on 21 August 2026; the governance changes of 22 August did not close all runtime gates thereafter. There is no production signing and no store distribution, and a catalogue of 22 items is not 22 workspaces, nor does the existence of the September package establish that these results were repeated on it

How to read the register

The presence of a case in the register preserves its usefulness and prevents its loss, but it does not grant it the rank of ‘field-documented’ or make it suitable for the reader’s context. Before using it, the reviewer determines whether it is a study, review, simulation, implemented program, or internal IMP, then establishes the version, source, context, comparator, metric, outcome, and limits.

When transferability is examined, fields that may not be omitted are added: the agricultural material, its stage, and its destination; the site, season, and production system; the user, role, language, units, and device; the body that authorises the decision; the service, maintenance, and local alternative; and the outcome to be measured. If one of these elements changes in a way that affects the decision, the case remains useful knowledge but reverts to a hypothesis that requires local verification.

Appendix B: Directory of institutions and data — for discovery, not endorsement

The list below is a discovery directory, not a ready-made source list for every country. It separates the name of the institution or provider from the specific dataset, and separates downloadability from the suitability of the data for agricultural decision-making. At the time of use, the product, version, access date, licence, unit, resolution, and review method must be documented, then compared with the authoritative national or local source where one exists.

Institution/provider	Category	Potential use	Verification required before use
FAO / FAOSTAT	Organisation/statistics	Production, food, inputs, and indicators	Series definition, year, and revision
AGRIS (FAO)	Knowledge index	Discovery of agricultural literature	Record quality and access to the text
World Bank Data	Development data	Infrastructure, economy, and indicators	Definition, original source, and currency
OECD Agriculture	Policy and data	Governance, markets, and policies	Membership, scope, and definition
CGIAR	Research network	Crops, climate, data, and cases	Programme, version, and licence
IFPRI	Food policy research	Markets, policies, and extension	Method and geography
NASA Earthdata	Earth observation	Imagery, climate, and derived products	Product, resolution, and algorithm
USGS Landsat	Satellite imagery	Long-term change	Cloud, correction, and resolution
Copernicus/Sentinel	Imagery and services	Crops, moisture, and cover	Product, sensor, and licence
ECMWF/Copernicus Climate	Weather/climate	Reanalysis and forecasting	Spatial and temporal resolution, and uncertainty
WMO	Standards and weather	Institutional reference	National source and version
GBIF	Biodiversity	Species records	Sampling bias and resolution
SoilGrids/ISRIC	Soil maps	Initial spatial estimation	Resolution, uncertainty, and local verification
OpenStreetMap	Spatial data	Roads, facilities, and supporting boundaries	Completeness, licence, and currency
Zenodo	Research repository	Data, code, and releases	DOI, licence, and documentation
Kaggle	Dataset hosting	Training and experimentation	Origin, licence, and leakage
GitHub	Code/data	Discovery and implementation	Owner, version, licence, and maintenance
Crossref	Metadata	DOI, title, and author	Does not replace the full text

Rules for using the data directory

Do not cite the home page in place of a specific dataset or report.
Keep a copy, download date, and hash where permitted.
Read the documentation for variables, units, and missing values.
The licence, restrictions, and privacy are reviewed.
Representation and spatial and temporal bias are tested.
A competition dataset is not used as evidence of field success.
Every result is linked to the raw record and the transformation.
A global product does not replace the national record or local inspection merely because its coverage is broader.
Administrative boundaries, local names, the calendar, and units are set explicitly and not standardised silently.
If the data are old, low-resolution, or do not represent the production system, the limitation appears with the output, and the model does not fill it in by guessing.

Appendix C: Guide to the tables in the main text

This is a navigational guide to the main analytical tables, not a count of every small table or operational record. The short title describes the table’s function, while the definition of fields and boundaries remains in the chapter itself. Before final publication, this list is matched automatically and manually against the manuscript headings so that no table retitled in revision remains listed under an old name.

Map of pressures and their transmission pathways — Chapter 1\.
Dimensions of the transformation test — Chapter 2\.
The most suitable source for each question — Chapter 3\.
Case study sheet — Chapter 3\.

4-A. Regional transferability limits (Regional Transferability Limits) — Chapter 3\.

AI method selection matrix — Chapter 4\.
Sample observation record — Chapter 5\.
Spatial monitoring platforms — Chapter 6\.
Advisory assessment layers — Chapter 7\.
Forecasting and interpretation case — Chapter 9\.
Nutrient data and their limits — Chapter 11\.
Comparison of maize disease and CropGPT — Chapter 12\.
Herbicide modes of action — Chapter 13\.
Map of the seven terms and their five axes — Chapter 14\.
AI functions, their limits, and human authority — Chapter 14\.
Harmonising the terms cycle, sequence, and cover — Chapter 15\.
Cropping-cycle data contract — Chapter 15\.
Ledger of the previous crop’s causal effect — Chapter 15\.
The eight decision gates of the cropping cycle — Chapter 15\.
AI functions and the limits of their authority in the cropping cycle — Chapter 15\.
Calculating the environmental impact of the cropping cycle — Chapter 15\.
Expanded teaching-case alternatives — Chapter 15\.
Cropping-system scorecard — Chapter 15\.
Animal assessment and welfare — Chapter 18\.
Selected harvest cases — Chapter 16\.
Food quality and safety authority — Chapter 17\.
Investment economics model — Chapter 20\.
Market-valuation transparency — Chapter 21\.
Supply-chain risk map — Chapter 22\.
Service models for smallholders — Chapter 23\.
The difference between task, function, and livelihood — Chapter 24\.
Regional pathways matrix — Chapter 26\.
Agricultural authority matrix — Chapter 27\.
Data rights agreement — Chapter 28\.
Model card — Chapter 29\.
Maturity ladder — Chapter 30\.
Types of evidence in the Aladdin case study — Chapter 31\.
Architectural barriers — Chapter 32\.
Aladdin capability register — Chapter 32\.
Controlled evidence landscape — Chapter 34\.
Product/case evaluation register — Chapter 35\.
Matrix of localised research questions, their interventions, outcomes, and transfer limits — Chapter 36\.
Matrix of research-question data, verification design, fairness, safety, and economics — Chapter 36\.

Additional tables in the appendices: evidence density by chapter and source-base gateway (Appendix Z) · implementation-evidence register and its promotion conditions (Appendix N) · the seven inclusion conditions, the map of application families, the four evidence dimensions, the master register of implemented programmes, and the gap matrix (Appendix S).

Appendix D: Guide to the proposed figures

These figures are editorial specifications for figures not yet produced, not independent evidence. For each figure, when implemented, there is an editable source version, a source and licence record, alt text, and a review of terminology, colours, and directionality. If a figure shows a map, unit, authority, or service pathway, its local version is produced only after the boundaries, names, units, and entities have been verified; a generic map or stereotyped icons are not carried over as a representation of a specific region.

Agriculture as a system of interconnected pressures.
The decision chain from measurement to follow-up.
The multi-authority evidence pyramid.
The data life cycle and transformation provenance.
Spatial monitoring–ground verification–action.
The governed agricultural RAG chain.
Layers of plant-disease decision-making.
The perception–action chain for the weeding robot.
Water, energy, and yield balance.
Transfer of tasks and skills.
Cascading-failure network in supply.
Export, transition, and exit rights.
Maturity ladder from idea to impact.
Public AiBridge chain, safe for publication.
Protocol for source, methodological, and temporal triangulation.

Do not include a generated image, a marketing diagram, or an undocumented map until it has been linked to the asset register and to its source, permission, and alt text. Nor may a figure introduce a number, a causal relationship, or an operating condition that the main text does not establish.

Appendix E: Glossary of Terms

This glossary defines how each term is used within the book; it does not claim to impose a single term on all countries or disciplines. The local name is retained as used by the people who use it, and linked to the standard concept when needed rather than erased. When a difference in wording changes the meaning of the variety, stage, tool, or authority, documenting that difference becomes part of the data and of the localisation decision, not a matter of style.

Remote sensing: collecting information about a surface or crop without direct contact, via satellite, aircraft, or a nearby sensor.

Inference: running a model on an input to produce a prediction, classification, or text.

Retrieval-augmented generation (RAG): retrieving selected evidence and then using a generative model to formulate an answer from it.

Aquaculture: organised production of aquatic organisms; not equivalent to capture fisheries.

Abstention: the system’s decision not to issue a judgement when confidence, authority, or data are insufficient.

Drift: a change in the data, relationship, or use after deployment that may weaken performance.

False alarm: the operational effect of a false positive, that is, an alert that calls for a human response to an event that is not present. It is measured by the burden it places on the worker and by the rate at which alarms are ignored, not by the statistical error rate alone (see: false positive).

Transferred impact: harm or consumption that moves to another stage, resource, or party instead of disappearing.

Automation: carrying out a task according to rules or commands without manual intervention in every cycle; by itself it does not mean the presence of artificial intelligence or full operational autonomy.

Integrated pest management (IPM): monitoring, prevention, and compatible cultural, biological, mechanical, and chemical interventions under local regulation.

Seed bank: the stock of viable weed seeds present in the soil and capable of germinating later; therefore, assessment of control may extend across seasons, not only to the plants visible on the day of intervention.

Raw data: the value as received from the source before normalisation.

Normalised data: a value converted to a standard unit or term with a record of the conversion.

Approved data: a record that has passed review and is fit for a specific purpose.

Evapotranspiration (ET): the sum of water evaporation and plant transpiration, used in estimating water demand.

Triangulation: comparing multiple sources, methods, or points in time to understand a claim and any conflict around it.

External validation: testing the model on spatially, temporally, or institutionally independent data.

Ground verification: on-site measurement or inspection used to interpret or evaluate a remote-sensing product.

Local verification: checking that the data, metric, function, outcome, authority, and service are suitable for the place, season, and user concerned; it is not the same as merely translating the interface or running it once.

Digital transformation: changing process and outcome using data and technology; broader than digitising a paper form.

Diagnosis: determining the cause of a condition on the basis of suitable evidence; broader than classifying an image.

Interoperability: the ability of systems to exchange and use data while preserving meaning.

Explanation: information about the factors influencing a model output; it does not establish causality.

Classification: assigning an input to a class from a defined set.

Machine learning: methods that learn patterns from data to perform a task.

Deep learning: multi-layer models with high representational capacity, requiring data, computing, and tuning.

Reinforcement learning: learning sequential actions from reward signals, with safety risks during exploration.

Digital twin: an updated computational representation of a system, whose value depends on relevance and verification.

Actuator: a physical component such as a valve, pump, or fan that converts a command into action; it is not the human operator.

Minimum safe function: a local function that remains available if a component or connection is lost.

Accuracy: the proportion of correct answers under a given definition; it may conceal class imbalance.

Baseline: the method, service, or outcome that actually exists before the intervention and is used for comparison; not an assumed ideal alternative that the farm does not have.

Spatial resolution: the size of the smallest unit represented by spatial data.

Primary study: research that presents specific data or an experiment, as distinct from a review.

Case study: a description or evaluation of a system in context; it does not automatically generalise.

Repository evidence: code, a test, or an engineering report that establishes a dated implementation state.

Smart farming: the use of data, connectivity, models, and digital services to improve a farm decision; it does not by itself mean artificial intelligence or automated execution.

Precision agriculture: managing spatial and temporal variation through measurement that can be converted into an appropriate treatment, then verifying its execution and outcome.

Vertical farming: arranging production units in vertical layers or across multiple levels; the term alone does not specify the root environment, light source, or profitability.

Hydroponics: growing plants without field soil, in a nutrient solution or a soilless medium that receives water and nutrients; it is distinct from aquaculture of aquatic organisms.

Aeroponics: soilless cultivation in which the roots are suspended inside a chamber and receive water and nutrients as mist or fine droplets.

Protected cultivation: production under a covering that changes how much the plant is exposed to weather, from netting or a tunnel to a plastic house or glasshouse; the covering does not imply full control.

False negative: a present condition that the system did not detect.

Scenario: a hypothetical case for testing a decision; not a certain forecast.

Chain of custody: a record of a sample's or batch's transfer and of who handled it.

Functional safety: design that makes the system move, in the event of a fault or hazard, to a known response appropriate to the consequence, while keeping protection limits independent of model confidence.

Edge conditions: operating cases at the boundaries of the range in which the system was tested, such as deep shade, occlusion, dust, or high density; they must be measured separately from the overall average.

Intelligence sovereignty: governed control over the flow of requests, resources, policy, and records rather than scattered connections.

Uncertainty: limits of knowledge or variability in the phenomenon; it must be represented, not concealed.

Vapour pressure deficit (VPD): the difference between the saturation vapour pressure of water and its actual vapour pressure; it helps explain the driving force for transpiration and is not a fixed universal target.

Pre-harvest interval (PHI): the regulatory time between applying a substance and harvest, according to the label.

Restricted-entry interval (REI): the regulatory period during which entry to an area is prohibited or restricted after application, according to the label.

Portability: the user's ability to take their data and its meaning to another system.

Repairability: the ability of the user organisation to diagnose a fault and obtain the tool, part, technician, and documentation needed to restore function within a time appropriate to the consequence of downtime.

Missing value: the absence of a measurement; it is not zero.

Confidence calibration: how well the probabilities reported by the model match actual outcomes.

Provenance: a record of data origin, transformations, owners, and uses.

Professional localisation: documented adaptation of the decision, agricultural input, place, season, user, language, term, unit, device, connection, institution, source, service, alternative, and outcome, followed by local verification before adoption.

False positive: the statistical term for a case that the system classified as positive when it is in fact negative. Its counterpart in this book is 'false alarm' when describing the operational effect.

Large language model: a model that generates or analyses language; its fluency does not establish truth.

Mode of action: the biological, chemical, or physical way in which a control measure affects the target; repeating the same mode may increase selection pressure even if the product name changes.

Shadow mode: running the model to produce an estimate or recommendation and comparing it with reality without granting it authority to issue a physical command.

Decision window: the period during which intervention can still change the outcome safely and with value; it may be minutes, days, or a season depending on the task.

Functional unit: the common basis to which resources or impacts are attributed for comparison, such as a treated hectare, a weed whose growth was prevented, or a tonne of marketable product over a stated period.

Agent: a software component that selects actions or co-ordinates tasks according to goals and tools.

Appendix F: Topical and Regional Index

The number refers to the chapter that addresses the topic primarily or revisits it within another decision. A regional cross-reference does not mean that the chapter covers all countries in the region or that what it presents is transferable across them; it is an entry point, followed by verification for the country, locality, production system, season, and institution.

Topical index

Evidence for implementation and its limits: 3, 30–34\\. Cybersecurity and service continuity: 5, 10, 16, 22, 27–28, 30, 32–33, 36\\. Sustainability, total resources, and transferred impacts: 1–2, 10–11, 14–15, 18–20, 26, 34–36\\. Retrieval, language models, and advisory services: 4, 7, 23, 25, 27, 29–30, 32–33\\. Aquaculture and integrated farming: 17–19, 26, 34–36\\. Markets, finance, feasibility, and investment: 1, 20–23, 26, 34–35\\. Weeds, herbicide resistance, and robotics: 12–16, 27, 30, 34\\. Diseases, pests, and integrated management: 1, 7, 12, 15, 17–18, 27, 29, 34\\. Data, provenance, and rights: 3, 5, 7–11, 20, 28–30, 32–33, 36\\. Greenhouses, vertical farming, hydroponics, and aeroponics: 10–11, 14, 25–27, 36\\. Soil, land, and crop suitability: 1, 5–6, 8, 11, 14–15, 26\\. Fertilisation and nutrients: 8, 10–11, 14–15, 20, 26, 34\\. Bias, uncertainty, interpretation, and verification: 3–4, 6–9, 12, 23, 27, 29–30, 34, 36\\. Harvest, post-harvest, and food safety: 15–17, 20, 22, 26, 34–36\\. Irrigation and water: 1, 6, 8, 10–11, 14–15, 18–20, 25–26, 34–36\\. Robotics, machinery, and automation: 4–5, 13–16, 20, 24, 27, 30, 34–36\\. Safety, human authority, and abstention: 4–5, 7, 10–19, 22, 27, 29–30, 34–36\\. Supply chain, traceability, and resilience: 5, 16–17, 19, 21–22, 25–26, 30, 34–36\\. Smallholders and shared services: 2, 7, 10, 16, 20–21, 23–26, 35–36\\. Labour, skills, and a just transition: 1–2, 13, 16, 18, 20, 23–24, 27, 35–36\\. Repairability, exit, and rollback: 5, 10, 16–17, 20, 22–23, 25, 27–28, 30, 32–36\\. Local knowledge, translation, and localisation: 3, 5, 7–8, 12, 15, 18–19, 23, 25–29, 31–36\\. Crop monitoring and yield forecasting: 6, 8–9, 15–16, 20–22, 29, 34\\. Research method, experimentation, and stopping: 3, 29–30, 34–36\\.

Regional index

Arab world and the Gulf: 1, 7, 10–11, 14, 20–26, 35–36\\. Turkey: 7, 11, 14, 20–21, 24–26, 31–33, 36\\. Indonesia and South-East Asia: 7, 19, 23, 25–26, 35–36\\. Sub-Saharan Africa: 7, 10, 18–19, 21, 23, 25–26, 35–36\\. South Asia: 7, 10–11, 18, 21, 23–26, 35–36\\. South America: 6, 9–10, 13–15, 20–22, 24, 26, 35–36\\. Contexts with limited connectivity, power, and service: 7, 10, 16–19, 22–26, 30, 35–36\\.

Case study index

Aladdin Agri AI, AiBridge, and Chat V2: 7, 25, 30–33, 35–36, and Appendix N. These passages describe date-stamped internal effects and engineering limits; they should not be read as independent evidence of localisation, live operation, the quality of agricultural decision-making, or field impact.

Appendix G: Release and Translation Provenance

Source corpus (corpus)

The work began from an audited baseline of 96 files, with additional material in For Review, the Chapter21_Unified branches, and multilingual versions. The inventory and fingerprints in the audit logs are the reference for file identity, not the name or language alone. The presence of a version in a country's language does not prove that its examples, units, institutions, and services are valid for that country; provenance preserves the text's origin, whereas localisation requires a separate decision and independent verification.

Consolidation method

English RC1 was used to align structure and provenance, not as a superior authority on local meaning.
Unique material was extracted from the detailed branches and regional lists, while preserving its source and date.
Claims were broken down into existence, property, performance, and impact, and the measured outcome was separated from the promise.
Each type of claim was matched to the source that could substantiate it, and what was missing remained missing.
The narrative was rebuilt in Arabic from the agricultural decision, and the local term was retained where it changed the meaning.
Examples, thresholds, units, and authorities were not carried across regions without qualification, and transferability became a question for verification.
Unresolved claims remained in the quarantine log, and their limits appeared in the main text and appendices.

What the Arabic version does not mean

This version is not a certified legal translation of the references, nor an operating or procurement guide, nor a systematic review, nor an advisory recommendation for any particular country. It is an originally drafted Arabic manuscript built from multi-source materials, in which the questions were returned to the agricultural decision and the limits of the evidence. Its localisation should not be considered complete for every country, crop, or institution; that requires local human review in language, agriculture, law, and operations, and testing appropriate to the consequence before adoption.

Version log

Baseline for external evidence: 2 September 2026, with later controlled additions documented in the references register.
Last reviewed internal distribution artefact for the plugin: package Koot-EL-Kuloob-2.1.39.zip generated on 22 September 2026 and reviewed on 10 October 2026 under IMP14.
Manuscript status at this appendix: expanded working draft; professional rewriting does not, by itself, amount to independent publication acceptance.
Condition for transition to release candidate: consistency across chapters, tables, references, and figures; resolution of conflicts; then documented human reviews covering language, applied domains, localisation, and accessibility.

Evidence density by chapter

Evidence is not distributed evenly across the chapters, but citation counts do not measure quality or local suitability. The following table is an automated inventory of the current manuscript: it counts each occurrence of an SRC or IMP identifier, then counts the distinct identifiers within each chapter. It separates accessible external references from internal implementation evidence unavailable to the reader, and it does not classify sources or judge their fit to the claim; that is a sentence-level human review task.

Chapter	SRC citations	Distinct SRC sources	IMP citations	Distinct IMP evidence
1	6	6	0	0
2	7	7	0	0
3	2	2	0	0
4	6	6	0	0
5	3	3	0	0
6	4	3	0	0
7	27	16	10	4
8	1	1	0	0
9	8	7	0	0
10	25	19	0	0
11	28	25	0	0
12	18	16	0	0
13	7	7	0	0
14	23	23	0	0
15	24	8	0	0
16	22	14	0	0
17	9	5	0	0
18	29	12	0	0
19	4	2	0	0
20	6	3	0	0
21	0	0	0	0
22	4	4	0	0
23	3	3	0	0
24	6	3	0	0
25	1	1	1	1
26	1	1	0	0
27	2	2	0	0
28	2	2	0	0
29	1	1	0	0
30	1	1	12	7
31	19	14	0	0
32	15	15	28	8
33	45	30	23	8
34	19	19	2	2
35	0	0	0	0
36	3	3	0	0

Zero citations do not mean that a chapter is wrong, but they do prevent it from being used to support a claim of performance, impact, or transferability. Nor does a high citation count mean that a chapter is valid for a given country or crop; the references may be domain reviews or official sources that describe a function without testing a local outcome. These numbers should therefore be read alongside the type of argument, while matching the claim to the source, comparator, place, season, and outcome remains an independent gate.

What the table reveals

Three observations bind reader and reviewer alike:

First: Chapters 15 and 16 do not contain primary regional or field studies that directly match their applied decisions. Chapter 16 currently relies on a scientific review, institutional references, and official vendor pages that map market functions, not on independent verification of product performance or on local testing. This is a declared gap, not a hidden error, but it remains a gap: neither chapter may be used as a basis for an advisory recommendation before context-matched primary sources are added and performance is verified locally.

Second: Chapters 31–33 now cite accessible external references for the general principles that govern the case study: transparency, robustness, accountability, generative-AI risk management, pesticide governance, provenance, data quality, internationalisation, translation quality, and locale handling. Those sources give the reader an independently inspectable basis for judging the proposed rules. They do not independently verify any Aladdin-specific capability, release state, price, reach, localisation result, completed user journey, or agricultural outcome. Factual claims about the platform itself therefore remain bounded by the private implementation evidence IMP01–IMP14 recorded in Appendix N.

The distinction is substantive. Part Nine now has public normative and technical grounding, but it still does not contain an independent evaluation of Aladdin’s operation or agricultural value. A reader may use the external references to challenge the criteria applied to the case; the reader may not use them as a substitute for field trials, user studies, release verification, or an independent examination of the platform.

Third: the inventory shows substantial unevenness, and applied chapters with a limited primary base require qualitative review, not the accumulation of citations without matching. This is the broadest gap in the work, and it determines the condition for transition to an approved edition in the next section.

Source-base gate for the approved edition

This edition is based on two hundred and eighty external references for thirty-six chapters covering computer vision, soil, water, nutrients, diseases, robotics, protected cultivation, livestock, aquaculture, harvesting, food safety, economics, markets, supply chains, inclusion, six regions, governance, and reliability. The source base was strengthened by adding sources on Farmer.Chat and GAIA, followed by sources on crop monitoring, yield forecasting, and irrigation management; references on nutrient management, fertilisers, and modern production systems; four specialized references on crop rotation, its mechanisms, and its techniques; seven specialized references on animal welfare and precision livestock farming; three references on food loss, feed safety, and integrated agriculture–aquaculture; nine references on the diversity of harvesting systems and the landscape of their software and machinery; and three references on agricultural labour, the just transition, and gender. On 15 October 2026, seventy-two references were added to cover the development of Chapter Nineteen: seven scientific papers, fourteen professional references, thirteen technical documents, thirty-seven commercial descriptions, and one research-project report. On the same date, seven original references were added for the development of Chapter Twenty-Two: four commercial descriptions, two technical documents, and one professional reference, with SRC216 reused for document processing and a crosswalk to the chapter’s sections. Then, on 19 October 2026, fifty-one unique external references resulting from the development of Chapters Twenty-Three to Thirty were added, with existing references reused where the source was the same and the duplicate NIST reference in the Chapter Twenty-Nine draft not counted a second time. On 22 October 2026, fifteen external references used in Chapters Thirty-One to Thirty-Three were registered to support the public principles governing the case study; they do not verify Aladdin-specific implementation or impact. These additions strengthen the documentation of available functions, methods, and governing criteria; they do not establish local performance, service readiness, or suitability as a substitute. Even so, the source base remains limited relative to the breadth of the book’s subjects, and verbal caution alone does not remedy that unevenness.

The figure of twenty-four was the accurate one after the nine implementation-evidence items had been removed from the reference count. Fourteen accessible public sources were then added to document Farmer.Chat and GAIA, bringing the total to thirty-eight. After that, twenty-three public references were added to document the tools, methods, and new studies in Chapters Nine and Ten, bringing the total number of external references to sixty-one. Then twenty-four new references specific to plant nutrition, nutrient management, protected cultivation, modern systems, and the tools related to Chapter Eleven were added, bringing the total to eighty-five. Subsequent public additions raised the number to one hundred and nine; SRC124–SRC127 were then added to develop Chapter Fifteen, bringing the total to one hundred and thirteen; SRC128–SRC134 were added to develop the livestock chapter and responsible animal care, bringing the total to one hundred and twenty; SRC135–SRC137 were added for food loss, feed safety, and agriculture–aquaculture integration, bringing the total to one hundred and twenty-three; SRC138–SRC146 were added to develop Chapter Sixteen, bringing the total to one hundred and thirty-two; and SRC147–SRC149 were added for agricultural labour, the just transition, and gender, bringing the total to one hundred and thirty-five, without restoring any private repository artefact to the list. On 15 October 2026, entries SRC150–SRC221 were then added to document the development of Chapter Nineteen, bringing the total to two hundred and seven external references, with SRC117 reused without duplication and all earlier identifiers, reserved gaps, and the historical inventory in Appendix S preserved through 31 August 2026. Entries SRC222–SRC228 were then added to document the tracking, forecasting, optimization, monitoring, simulation, and risk-management functions in the development of Chapter Twenty-Two, bringing the total to two hundred and fourteen external references. SRC216 was reused without creating a duplicate entry. On 19 October 2026, fifty-one unique external references were added for the development of Chapters Twenty-Three to Thirty, distributed across entries SRC229–SRC281 with the two deliberate gaps, SRC235 and SRC251, bringing the total at that stage to two hundred and sixty-five external references. SRC251 was not counted because its material is the same as SRC033; SRC235 remained unassigned; and all previous entries and the historical inventory in Appendix S remained preserved. On 22 October 2026, entries SRC282–SRC296 were registered for the development of Chapters Thirty-One to Thirty-Three, bringing the current total to two hundred and eighty external references. By contrast, the figure of thirty-three that appeared in earlier drafts was inflated by material that could not be externally verified and is not used as the basis for the current count.

Caution prevents overclaiming, but it does not create coverage. When an entire applied chapter depends on a single scoping review, the text can be honest about its limits and incomplete in its source base at the same time; these are two different shortcomings, and neither compensates for the other.

Accordingly, any edition described as authorised — not this draft — must satisfy at least the following minimum requirements:

Requirement	Minimum requirement	Applies to
Preliminary studies for each applied chapter	Two peer-reviewed sources matched to the task	Chapters 8–25
Regional source	At least one reference for each region mentioned	Chapter 26
Evidence independent of the sponsor	One non-repository source for each claim of capability	Chapters 31–33
Upgrade of implementation evidence	Persistent identifier (DOI) \+ tagged release \+ CITATION.cff for each IMP entry	Appendix N
External verification	A verification or reproduction study for each published performance figure	Every figure in the main text
Specialist review	One independent domain reviewer for each sensitive chapter	10, 11, 12, 16, 19, 24, 27

Chapters that do not meet their minimum requirement may be published in only one of two ways: either their rank is explicitly lowered to ‘a design framework that presents no performance evidence’, or they are deferred to a later edition. The common third option is not permitted: leaving the chapter in its current form and adding a general reservation in the introduction.

Share of repository sources in the evidence base

Earlier drafts carried nine sources out of thirty-three — about twenty-seven per cent of the evidence base — that were in fact an examination of the sponsoring entity’s own repository. This was corrected in the current edition: those nine items of evidence were removed entirely from the reference list and transferred to a separate register with identifiers IMP01–IMP09 in Appendix N.

IMP10–IMP12 were added to the register on 16 September 2026, then IMP13 on 18 September to document the AgroGenie development package and its local tests, then IMP14 on 10 October after examination of the plug-in’s official distribution package. These artefacts did not enter the reference list at any stage, and therefore do not change the number of external references.

The numerical result after correction and update is this: the book’s references comprise two hundred and eighty external references, all published and accessible via a DOI or a stable link. The reference list itself contains no private material that cannot be externally verified.

The reason for the correction is methodological, not cosmetic: an implementation artefact that the reader cannot access is not a reference, and listing it among the references gives it an authority it does not possess and inflates the total count with material that cannot be examined.

Implementation evidence IMP01–IMP14 is used where a dated engineering case needs to be described, while keeping each artefact separate from the others. These items remain private implementation evidence, not public references or evidence of live operation, completed localisation, or agricultural impact. The accessible sources cited in Part Nine independently support the criteria used to examine the case, but they do not independently verify that Aladdin satisfies those criteria. The methodological tension is therefore narrower and more precise: the part that discusses the sponsor’s product has public external grounding for its principles, yet still lacks independent evidence for the product-specific claims and outcomes.

The book addresses this paradox through three declared limits in the editorial independence statement and in Chapter Thirty-One. But those limits govern the wording; they do not alter the nature of the evidence.

The explicit methodological position is this: repository evidence establishes existence and structure. It does not establish operation, impact, or quality.

Anyone wishing to judge the platform’s agricultural value will not find enough in this book. The book presents no independent field trial, user study, release verification, or outcome evaluation sufficient for that judgement. This limitation is disclosed rather than filled with inference from general standards or with confident language that the evidence cannot support.

Data and materials availability statement

External sources: every reference in the ‘controlled references’—two hundred and eighty in total—is available through the digital object identifier (DOI) or the stable link recorded in its entry. The references section contains no source that is not publicly accessible, after the implementation evidence was removed from it and transferred to Appendix N.

Implementation evidence (IMP01–IMP14): IMP01–IMP09 concern a dated private source artefact from 22 August 2026; IMP10–IMP12 concern separate internal artefacts; IMP13 concerns the AgroGenie development package; while IMP14 concerns the official distribution package generated on 22 September and examined on 10 October. Each layer has its own version or date, fingerprint, and limits in Appendix N, and a later layer does not erase the judgement on an earlier one. These artefacts are unpublished and carry no public persistent identifier; therefore they do not enter the reference list and do not support any claim of performance, acceptance, localisation, or agricultural impact. The route to verification is independent examination of the specified artefact, and the conditions for its upgrade are stated in Appendix N.

Internal editorial records: the source-code inventory and the logs for verifying claims, statistics, visual assets, and integration — internal working records, not public scientific references, and not cited to support a finding.

Generated data: this work did not produce new experimental data, conduct field measurements, or reproduce a published result. Every figure in it is taken from its source with its declared limits.

Funding and conflict-of-interest statement

The sponsoring entity: Aladdin Agri AI, which is also the subject of the case study in Part Nine. This is a direct and declared conflict of interest, addressed not through denial but through the three limits set out in the editorial independence statement and through a strict separation between accessible external references and private implementation evidence. The external sources make the governing principles independently inspectable; they do not make the sponsor’s testimony independent or verify the platform-specific claims bounded by IMP01–IMP14.

Editorial decision-making authority: the sponsor has no authority over the book’s findings, over the removal of any limitation, or over the wording of any conclusion. Where a marketing description conflicts with a primary source or with the state of the repository, priority is given to the higher-authority source and to the more cautious description.

What was not funded: the work received no funding from any supplier or product mentioned in Appendix H or Appendix S, and no party paid in exchange for the inclusion of a name or a case, or for raising an evidence grade. A product’s presence in the book is not an endorsement, and its absence is not a negative judgement. This neutrality applies to the sponsoring entity’s own entry in Appendix S (AIP019) just as it applies to others.

Statement on the use of artificial intelligence

What the tool was used for: organising materials, detecting duplication across versions and languages, structural comparison of files, suggesting wording, and checking the internal consistency of terminology, cross-references, and figures.

What it was not used for: no scientific finding was generated, no reference was created, no figure was estimated, no agricultural judgement was inferred, and no missing field in a case card was filled. Every claim in the book is traceable to a human-verified source.

Limits of the tool: it was not treated as a scientific source or as an agricultural authority. Fluency of wording does not prove the correctness of the content, and this is the same rule that the book applies to the systems it analyses.

Responsibility: responsibility for verification, editing, and final approval remains entirely human and does not pass to the tool under any circumstances.

Appendix H: Landscape of applications, products, and programmes

Purpose of this appendix

This appendix preserves the nominal details found in the expanded branches instead of dropping them from the book, but it does not turn a name into evidence or into a procurement list. Each case is read across six separate layers: the appearance of the name in the book’s material; resolution of its identity and of the entity responsible for it; the existence of a specific implementation; its state of availability in the intended place and time; its suitability for the language, unit, device, connection, and local service; and what an independent evaluation has established in an agricultural or operational outcome. One layer may be established while the next is not.

Anything for which no verified source appears in the reference register or in an implementation register with declared limits remains only a source lead, requiring a dated official page and identity resolution. Even after existence is established, it is not used as evidence of accuracy, impact, or local suitability without an evaluation proportionate to the claim. The availability of an interface in a given language does not mean that the agricultural terminology, warning, unit, or support pathway has been professionally localised.

Its relationship to Appendix S — a division of labour, not duplication: this appendix is narrative and broader in scope; it preserves everything that appears in the book’s source corpus of names, cases, source leads, concepts, and technology families, and explains each in the language of the chapters. Appendix S, by contrast, is a guide restricted to a single unit: the implemented programme; nothing enters it unless it has met seven conditions, and every entry there has an official source and a dated availability status.

What appears here and not there — a study, a review, a simulation, a concept, or a name whose attribution has not been settled — is not omitted, but placed in its proper rank. Those seeking explanation and context will find them here; those seeking programmes that can be verified and compared should go to Appendix S.

The localised reading rule

A status from this appendix must not be carried over into a field decision merely because the crop or climate seems similar. Before any comparison, the required task must be identified, along with the agricultural material, its stage, and its local name; the place, season, and decision window; the user, their authority, language, and unit; the device, power, and connectivity; the reference body; and the availability of training, technical support, spare parts, and a fallback. Anything not verified in these fields remains ‘unknown’; it must not be filled in from a vendor description or by automated guesswork.

Farm and production management

Cropin Cloud / Cropin Intelligence

They appear in the source corpus as a family of platforms for production management, analytics, and models. This case is useful for discussing the agricultural cloud and the linking of multiple fields. Claims such as ‘the world’s first’, and counts of models, crops, acres, and data points, are not published without a dated document and a separately stated definition. Status: vendor description/source lead.

Agrotics

It appears as a SaaS platform for smart agriculture and operations management. It is retained as a case for the subscription model and integration, and any evaluation must first document the version, modules, market, language, data export capability, cost, and user outcomes. Status: commercial source lead.

Terra Oracle AI

It was cited as a multilingual agricultural adviser. Its value lies in testing the quality of terminology, sourcing, and abstention across languages, not in assuming that multilingual functionality has been achieved in practice. Status: source lead; a formal, current test is required.

AgriNEXT

It was cited as a system combining satellites and the Internet of Things. Observation data should be separated from analytics and from commercial services, and the partners and coverage should be verified. Status: source lead.

FarmMind

It was cited among agent applications for farmers. It must be specified whether the ‘agent’ is a conversational interface or a tool-using system, and what permissions it has and what action log it keeps. Status: source lead not qualified for an impact claim.

Plant diseases and health

AGMRI

It appears as an AI application for crops. The name alone is not enough to identify the company, version, or study; the identity must be resolved before public release. Status: unresolved source lead.

CropGPT

It has a specific peer-reviewed study \[SRC008\] supporting the existence of a multimodal model and the scope of its diagnostic and explanatory functions. The inherited figures remain quarantined until the full text has been examined. Status: partially controlled scientific example.

TatarAI

It was cited as a mobile application for plant health. An official source and a study covering the data, diseases, and geography are required. No accuracy figure is to be transferred from an inherited table. Status: source lead.

Mobile CNN for Maize

It has a primary source \[SRC007\] establishing the mobile classification task. The figure of 95% is not currently accepted. Status: partially controlled scientific example.

Three-tier multi-crop framework

It was cited as a multi-level, multi-crop diagnostic framework. It is not to be treated as a product or as a specific study until the reference has been resolved. The idea remains architectural: screening, then verification, then guidance. Status: concept/source lead.

Yield forecasting

Yield-XAI study

The study \[SRC006\] establishes identity and scope, with R² up to 0.92 in its own setting. It is used to illustrate the linkage between prediction and explanation, and the limits of generalisation. Status: controlled primary study.

Stability zones \\+ ML

The study \[SRC005\] establishes the integration of stability zones and machine learning. No result should be imported if its text has not been examined. Status: primary study, scope only.

IoT/ML crop forecasting frameworks

The drafts include several frameworks combining sensors, weather, and models. They are retained as a technology family, and their results are not merged into a single value. Status: family of source leads.

Soil and nutrients

AI-enabled soil fertility frameworks

Fertility frameworks are emerging that link sensors, analysis, and decision-making. Their value lies in classifying components, but any recommendation requires a sample and a local standard. Status: a family of concepts; the preliminary evidence is insufficient.

Smart sensor fusion for nutrients

This illustrates the idea of combining several sensors to reduce reliance on a single signal. The laboratory reference, the soil context, and transferability must all be verified. Status: research lead.

Field-scale soil-moisture prediction

It represents the use of time and space in estimating moisture, and it is kept separate from the irrigation decision. Status: a family of studies that requires a clearly specified record.

Weeds and robotics

Soybean ConvNet

Study \[SRC009\] concerns weed detection in soybean images using convolutional networks. It is neither Q-learning nor a removal robot. Status: a narrowly scoped vision study, used to demonstrate the perception layer rather than the execution chain.

Carbon Robotics LaserWeeder / Large Plant Model

It is presented as a laser product and a vision model. Its claimed capability should be presented only after reviewing the supplier’s dated documentation, and the results require independent evaluation. Legacy weed-removal/herbicide figures are quarantined. Status: commercial-source lead.

Autonomous solar-powered lightweight robot

It represents a lightweight design and perhaps a service suited to farm holdings, but the name, source, and testing must be resolved. Status: research/product lead.

EM-GROW

It was cited as a robotics project for organic farms supported by space technologies. It requires an official programme page and results. Status: programme lead.

Harvesting

Eternal.ag tomato harvester

It was cited as an autonomous tomato harvester. The claim “fully autonomous”, or any figure for operating hours, must not be published before verifying the version, location, time of availability, and damage. Status: commercial lead.

Strawberry robot with silicone fingers and fan

It represents the challenge of gentle gripping and fruit quality. It requires a specific paper/programme and post-storage metrics. Status: research lead.

Occluded cucumber harvesting

It represents occlusion, three-dimensional estimation, and reach through foliage. Commercial maturity must not be inferred. Status: research lead.

Apple harvesting and quality inspection

It represents the integration of harvesting and sorting. Picking performance must be separated from quality inspection. Status: source lead.

Irrigation and water

AS-MO precision irrigation

Study \[SRC038\] is the controlled case: a design that combines automatic scheduling with manual operation, alongside stakeholder evaluation in Kenya, Jordan, and Morocco. It does not establish a savings percentage. Status: design and user study.

Smart irrigation scheduling / IoT irrigation

Families of studies link sensing, prediction, and control. Each case must be evaluated against crop, water, yield, energy, and baseline. Status: a source family, not a unified result.

Livestock and aquaculture

PoultryFI

Preliminary draft \[SRC015\] describes a multi-sensor platform for welfare and productivity. It does not constitute a production standard. Status: emerging evidence.

BirdWatch

It was cited as a poultry-health monitoring system with a space link. The name requires identity and source resolution, and it must not be described as a deployment without that. Status: source lead.

BroBot

It was cited as a poultry-monitoring robot. The platform, its status, and its safety must be documented. Status: source lead.

Aquaculture predictive systems

Summarised by review \[SRC010\] at the domain level: water, feeding, disease, decision-making, and edge computing. There is no single platform or general cost percentage. Status: review evidence.

Extension

Digital Green FarmerChat

Documented by the programme, release, and open-data pages \[SRC039\]–\[SRC046\] as a multilingual, multimodal assistant linked to local extension services. There is also a user study covering 450 farmers in Kenya; it allows description of use, trust, and self-reported action, but does not establish general agricultural accuracy or a causal effect on yield and income. Status: institutionally operational service, with geographically limited evaluation evidence.

Vayazh

Mentioned as an adviser using RAG. Its identity must be resolved, and its knowledge sources, language, and safety must be stated. Status: source lead.

Kisan AI

Mentioned as an adviser that takes profitability into account. It requires documentation of pricing, data, and scope, with governance for financial projections. Status: commercial lead.

CottonBot

Mentioned as an LLM assistant for cotton. The source, crop focus, geography, and testing need to be examined. Status: research/product lead.

Agro Bot

Mentioned as a use of ANN and NLP in advisory services. The name is generic and requires identity resolution. Status: open lead.

GAIA

Documented by IFPRI and CABI pages, and by testing, governance, and licensing materials \[SRC047\]–\[SRC052\] as a multi-partner research and development programme with prototypes, user testing, and governance documents. The evidence supports describing the programme and its early stage, but does not establish a complete commercial product or broad field impact. Status: documented programme with limited early results.

AgroGenie

The 22 August artefact establishes an optional integration connected to the front end and to the conversation path, with an acceptance constraint specific to that package \[IMP06\]. The subsequent AgroGenie development package establishes successful local tests for search, languages, governance, and integration, without establishing WordPress, database, browser, or load operation in a live environment \[IMP13\]. As for the official distribution package of the generated plugin dated 22 September, it establishes the presence of contracts and components within a standalone distribution artefact, and remains limited to what \[IMP14\] recorded; it does not overturn the August judgement, nor does it establish AgroGenie activation, completion of its localisation, or field impact. Status: dated internal implementation layers, each with a different version, environment, and limits.

Sustainability and protected agriculture

Cropin sustainability tools

Mentioned as impact/carbon-tracking tools. They require a methodology, system boundaries, data source, and verification; a dashboard display is not sufficient. Status: vendor description/lead.

CinSOIL

Mentioned as a soil-carbon measurement/input case. The methodology, baseline, permanence, and leakage must be verified. Status: source lead.

Farmdee-Mesook

Mentioned as an emissions-awareness platform in a smart-agriculture context. It requires an identity, a formal programme, and results. Status: source lead.

AI \+ 5G greenhouse

It represents an integration of connectivity and control, not evidence that 5G is required or that the system is sustainable. Status: application family.

Supply chain and food safety

Bai GenAI \+ blockchain

A simulation study \[SRC011\] for supply-chain coordination and trust. It does not describe deployment. Status: controlled simulation.

Khanna cold-chain framework

A computational study \[SRC012\]; the 30% travel-time result is supported only within its experiments. Status: constrained experiment.

Spectral-AI

A review \[SRC013\] maps the field. It does not establish carry-over accuracy. Status: review evidence.

Cloud-based grain-quality monitoring

Mentioned as a family of cloud-based monitoring systems. The 96% figure is withheld pending identification of the task and study. Status: source lead.

Appendix update rule

When verifying any source lead, it is not moved directly to ‘Approved’. A card is completed that separates the name, organisation, version, source, and cut-off date from implementation, availability, independent evaluation, and impact. A local suitability card is then added: the agricultural task; the place and season; the user, language, and unit; the device and connectivity; the authority, service, maintenance, and fallback; and the outcome to be measured. If a source establishes only that a feature exists, availability, suitability, and impact remain ‘unknown’. If a primary source contradicts the vendor’s description, the contradiction, its source, and its date are presented instead of being resolved through middle-ground wording or an automated summary.

Appendix T: Practical workbooks

These templates are neither formal questionnaires nor ready-made recipes. Their purpose is to prevent a team or a farm from jumping from a general problem to purchasing a system before defining the decision, the evidence, the responsibility, and the context in which it will operate. The same template may be suitable in more than one place, but the values, thresholds, terms, authorities, and continuity plans do not transfer with it automatically.

Localized context card — complete before any workbook

Agricultural commodity, its stage, and its local name: Place, production system, season, and decision window: Actual user, role, authority, language, and terminology: Unit, calendar, device, power, and connectivity: Professional or regulatory body responsible for the reference standard and accreditation: Local service: training, technician, spare parts, response time: Fallback in the event of interruption, refusal, or objection: Required outcome, the consequence of error or delay, and who bears it:

If any element affecting the decision changes — such as the crop, season, language, unit, authority holder, or service availability — the card must be reviewed before the record is reused. AI does not turn an unknown field into a fact; it can reveal gaps or compare records, but establishing context, authority, and fallback is field and institutional work.

1\. Problem definition card

Decision name: Crop/animal/material, its stage, and its local name: Place, season, and action window: Actual user: Language of use, unit, device, and connectivity: Authority holder: Decision frequency and window: Current practice: Loss/opportunity by unit and period: Consequence of delay: Consequence of a false positive: Consequence of a false negative: Information available at the moment of decision: Possible action after the output: Governing authority or rule: Service and fallback if the system is unavailable: Why the technology may be better than changing the process:

Acceptance gate

The project does not proceed unless the user, context, baseline, action, authority holder, and fallback are known. The phrase ‘improving productivity with AI’ is not a knowledge problem, and similarity between two crops or two regions is not enough to transfer a decision from one to the other.

2\. Dataset card

Section	Recording fields
Identity	Name, version, owner, contact point
Purpose	Reason for collection and permitted/prohibited uses
Scope	Crop/animal, local name, and stage; place, production system, and season; device, channel, and connectivity; user category and language
Sample unit	Image/plant/field/day/batch/animal
Collection	Protocol, timing, and instrument
Expert-verified annotation	Definition, expert, agreement, unresolved class
Local meaning	Original term, linked standard name, unit and calendar, body responsible for the threshold definition
Quality	Missingness, noise, exclusion, calibration
Representation	Distribution of reference classes, regions, and languages
Privacy	Identity, place, re-identification risks
Rights	Consent, rights holder, licence, training use, sharing, correction, objection, deletion, local benefit
Processing	Transformations, code, and versions
Split	Training/validation/test and unit of separation
Limits	Where are the data not suitable?

3\. Model experiment card

Hypothesis: what change is expected, and why? Baseline: the actual local method, rule, or expert currently relied on. Metrics: according to the consequence of error, with units. Partitioning: spatial/temporal/farm/user. External test: a set not used in development. Transfer limits: which place, season, material, user, device, language, and institution were not tested? Calibration: how was confidence assessed? Unknown class: how does it abstain? Resources: time, memory, energy, connectivity. Disparity: results by region/language/size/device. Negative results: representative failure cases. Stop condition: harm threshold or absence of benefit.

4\. Field pilot card

Design

Site, crop/species, and season;
local name, stage, and decision window;
participants and selection method;
roles, authorities, language, unit, device, and connectivity;
baseline or comparison;
duration sufficient for the decision cycle;
training and support;
technician, spare parts, response time, and local alternative;
consent and data rights;
safety and insurance plan;
pre-registration of outcomes.

Operations

availability and outage rates;
data quality and missing data;
model updates;
false alarms and false negatives;
human interventions;
failures and maintenance;
escalation time and continuity of the critical function during an outage;
user withdrawal and the reason for it.

Results

model metric;
agricultural outcome;
economic outcome;
total water/energy/inputs;
labour, skills, and safety;
disparities across groups;
scalability and exit;
contexts that remained outside the evidence and to which the result must not be transferred.

5\. Irrigation decision logbook

Date/sector	Crop stage	Moisture and depth	Weather/rain	ET/assumption	System command	Worker decision	Actual water	Energy	Note/outcome

The local threshold, its source, and its approving authority are recorded, along with the reason for manual override and who holds that authority. At the end of the period, quantity, yield, quality, and energy are compared; savings must not be inferred from valve time alone, and a threshold must not be transferred between soil, crop, and season without verification.

6\. Plant disease logbook

Case	Crop/stage	Location/date	Required images	Symptoms/spread	System probabilities	Confidence/abstention	Expert/laboratory verification	Action	Outcome

It separates the 'model class' from the 'confirmed diagnosis' and from the 'intervention'. The local names of the symptoms and the likely causal agent are recorded, along with the authority authorised to diagnose or refer. If a substance is used, the label, version, and local authorisation are recorded, without copying a general prescription into the book.

7\. Field robot logbook

Task	Density/occlusion	Conditions	Detection	Localisation	Successful action	Crop damage	Intervention	Stop	Energy/input

This structure prevents vision-accuracy rates from masquerading as removal or harvest performance. Local safety limits are recorded, along with who holds stop authority, and how the task continues or stops safely if the technician, the required part, or connectivity is unavailable.

8\. Animal/Aquaculture Logbook

It records the species, local name, age/stage, environment, device, calibration, alert, interpretation, specialist examination, action, welfare, and outcome. For critical measurements, a locally defined threshold of known provenance, approved by the authorised body, is used, together with a manual mode or care alternative that does not depend on the cloud. Cameras are reviewed from the standpoint of worker privacy and the worker’s authority to correct and stop.

9\. Cold-Chain Logbook

Batch	Point	In/Out	Temperature/Humidity	Calibration	Opening/Shutdown	Route	Alert	Decision	Final quality

Batch linkage is tested during merging and splitting, and the names of sites, the product, grade, units, and the local authority responsible for recalls are recorded, and an actual recall drill is conducted. Quality is not inferred from travel time alone, and the product’s destination or specification is not changed to make the route appear successful.

10\. Labour Impact Logbook

Task before	Who performs it	Time/Risk	Task after	New skill	Training	Who benefited	Who lost	Transition measure

Measurement is repeated after stabilisation, because the initial learning period does not represent steady operation. The groups that lost a task or were given an added monitoring burden are consulted.

11\. Exitability Test

Request a complete copy of the data.
Verify the raw data, the normalised data, and the metadata.
Review the format, dictionary, and identifiers.
Import a sample into an independent system.
Examine relationships, dates, and units.
Revoke a test key and confirm safe operation.
Request a statement of what will be deleted or retained.
Calculate the time, cost, and stoppage.

The test is carried out in the format, language, unit, and device available to the actual user, and it records who provides local help if the export cannot be understood or restored.

Failure of the test makes ‘portability’ an unproven promise.

12\. Cascading Failure Exercise

Scenario: the identity provider fails on a day of high heat, and several houses need ventilation. Questions: do the thresholds work locally? Who receives the alert? Does manual mode require a login that is down? Do the messages depend on the same provider? What is the recovery time? Required result: a dependency map, a mitigation procedure, a responsible person, and a retest date.

The scenarios are repeated for sensor failure, stale price data, batch misidentification, a bad model update, and a supplier outage.

13\. Total Resource Assessment

The system boundaries and period are defined, then the following are recorded:

Direct and indirect water;
Electricity, fuel, and energy source;
Manufacture and transport of devices, and their service life;
Replacements, batteries, and waste;
Inputs and chemicals;
Changes in area or production cycles;
Impact shifted to the supply chain or region;
Production and quality outcome.

The total quantity and the intensity per unit are presented. If one improves and the other worsens, the trade-off is shown.

14\. Translation and Terminology Review

Item	Reviewer’s question
Term	Is it natural and agriculturally specific?
Local name	Has it been retained and linked rather than erased?
Unit	Are its direction and separator correct?
Warning	Has it remained strong and clear?
Authority	Is the local authority explicit?
Example	Has it been transferred to another region without justification?
Interface	Are RTL, audio, and font readable?

15\. General Procurement List

Problem, baseline, and outcome;
Feature definition and version;
Independent study or evaluation plan;
Data, privacy, and rights;
Interoperability, export, and exit;
Security, incidents, and updates;
Support, maintenance, and parts;
Demonstrated local availability of a technician and parts, response time, and a fallback for the critical function;
Total cost and financing;
Training, work, and fairness;
Termination conditions and penalties;
Preservation of evidence after the contract ends.
Rights to object, correct, export, transition, and exit in users’ languages.

16\. Gate for publishing a case in the book

A case may not be promoted from a source lead to a case published as evidence until after:

resolving identity and version;
determining the source type and conflicts of interest;
obtaining the original method and result;
determining the context, comparison, and denominator;
separating model output from impact;
examining transferability and limitations;
writing and reviewing an Arabic card;
updating the claim register and reference.

17\. Human decision record

Localised context and its card version: Automated output: Sources/version: Confidence and limitations: Human decision: accept/modify/reject/escalate. Role, authority, and why the time and information were sufficient: Reason: Action: Outcome: Does it require a rule/model update?

The record allows genuine learning from disagreement and prevents the human from becoming a silent stamp. The program must not attribute the decision to ‘human review’ if the reviewer does not understand the language or the unit, or lacks time, authority, or a practical alternative.

Appendix Y: Scientific Quarantine Register and Reopening Claims

Why keep a rejected claim?

Silent deletion erases the editorial history and may allow the figure to return from another version or from an automated summary that does not carry its limits. The quarantine log records the text, the value, the purported source, its version, its context, the reason for blocking, and what is required for reopening. ‘Quarantined’ does not necessarily mean the result is impossible; it means it is not fit for publication at the required confidence level, or not transferable to the intended local context.

Quarantined quantitative claims

Inherited claim	Reason for quarantine	What is needed for reopening
CropGPT 93.1%	Full text/result location not recovered	Original table, task, data, split, and reference classes
VGG16 95%	Figure not verified against the study	Original location, context, and test
Weed detection 92–97%	Studies and metrics conflated	Source for each value and task
Robotic removal 80–85%	No comparative source chain	Field test and definition of removal
Soil disturbance reduction 70–80%	Unverified vendor claim	Independent measurement and baseline
Pesticide reduction about 75%	Definition/source insufficient	Substance, area, effectiveness, and impact
Irrigation 30% water/fertiliser	No matching crop/place/period/source	Primary study and full balance
Yield increase 20–25%	Attributed to a source that does not substantiate it	Primary result and comparison
Aquaculture 30% cost	Review \[SRC010\] does not support it	Primary cost study
Spectrum 99.85%	Review does not identify the origin	Primary study, task, and risk
Grain quality 96%	No matching source	Data, categories, and test
Harvester 22 hours/day	Unverified operating description	Availability, downtime, and context log
400 crops/10,000 varieties/millions of data points	Vendor figures without version control	Dated document, definition, and verification
‘A fraction of conventional cost’	No denominator, currency, or context	TCO and comparison
‘First globally’	Priority claim cannot be settled	Market survey, definition, and date
Millions of hectares	No definition/independent source	Dated report, method, and area

Methodological claims under quarantine

That the book systematically reviewed more than 150 sources: there is no complete search-and-screening record.
That mixed triangulation was carried out in the present study: it is a future protocol.
That the language files constitute independent evidence: translations do not count as additional instances of the source.
That a large number of citations proves the quality of a claim: what is required is claim-level matching.

Conditional claims about Aladdin

The following historical claims are to be read against the source artefact Koot-EL-Kuloob-2.1.39 dated 22 August 2026 \[IMP01\]–\[IMP09\]. Neither the subsequent development package for AgroGenie \[IMP13\], nor the official distribution package generated on 22 September \[IMP14\], retroactively rewrites the status of that artefact. Each layer has its own date, fingerprint, test type, and inference limit.

All services are live and approved for production: not established.
The seventeen cards are always active: this depends on the feature, entitlement, and adapter.
The desktop catalogue of twenty-two items is equivalent to twenty-two ready workspaces: this contradicts the source’s own classification.
The ten languages passed human and operational review: not established.
AgroGenie is active in every installation: both gateways are closed by default.
AgroGenie passed its bundled test in this historical artefact: not established — two required cache files were missing, which halted the test. This judgment is not carried over to the current development package, which passed its local tests according to \[IMP13\].
Humanizer is a stable core layer: its linkage is conditional, and the class is not included.
The phone app is complete and all AI works offline: this contradicts the implemented contract itself.
Bidirectional synchronisation for the phone works: the contract declares it unsupported, and the payment path returns a ‘Not Implemented’ status.
The phone client is bundled in the package or available in a store: there is no executable trace of it in the package.
The desktop is production-signed or ready for the store: not verified.
An annual price is published and in use: proposed material only.
The release as a whole is fit for publication: the readiness log records INCOMPLETE / NO-GO.
The quality of the agricultural answer is established: there is no independent evaluation.
The official package proves that all functions are enabled and operating live: not established; \[IMP14\] proves only the distribution artefact, the contracts, and the specified tests, and does not prove that WordPress is live, the status of flags, vendor connectivity, interface acceptance, localisation completeness, or field impact.
The existence of ten language packs proves full professional localisation: not established, and \[IMP14\] recorded missing visible messages within the specified test; text completeness alone also does not prove the correctness of agricultural terminology or local practical usability.

Conversely, negatively framed claims under quarantine must not be made either, because the evidence refutes them:

‘The cloud path was never tested live at all’: this contradicts evidence of local live verification for the conversation, retrieval, and citation paths recorded in the package.
‘The phone is merely a project or a release candidate’: this contradicts a server gateway implemented with contract, routes, signature, and sessions.
‘The desktop was not built’: this contradicts successful deployment, packaging, and test signing dated 21 August 2026\.
‘AgroGenie is an isolated tree’: this contradicts its linkage to the front end and to the conversation path.

Quarantine works in both directions: a claim that exceeds its evidence is quarantined, and a denial that exceeds its evidence is likewise quarantined.

Reopening decision

A claim is not reopened on the basis of an AI summary, a secondary excerpt, or a vendor page alone when the claim concerns impact, safety, or transferability. Such materials may lead to the source, but they do not replace it. The reviewer fills in:

Claim identifier: Exact wording: Primary source: Location of the finding: Context and setting: Does it match the stated system? The material, place, season, user, language, unit, and service covered by the evidence: Contexts it does not cover and to which it must not be transferred: Is there a conflict? Permitted wording: Prohibited wording: References and date:

The reviewer then updates the claims register, statistics, status, chapter, and references in a single change, so that the value does not become detached from its limits. If the new source is a later artefact, a dated layer is added and does not erase the historical judgment on an earlier version.

Appendix K: Cross-chapter synthesis

1\. From data to responsibility

The different applications reveal a single pattern. A decision begins with an observation that has an owner, a unit, and a time; it is then transformed into a feature, a model, an output, and an action. Every transition can alter the meaning. When harm occurs, responsibility must not be lost between those who collected it, those who built it, those who operated it, and those who approved it.

For this reason, provenance is a continuous thread from the file or sensor to the answer. If the worker modifies the recommendation, the reason is recorded; if the model changes, it becomes known which decisions were affected; if the laboratory corrects the diagnosis, the related cases can be updated. It is this structure that makes ‘learning’ institutional rather than mere retraining.

2\. The general decision loop

The following loop structure works for irrigation, disease, livestock, and the supply chain, but its values, thresholds, and authorities do not transfer across contexts:

Define a case or question for a specified commodity, place, season, and decision window.
Collect the minimum sufficient data.
Check quality and provenance.
Generate a probability estimate or alternatives.
Apply local rules, authority, safety, and a known fallback.
Make a human decision, or a constrained automated one.
Implement and monitor.
Measure the outcome and compare.
Learn and update, or stop.

The loop may run in seconds for ventilation or over a season in a crop cycle, but its logic remains constant. Each time, the language, unit, device, connection, and the authorised party empowered to exit the loop or halt it must be fixed. If the outcome cannot be measured, the claim must be limited to a monitoring or co-ordination function.

3\. Four types of failure

Epistemic failure: the model does not know, or the source is out of date. Remedy: abstention, sourcing, and updating. Technical failure: device, connection, or software. Remedy: detection, redundancy, fallback, and recovery. Operational failure: the output arrived but was not acted on, or arrived too late. Remedy: workflow design, responsibility, and support. Institutional failure: conflicting incentives, the absence of a right, or dependence on a single provider. Remedy: governance, contract, plurality, and exit.

Failure may also be a localisation failure that cuts across these types: a local name was matched to the wrong variety, a unit was misinterpreted, a warning lost force in translation, or a service does not reach the region within the decision window. Its remedy is not retraining alone, but correcting meaning, pathway, authority, and local support.

Many projects address the first and leave the rest, then attribute the failure to ‘lack of farmer adoption’.

4\. The difference between accuracy and usefulness

Accuracy is a model metric, whereas usefulness is a change in a user outcome after all loops. It can be represented roughly as follows:

Decision usefulness \= prediction quality × timeliness of delivery × actionability × compliance × outcome value − system cost and risk

This is not a final calculation; it shows that any factor close to zero reduces usefulness. So the model should not be improved if the bottleneck is maintenance, an official source, or ability to pay.

5\. The shared structure of the named cases

The cases in the book fall into six evidentiary categories:

Preliminary studies such as Yield-XAI, the maize CNN, CropGPT, and the weed ConvNet;
Reviews such as those on aquaculture and spectral analysis;
Simulations and computational experiments such as supply-chain agents and cooling;
Drafts such as PoultryFI;
Products and programmes such as FarmerChat, GAIA, Carbon Robotics, and Harvest List;
A repository and an engineering report such as the Aladdin components.

These ranks are not an absolute ladder of quality; each establishes something different. The error is to let one substitute for another.

6\. The shared regional pattern

Priorities vary across regions, and a region cannot be reduced to a map or a country name. Even so, three recurring layers appear:

General foundation: weather, soil, and market data; connectivity; standards; extension.
Sectoral service: irrigation, disease, logistics, or finance.
Governance: authority, rights, security, and evaluation.

Investing in the second layer without the first produces fragmented solutions, and without the third produces risk and lock-in. The general foundation must also not become a tool of surveillance or exclusion.

7\. Water, energy, and data

A three-way trade-off recurs. Measuring water requires devices, connectivity, and energy; control may save one resource while increasing another; and data themselves require infrastructure. Decisions should therefore be made within clear system boundaries. In the Gulf, cooling may be decisive; in an area with intermittent connectivity, local computing; and on a smallholding, a shared service.

8\. Labour and authority

Automation does not remove people; it redistributes visibility, action, and responsibility. If the decision is hidden inside a system, the worker may bear the consequences of a result they cannot understand. Therefore, the designated human must have:

sufficient knowledge;
time to intervene;
real authority;
access to the source;
a fallback if they refuse;
protection from punishment for reasonable objection.

Institutions also need to build the new skills and fund that effort.

9\. Exitability as a test of maturity

A system that ties devices, data, and history to a single supplier may be easy to start with and hard to leave. Testing export, migration, and exit forces the design towards standards and provenance, and exposes dependency before a crisis. It is also a test of sincerity: is ‘your data is yours’ something that can actually be practised?

10\. Local knowledge as a living asset

Algorithms learn from repeated data, whereas local knowledge may carry rare cases and long memory. It should be treated neither as unchallengeable truth nor as noise. It should be documented with context, source, and consent, tested, and kept open to plurality. A difference in name may indicate a difference in variety, stage, or practice.

11\. Why do big numbers look attractive?

A single percentage is easy to market; a chain is hard to explain. But a number may migrate from a study to a product, from a metric to an outcome, and from a laboratory to a continent. The book’s record of scrutiny is not hostile to measurement; it asks for denominator, context, and source. A small, documented value is more useful than a large percentage with no traceable path.

12\. A practical definition of a responsible system

A responsible agricultural system is one that:

addresses a known decision;
uses lawful and relevant data;
preserves the local name, language, unit, meaning, and authority;
is compared against a baseline;
presents uncertainty;
abstains and escalates;
fails safely;
has service, maintenance, and a workable alternative within the decision window;
preserves rights of transfer and exit;
measures outcomes, disparity, and resources;
learns or stops according to the evidence.
states where it has not been tested, and does not carry a result beyond its context without verification.

By this definition, responsibility is not a constraint on innovation, but the standard that distinguishes a localised tool that is repairable, open to challenge, and capable of exit from a temporary showcase.

Appendix L: Brief for producing figures and tables

Purpose

Images are not added for decoration. Every figure must explain a relationship that is difficult to convey in a paragraph, and must have a title, a data source, and alt text. The proposed graphics below can be produced after the main text is approved, and do not give the designer the right to add any number, causal relationship, or operating condition not established by the text. When a map, unit, calendar, or entity name is used, the naming, boundaries, orientation, and appropriate unit should match the publication context, while keeping the original reference visible.

Figure 1 — The compound pressure network

Type: causal map. Nodes: climate, water, soil, disease, labour, energy, price, supply chain, access. Message: pressure moves between resources, decision, and outcome; AI is a layer of observation and co-ordination, not an independent cause of the solution. Warning: do not place percentages or magnitudes without a source.

Figure 2 — The decision chain

Type: flowchart. Stages: observation → quality → context → model → uncertainty → authority → action → outcome → learning. Failure points: mark every transition. Use: Chapters 1–5 and 30.

Figure 3 — The multi-evidence pyramid

Type: a matrix, not a linear pyramid. Horizontal axis: question type (legal, performance, existence of a feature, impact). Vertical axis: official source, primary, review, operational, supplier, scenario. Message: the highest-ranking source varies by question.

Figure 4 — Observation provenance

Type: data lineage. Content: file/sensor → raw → normalised → validated → feature → model → answer, with identifier and version. Alt text: explains the ability to trace back from the decision to the measurement.

Figure 5 — Spatial monitoring and verification

Type: spatial loop. Content: satellite → change → inspection points → interpretation → action → follow-up image. Message: an indicator is not a diagnosis.

Figure 6 — Governed agricultural RAG

Type: flow with gates. Content: question and context, risk classification, source index, country/date/authority filter, generation, checking, answer/refusal. Warning: do not present a model name as a source.

Figure 7 — Plant disease decision

Type: three layers. Content: visual inspection, diagnosis/confirmation, IPM/official intervention. Message: the machine does not move across the layers without data and authority.

Figure 8 — Robotics chain

Type: timeline. Content: detection, localisation, planning, action, verification. Metrics: each stage shows its own metric beneath it, and the final outcome is shown separately.

Figure 9 — The water, energy, and yield triangle

Type: trade-off triangle. Content: do not improve one corner without monitoring the other two; total consumption is added around the triangle. Use: irrigation, greenhouses, and sustainability.

Figure 10 — Labour transition

Type: before-and-after diagram. Content: manual task → monitoring/intervention/maintenance; with skill, training, and risks. Message: task transformation, not simply the disappearance of the job.

Figure 11 — Cascading failure

Type: network. Content: identity/cloud/power/cooling/inventory/market, and propagation arrows. Teaching point: central optimisation may increase shared dependency.

Figure 12 — Data rights

Type: contract cycle. Content: collection, use, sharing, retention, export, transformation, deletion, exit. Message: ownership is a broad term; rights are operations.

Figure 13 — Maturity ladder

Type: levels. Levels: idea, experiment, pilot, candidate, production, impact. For each level: the evidence required and what it does not prove.

Figure 14 — Aladdin general architecture and evidence status

Type: architectural flow topped by two dated evidence layers, not a single unified operational-state diagram. Content: consumer → governed service → Orchestrator → policy/provider registry → adapter → gateway/managed identity → approved target. 22 August layer: shows what the historical source trail \[IMP01\]–\[IMP09\] established, including limited local live verification of specific paths and the still-open release decision. 22 September layer: shows what the official distribution package \[IMP14\] established about the integrity of the distribution artefact, the presence of contracts and components, and the results of the specified checks, while also showing the constraints on integrity, language, the interactive book, and testing; it does not inherit operational results that were not rerun on it. Reading rule: the arrow means ‘this is the declared architectural path’, not ‘this works now for every user’. The layer colour or the completeness of the arrow does not mean feature-flag activation, provider connection, interface acceptance, recommendation validity, or localisation completeness. Warning: do not display keys, addresses, internal resource names, or threat details.

Figure 15 — Research triangulation

Type: three intersecting axes. Axes: source (official/primary/field), method (quantitative/operational/qualitative), time (before/during/after/shock). Centre: a defensible claim, not ‘absolute truth’.

Drawing specification

Arabic runs RTL, and the titles are concise.
Crop, animal, and local place names are retained, and variant forms are linked to the standard term rather than erased.
Boundaries, administrative names, the calendar, units, and axis orientation are reviewed with a competent local authority.
Use colours distinguishable to people with colour-vision deficiency.
Do not use green alone to indicate success.
The figure remains readable when printed in black and white.
Every abbreviation is defined in the title or key.
Every number is accompanied by its source, period, and unit.
The alt text explains the conclusion, not only the appearance.
The alt text has a localised version that preserves the warning, the negation, and the limits of transferability, rather than a shortened literal translation.
Both an editable source version and a publication version are retained.

Acceptance gate for the source visual

It answers a single question.
It matches the main text and introduces no undocumented claim.
It has a source, a licence, and a fingerprint.
It has passed scientific, linguistic, and accessibility review.
It has passed localisation review for names, units, boundaries, orientation, and the intended conclusion.
It does not use a disproportionate visual comparison.
It does not turn one case into an average or a forecast into a fact.

Appendix M: Field decision scenarios

These scenarios are synthetic; they are not reports on real farms or ready-made implementation plans. Their purpose is to show how the book’s principles are translated into decisions, and they do not include dosages or site-specific recommendations. A scenario must not be applied merely because its description appears similar; it is rebuilt after local inspection and the approval of the competent authority.

Scenario localisation card — a precondition for use

Before any case below is turned into a trial or an action, the team documents:

the agricultural material in question, its stage, and its local name;
the location, production system, season, and decision window;
the user, their role, authority, language, and unit;
the device, power, connectivity, and how work continues during an outage;
the institution, the professional or regulatory reference, and the approval and escalation path;
the technician, spare parts, service, and the alternative actually available;
the outcome to be measured, the baseline, the consequence of error or delay, and who bears it.

If any of these fields remains consequential and unknown, the scenario remains a discussion tool, not operating instructions. The software can organise the fields, surface conflicts, and compare alternatives, but it cannot invent an authority, threshold, service, or alternative that does not exist.

Scenario 1 — A map says 'thirst' and a valve says otherwise

A farm engineer receives a heat map indicating stress at the edge of the field. If the map were turned directly into an irrigation command, water might be increased in a sector that does not receive it evenly. The correct path begins by comparing the image date with the weather, then checking the flow meter and pressure. The readings show that the valve was given an open command, but the flow is lower than usual.

A worker checks three points and finds a partial blockage. He repairs the line, measures distribution, and then captures a follow-up image. The satellite did not diagnose the blockage; it narrowed the search area. The system’s value is measured in detection time, water, and damage avoided, separately from the effect of repairing the network. If there had been no meter or worker, the map would have been an incomplete signal.

Scenario 2 — A disease image and high confidence

A farmer uploads an image of a leaf to an application, and the model assigns it to a class with high confidence. Before suggesting a treatment, the system asks about the crop, stage, location, and spread of the symptom, and requests images of the plant and the field. A second image conflicts with that class, so the system shifts to 'undetermined' and suggests keeping a sample and contacting the extension officer.

The extension officer finds that the symptom is linked to salinity in part of the field, not to the causal agent classified by the model. The case is logged as an error, and the images are reviewed. If the application had recommended a pesticide, the cost, and perhaps the harm, would have increased. Numerical confidence was not authority; the quality of the abstention path prevented the error.

Scenario 3 — A weeding robot in an imperfect field

The robot succeeded in a demonstration on regular rows and small weeds. On the client’s farm there is residue, overlapping crop growth, and larger weeds. Before purchase, a trial is run on strips that represent the density and occlusion. The team measures detection, action, damage, time, and intervention, not a single vision metric.

The result shows that performance is good on small weeds between rows and that a worker is needed near the crop. Instead of rejecting the system or claiming full autonomy, a hybrid service is designed and the timing of the pass is changed. The cost is calculated including transport, cleaning, and maintenance. The farmer keeps the work map and the right to export it.

Scenario 4 — A greenhouse during an outage

The controller predicts a rise in temperature and begins opening the vents, then the cloud provider goes down. A local unit continues to monitor critical limits and run the fans, and an independent alert is sent. The interface shows that advanced optimisation is stopped and that safe mode is operating. The operator can intervene from a local control panel.

After the incident, it becomes clear that the identity service itself was required to log in to a remote panel. The design is modified so that emergency authorisation does not depend on that same endpoint, and a recovery drill is run. This shows that resilience is not just the presence of a generator; it is the separation of dependencies and the availability of human authority.

Scenario 5 — An alert in a poultry house

A multi-sensor system detects reduced activity and an unusual sound in one area. Instead of diagnosing a disease, it displays the temperature trend, air quality, and sensor status, and directs the worker to inspect. The worker finds that a ventilation opening is stuck and that the birds are showing heat stress. The worker intervenes according to the house protocol and calls the veterinarian if the signs persist.

The system records that the alert preceded the problem, but it does not attribute any improvement in production or welfare before measurement. The analysis also shows that a second sensor was necessary to prevent silent failure. That failure mode is added to testing.

Scenario 6 — Yield forecast and storage decision

The tool presents a yield range, not a single figure, six weeks before harvest. The co-operative compares scenarios against storage capacity and transport contracts. It does not build a new facility; it books flexible capacity because uncertainty is wide. After harvest, it compares the error and finds that a new field outside the training data caused bias.

The tool then narrows the range for this type of field, and the team collects data from another season. The benefit lay in flexible planning, not in claiming exact knowledge of yield. They also calculate the value of the decision against the cost of the service.

Scenario 7 — Market platform and loan

A platform proposes a buyer, a price, and an input loan. The user discovers that the ranking of buyers is influenced by a commercial partnership, and that accepting the loan allows the production record to be used for other assessments. In a governed system, the disclosure appears before the decision, offers not tied to financing can be sorted, secondary use can be refused, and a human review of the credit assessment can be requested.

The outcome is measured by net price, payment time, and loan terms, not by clicks. The farmer can also export transactions and reputation when leaving. This turns the digital market from a closed gateway into a comparable service.

Scenario 8 — Exit from a vendor

A farm decides to switch platforms after a price increase. It requests an export and receives a PDF containing neither the raw sensor values nor the column definitions. The exit test reveals that ‘data ownership’ was not portability. Before the new contract, it asks for documented CSV/JSON, stable identifiers, a conversion log, and a transition-support period.

It tests importing a sample into an independent repository before moving everything. The devices retain basic functionality during the transition, and the keys are revoked after verification. The cost of exit is included in TCO. In this way, the right becomes a tested procedure.

Scenario 9 — Shock in the cold chain

A main road closes, so the optimiser proposes a longer alternative route. It does not choose on journey time alone; it calculates cooling capacity, fuel, the driver’s time window, and product condition. It allocates the batches according to their sensitivity and keeps a decision log. At the same time, a tracking provider fails, so local identifiers and deferred synchronisation continue to operate.

After arrival, temperature and quality are measured, not time alone. It turns out that fuel stock was a greater point of weakness than the algorithm. The result enters the risk map, and an alternative supplier is added.

Scenario 10 — Local knowledge conflicts with the model

A crop-suitability model proposes a crop, but farmers reject it because seasonal winds in one part of the valley cause a problem that does not appear in coarse climate data. Instead of classifying the rejection as resistance, the team records the description and locations, installs sensors, and compares. The pattern is confirmed at some sites and does not appear at others.

A local layer is added with source and confidence, and the term used by the community is preserved. The platform does not appropriate the knowledge; participants agree to its use, and attribution is shown. The map improves because it accepted a difference that did not match the initial lexicon.

Scenario 11 — A translator changes a warning

A safety paragraph is carried into a new language, and the phrase ‘do not use before verification’ is shortened into a less forceful form. A parallel test finds that users understood it as a suggestion. The feature is suspended in that language, the catalogue is corrected, and a case is added to the regression suite.

The existence of a translation file was not enough; verification covered meaning and action. The version is also recorded so that the old wording does not return from memory or from an application that has not been updated.

Scenario 12 — When is not investing a success?

A co-operative wants to buy a drone and a platform. The baseline shows that the greatest loss is caused by delayed transport and the absence of suitable crates, and that data will not change a decision before the chain is repaired. The co-operative funds improvements to packing and collection scheduling, and postpones sensing for one season.

Loss falls according to local measurement, and it then reassesses the need. This is not a rejection of technology; it is choosing the weakest link first. The digital project remains a candidate if a spatial problem appears that the current record does not solve.

Summary of the scenarios

These composite cases show that value arises at the boundaries: between an image and an inspection, between a forecast and a decision, between an instruction and an action, between a company and a user, and between connection and interruption. The best design is not the one that hides these boundaries, but the one that records them and places verification, authority, and an alternative at those points. Every outcome remains constrained by its material, place, season, user, language, unit, institution, and service; if any one of these changes, the verification question changes, and the scenario does not become a portable recipe.

Appendix N: Implementation Evidence Register — IMP01–IMP14

Why a register separate from the references?

The first nine items of evidence in earlier drafts carried SRC identifiers and were counted among the book’s references. That was a methodological confusion, and it has been corrected in this edition. Five later internal artefacts were then added directly to this register; they did not enter the reference list at all, and did not change its total.

Under the conventions of scholarly publishing, a reference is a document that the reader can access and verify independently. A file path inside a private repository does not meet that standard: the reader has no access permission, no one guarantees that the path will remain, and no third party can re-examine it. Including something of that kind in the reference list grants it an authority it does not possess and inflates the number of references with material that cannot be verified.

Internationally recognised software-citation standards — most notably the FORCE11 principles published in PeerJ Computer Science — require four things in any software citation: a unique identifier that is machine-resolvable, persistence beyond the life of the software itself, access, and version specificity.

The frozen artefact listed below meets only the fourth: the version is specified by a declared version number and SHA-256 hash, and this is a genuine documentation improvement that makes examination reproducible for anyone who has the package. The other three are not met: there is no machine-resolvable identifier, no guarantee of persistence, and no access. The later artefacts are specified by their hashes and dates to varying degrees, but they share the lack of access and of a persistent identifier. Three out of four suffice for exclusion, because access is a non-substitutable condition: what the reader cannot open, the reader cannot critique, and precision in describing the version does not turn it into a reference.

Therefore these artefacts are not placed in the reference list: the older evidence was removed from it, and the later artefacts were recorded from the outset with IMP identifiers (short for Implementation Evidence: implementation artefact). What follows from this is binding:

They are not counted among the book’s references. The number of references is two hundred and eighty, and all of them are external and accessible.
They do not support a claim of agricultural performance, impact, or general quality. Their maximum scope is to demonstrate a structure or an internal implementation result specified by artefact, environment, and date.
They may not be cited outside this book as scientific sources.

An independent localisation limit is also added: proving the existence of a component, contract, or language test does not prove that the system is suitable for any particular crop, animal, region, season, user, language, unit, device, or institution. That suitability requires testing of understanding, action, service, and outcome in the intended context, and cannot be inferred from the code or package.

Accessibility statement

A reader of this book cannot access any of the fourteen evidentiary items below. This is an explicit declaration, not an apology: the artefacts were private when the evidence record was closed, and they have no published public persistent identifier and no public archived copy. Specifying their names, fingerprints, and dates makes examination reproducible for whoever possesses them; it does not make them available to the reader or independent of the sponsoring party.

Anyone wishing to verify has only one route: to request an independent examination from the sponsoring party through the published contact details. A refusal by the party to provide access is not evidence that the claim is true, nor that it is false; the claim remains externally unverified, and that is its precise description.

Identity of the frozen artefact — IMP01–IMP09

All entries relate to: Aladdin Agri AI (software platform, private repository). Author/developer: Dr Alaa al-Din Ali.

Field	Value
Artefact	aladdin-pertanian-agri-ai-Koot-EL-Kuloob-2.1.39-desktop-p0-p1-source-2026-08-22.zip
Artefact date	22 August 2026
SHA-256	7B0CE4E1699E72958272821EA2A9083E2DC5EAB39069D395325B4AA6EE2BEC29
Plugin header	Koot-EL-Kuloob-2.1.39
Database schema	4.2.4
Minimum PHP	8.1
Package type	Source — contains no executable file or installation package for the mobile and desktop applications
Compliance review date	4 September 2026
Tag	Internally tagged with version and fingerprint
Public persistent identifier	None — no DOI and no public archived copy
Access status	Private — no reader access

Declared documentary correction: in an earlier edition, this paragraph described the artefact as ‘untagged’, and that description was inaccurate: the artefact carries a version and an identifying fingerprint. The correct description is that it is internally tagged and unpublished. This correction does not raise the evidential weight — the following point remains unchanged: the reader cannot access it, and it is not fit to serve as a reference.

Register

Identifier	Examined component	What it proves only	What it does not prove
IMP01	AiBridgeOrchestrator.php	The presence in the code of an orchestration component that evaluates policy, routes, executes, and logs	That it runs in every installation, or produces an agricultural effect
IMP02	ProviderGovernance.php	The presence in the code of a routing policy, provider registry, and circuit breaker	That the policy is applied at runtime for every request
IMP03	GovernedExecutionService.php	The presence of a governed execution interface that does not itself resolve the provider	That all consumption actually passes through it, with no backdoor path
IMP04	class-livingcardprovider.php	The definition in the code registry of seventeen cards identified by full names	That any of them are enabled or operate for every user
IMP05	FrontOffice/Localization/i18n \\+ class-frontofficegate.php	The presence of ten language catalogues and three character names	That it underwent human linguistic review, or that paths are operationally equivalent across languages
IMP06	includes/integrations/agrogenie	The presence of an integration connected to the front end and the conversation path, with staged modes and two gates closed by default	Its activation, service availability, or completion of acceptance for this package (see the completion constraint below)
IMP07	Mobile gateway contracts and routes within the package	The presence of an implemented server gateway: contract 1.14.1, 57 unique routes (59 method/path registrations), request signing, session derivation, pairing, heartbeat, release policy, and a declaration of non-support for bidirectional synchronisation	The presence of an installed client, store availability, readiness of every backend dependency, or a live user experience
IMP08	Desktop Application tree \\+ reports of 21 and 22 August 2026	Successful local test build, packaging, and signing on 21 August (release 1.1.0), followed by P0/P1 governance changes on 22 August that passed static tests	Closure of all runtime gates after the P0/P1 changes, or production signing, or store distribution, or that the twenty-two-item catalogue is a set of ready workspaces
IMP09	Chat V2 readiness log \\+ retrieval and citation log \\+ sovereignty tests	The presence in the code of contracts and tests, and local live verification of conversation, streaming, retrieval, and citation paths, with the release decision INCOMPLETE / NO-GO recorded and both the release-safety gate and its package gate open	The quality of any agricultural answer, or release acceptance, or availability of the external chain at the time of reading

Updates to implementation evidence after the frozen artefact — IMP10–IMP14

IMP10–IMP12 were added to the review register on 16 September 2026, then IMP13 was added on 18 September, and IMP14 was added after examination of the official distribution package on 10 October 2026\\. The date of addition does not replace the date of the original material, nor does it merge later artefacts into the 22 August artefact. The purpose is to preserve sequence: what the first examination knew remains as it was, and what later appeared for review is placed in a separate layer. In particular, IMP13 does not rewrite the status of IMP06 or erase its constraint, and IMP14 does not rewrite IMP01–IMP09 merely because the two packages bear the same version number.

Identifier	Artefact and version identity	Verified fingerprint	What it proves only	What it does not prove
IMP10	Project memory archive; 468,089 bytes; 103 entries, of which 102 are Markdown files; examined on 16 September 2026	SHA-256: 35E26CCD967B5B2B19900DE101D3618D4CD08A98A42B1BEFB012C96246F7D363	The presence of an internal historical record of faults, corrective actions, and failure patterns from which the Chapter 33 accounts were built	That every fault still persists, or that every repair is complete in the current source, or any performance or agricultural effect
IMP11	Final engineering audit report and its evidence statement for the Habiba Mobile 3.2.3 (12214) build, dated 19 August 2026; entered into review on 16 September	Report: 23DFFAF2CBE18BD7A68C30C1193CD2FDAF752C0BB5E6129C988D381852A04455; evidence statement: 409F49AFA5C6E178144363C326FF2405D3F3F0D7415E92040D1D4F7AC83FAD5B	The specific local audit findings: artefact integrity, static tests, Android execution, and local pairing, alongside blocked services and an installation fault; and its textual decision RELEASE\\_BLOCKED	Connection to production, or store acceptance, or readiness of every backend service, or generalisation of the result to another build, or an agricultural effect
IMP12	Humanizer v2.0 plan dated 8 June 2026 and the supporting source package dated 15 June; entered into review on 16 September	Plan: C6C1EA8434138556070CB39D9972C8F73FFA55A83BA7CD11C506D84902485DE5; package: F7DA1A703D690F8249DE705488C32EEC9671D75B4A0E8641FC30F08E219EDE81	The presence of design, defect review, and proposed tests for the risks of multilingual stylistic transformation; and the plan states that the integration code had not yet been written when it was prepared	Implementation of Humanizer within the platform, or its safety for every language and text, or its passage through a production path, or its improvement of an agricultural outcome
IMP13	Aladdin AgroGenie v4.2-developed package; 100 files; 1,423,275 bytes; examined and its tests run on 18 September 2026 using PHP 8.5.6	SHA-256 of the ordered file manifest: 7BC521E18884BD15CF534D68842B82927AED867BE7BEB3FC644885F4F5E8D460	The presence of a governed agricultural research structure, support for ten languages in the tests, and linkage to approved sources, AiBridge, and Chat V2; the success of 104 prototype queries, 149 isolated runtime checks, 196 phase-nine gates, and 33 boundary-guard checks with no failures; and the non-recurrence in this package of the historical test stoppage	Live operation inside WordPress or a real database, or a live connection to an AiBridge provider, or browser or 100k/1m load testing, or activation in every installation, or field accuracy or an agricultural effect
IMP14	Official distribution package `Koot-EL-Kuloob-2.1.39.zip`; generated on 22 September 2026; 65,023,351 bytes; 2,402 entries; examined on 10 October 2026	SHA-256 of the package: A86FD7A56DE5FDFF7CF13C60727A83A872FBBE861044DB43592A9135A84E38DA; and of the release manifest: 46E91AF93C910029F99DC791A0CF2979F0441C71356586A5EE31824178FF4E8D	The correspondence of the four release surfaces, the integrity of the ZIP paths, the uniqueness of its entries, and the presence of the required runtime files; the presence of AiBridge and Chat V2 contracts, the interface, agricultural modules, translation, and the optional interactive book; and the success of selected static and isolated tests; it also establishes the specific negative findings set out below	Live WordPress execution, or the status of flags and the database, or any external-provider connection, or interface and deployment acceptance, or completeness of translation and localisation, or the correctness of an agricultural recommendation, or a field effect

The IMP13 manifest fingerprint was built by ordering the relative paths, calculating SHA-256 for each file, then calculating SHA-256 for the aggregated text in HASH␠␠PATH format with LF line separators and no trailing line. For selective audit, the fingerprints of the core files are: README.md \\= 7E1709462278087E58DF1E7C6D56E53F1294D7BB81A637A224D487FEA8455109; root audit report \\= 1C74F68310A3AA63B88730E83EDAAD36433EE667CBD95800087E14A95C4E1B11; prototype test \\= 2144D7F735199ADE3B178C965F954C1BD5A197D28AD38DB1586B0DF198D0CB51; runtime test \\= 831DFCBECD0AFB74E3D2D9614EFBD04D6F8C634663F7650DF2D028090462C4B3; phase-nine gates \\= D928EC23E58E677E2108F27A7DA1A011EE52D38D9E7C6FA2CAF9317393AC0A3B; and boundary guard \\= 826AF1ED8E0E1C3B928859EBB89C5D0350CFA40F8E609AB95564F5E656025931.

This update does not create a new desktop entry; its status document dated 21 August is already included in IMP08. Nor does any later entry repeat old evidence under another name merely because the lessons file reused it. IMP13 was created because the current AgroGenie package, the identity of its files, and the running of its tests represent a new time-stamped artefact, not a renaming of IMP06. IMP14 was created because the official distribution package generated on 22 September is an independent artefact from the historical source package dated 22 August, even though both carry the same version number.

Recorded constraints in the official distribution package — IMP14

Constraint	Verified result	Limit of inference
Internal integrity chain	The external ZIP fingerprint and the release manifest matched, but four critical file samples carry `CRLF` while the sealed fingerprints correspond to `LF` versions. After line-ending normalisation, the samples matched, while the integrity monitor uses `hash_file` on the raw bytes	The integrity gate may classify the official package as modified and close the path; this is a packaging-and-sealing constraint, not proof that the entire plugin fails
Bundled AiBridge set	It ran 137 test files and 2,349 assertions, and recorded 77 failures. Some failures relate to missing encryption extensions, to test files that expect tools and reports excluded from the runtime package, or to a dummy host; the record also includes the fingerprint conflict above	The failures are not aggregated into a quality percentage, not all attributed to the code, and the integrity chain is not described as acceptable without reservation
Visible language	Isolated tests succeeded in separating language, schema, and channels, but the `visible-ui-language-remediation-regression.php` test did not find all seventy-six messages in Arabic, Turkish, Indonesian, and Simplified Chinese	The existence of ten language packs does not prove completeness of visible text, the correctness of agricultural localisation, or directionality and operational parity
Interactive book	Ten contract and isolated tests out of thirteen succeeded, and the editorial, plan, and release gates failed because the English-source fingerprint changed and because Chapter Four does not match the expected sequence	The module and contracts exist and fail closed, but the manuscript does not become published, accepted, or correct merely because they exist

The extracted package was examined as-is, not the mutable working tree. Some pivotal bootstrap files in the working tree matched the package by fingerprint, but that does not prove that the rest of the tree matches or turn it into a new distribution artefact. WordPress was not loaded, its actual database and flags were not examined, the interface was not acceptance-tested in a browser or on users’ devices, and no independent agricultural or linguistic review was carried out.

Completion constraints recorded in the frozen artefact — IMP01–IMP09

They are recorded here because they pertain to that package itself, and they must neither be generalised to the platform nor removed:

Constraint	Description	Effect on the claim
Two AgroGenie cache files in the historical package	The attached search test invokes the files AladdinAgroGenieCacheService and AladdinAgroGenieEvidenceFingerprintCache under engine/src/Cache/, and that folder is absent from the 22 August package; the test stopped before executing its checks	This prevents AgroGenie in IMP06 from being described as operationally verified; it does not invalidate the integration described in the code, and the constraint does not carry over to IMP13, where this stoppage did not recur
Isolated test installer	The P0/P1 report recorded a mismatch in the fingerprint and size of one test installer, so it was excluded from delivery	A release-integrity constraint resolved by isolation; it is noted and is not to be read as a product failure
Environmental failure during inspection	Some tests stopped in the inspection environment because PHP extensions were missing (sodium, openssl, mbstring)	An environment failure, not a package defect; it must not be recorded as a defect, and it must not be omitted because it explains the absence of an operational result

What is required to elevate this evidence into references

Elevation is possible, and its conditions are known. For any entry in this register to move into the reference list, it must satisfy all four principles in full:

Condition	Current status	Required action
Unique identifier	Not met for any entry	Deposit a tagged release in a public research repository that assigns a DOI (such as Zenodo, a CERN service)
Persistence	Not met	The identifier must remain resolvable even if the repository changes or disappears
Accessibility	Not met	A public landing page that displays the metadata and a link to the artefact
Version identification	Met for the frozen artefact, and partially met for later artefacts through releases, dates, and fingerprints	Add a CITATION.cff file and a standardised artefact statement for each entry

The frozen artefact therefore satisfies the version-identification condition, and the later artefacts have matchable fingerprints; but no entry satisfies accessibility, a persistent identifier, and persistence together. Therefore none of them has been elevated to a reference.

Until these conditions are met, none of the IMP artefacts becomes an external reference. Part Nine now cites accessible external sources for the principles by which the case is judged, but it still has no independent Aladdin-specific evidence of operational performance, completed localisation, affordability in practice, user benefit, or agricultural impact. This remains a material limitation and is stated without allowing general standards to stand in for evaluation of the platform itself.

How this appendix affects the reading of the book

Anyone reading Chapters Thirty, Thirty-One, Thirty-Two, and Thirty-Three must understand that they are reading an engineering description with its source declared, not a scientific result or a localisation certificate. The chapters describe architecture and implementation experience in materials not examined by an independent party, and they claim no more than that.

Anyone seeking a judgement on the agricultural value or local suitability of the sponsoring entity's platform will not find it in this book, because the evidence required for that — a localised field trial, a study of use, comprehension, and service, and an independent evaluation — is absent. Its absence is declared, not disguised.

Appendix C: Reference guide to software and systems implementing artificial intelligence in agriculture up to 31 August 2026

A dated register of commercial products, deployed platforms, operational institutional and government systems, development programmes, and operable open-source tools

Introduction: from the laboratory to the field

Artificial intelligence did not enter agriculture through a single gate. It entered through imaging, when a phone became able to recognise a spot on a leaf; through the sensor, when the soil began sending its moisture reading every hour; through the satellite, when the field could be read from hundreds of kilometres above it; and through language, when the farmer could ask in their own dialect rather than in textbook terminology. At each gate, tools emerged: some became products that could be bought and operated, some remained research papers, and some closed after a funding round.

That multiplicity is what makes the question, ‘What software is available?’, harder than it appears. Lists circulating on the internet conflate the name with the product, the marketing page with the working service, and the research paper with the operational system. This guide is an attempt to build a disciplined alternative around one question: what software and systems were actually implemented and whose existence could be verified by the cut-off date, who stands behind them, what agricultural task they perform, and what the evidence for them is.

Establishing existence is the function of this guide, not establishing local suitability. Software that exists and has been deployed may not be available in a given country, and may not support the required language, terminology, unit, device, connectivity, payment method, institution, maintenance arrangement, or alternative. The reading must therefore distinguish between the existence of the software, its availability status, the evidence for its evaluation, and its suitability for a specific local context. That last field must not be inferred from the developer's nationality, the number of languages, or the breadth of adoption.

1\. Unit of inclusion: the implemented software, not the publication that describes it

This is the governing rule for this entire appendix, and it is what distinguishes it from everything that came before it in the book:

A scientific paper may establish that a system was evaluated, but the paper is not the system. A review may map a field, but the review is not software. A simulation may test an idea, but a simulation is not an operational system. An architectural framework may organise thinking, but it is not software until an identifiable implementation exists.

In the previous edition of this book, the appendix mixed forty-seven heterogeneous items: commercial products alongside peer-reviewed studies, platforms alongside field reviews, operational systems alongside architectural concepts, simulations, and names whose identities had not been resolved. That has been corrected in this edition. The previous figure was not a count of software programs; it was a count of mixed cards, and it has been abandoned as a metric.

The scholarly material was not deleted — it was moved to its correct place: studies to their chapters and to the controlled reference list, reviews to ‘field-mapping sources’, simulations to their proper description as simulations, and unresolved names to the leaders register. The difference is that they are no longer counted as software.

2\. Scope boundaries — to be read before any entry

Scope boundaries: this guide aims to present a broad, verified, dated register of software and systems implementing artificial intelligence in agriculture whose public existence could be verified up to 31 August 2026\. It is not a complete global inventory and does not claim to be one. The absence of a given name does not mean that it did not exist, and the inclusion of any software does not imply endorsement or proof of its effectiveness, safety, production readiness, or suitability for a particular agricultural context.

Three caveats must be stated:

First, the guide is broad and verified, not exhaustive. Every entry in the main register has an official source or a public primary source; that makes it open to examination, but it does not make the register inclusive of everything that existed worldwide.

Second, the density of an application family does not measure its global activity. The number of entries in a domain reflects this guide's coverage, not market size or the number of companies operating in it.

Third, an empty family is not a failure. When no entry is listed in a domain, the meaning is that this edition did not establish a software program there that met the conditions — not that the domain is empty. Filling a slot with a study or a concept to avoid emptiness is a greater error than the emptiness itself.

3\. The seven inclusion conditions

No entry is admitted to the main register unless all seven conditions are met, not merely some of them:

\#	Condition	What it means in practice
1	An identifiable proper name	‘Smart irrigation with the Internet of Things’ is not a name; Climate FieldView is a name
2	A specified responsible entity	A company, government, university, institute, organisation, or maintainer of an open repository
3	Actual implementation	A system was built, released, published, operated, distributed, or made publicly available. A method described in a paper alone is not sufficient
4	A specific agricultural task	Agricultural, livestock, fisheries, forestry, food, financial, advisory, or regulatory
5	Official or primary source	An official product page, documentation, app store listing, public repository, or institutional publication page. An English source is sufficient to establish existence in this register, not to establish local availability, understanding, support, or suitability
6	Cut-off requirement	Evidence of existence on or before 31 August 2026
7	Determinable status	Active, commercial, institutionally published, limited availability, operational pilot, runnable open-source, discontinued, or acquired

4\. What is not counted as software — exclusion list

These are not counted as software, and do not enter the total, whatever their scientific value: a peer-reviewed paper; a thesis; a literature review or field review; a dataset; a benchmark; a model architecture without an implemented application; a conceptual framework; a general technical family; a general use case; a simulation without an operational system; a proposed design; an unnamed robot or prototype; a name whose responsible entity is unresolved; a company name without a specific smart-agriculture product; a marketing phrase that identifies neither a product nor a function; an internal feature within a larger platform counted as an independent product despite lacking an independent identity and independent distribution; and an agricultural system that uses only conventional automation without a documented artificial-intelligence function.

5\. The platform-and-component rule — to prevent double counting

The platform is counted once. Its internal modules and services are listed within its card or in an associated capabilities table, not as separate cards. A component is counted separately only if it is shown to have an independent product identity, independent distribution, and use outside the parent platform.

Changes of name, acquisition, merger, and discontinuation are tracked through an internal relationship field — component-in, renamed-to, acquired-by, merged-into — so that a single product does not appear twice under two names.

6\. Cut-off rule

Software and systems were included if their public announcement, publication, availability, or operation could be verified no later than 31 August 2026\. A source accessed after that date may be used only to verify prior existence or the product’s status at that date, not to add a product that appeared later. This use is explicitly marked as a retrospective source.

7\. Search and verification protocol

Accepted sources for discovery and verification are: official company and product websites, official documentation, mobile app stores, public software repositories, the pages of governments, universities, research institutes, and international and development organisations, as well as project reports and institutional publications. Commercial directories, news reports, and market reports may be used only to propose a candidate, and are not sufficient on their own for inclusion: the candidate must be resolved to an official or primary source before acceptance.

Language rule — stated plainly: searching and verification were conducted primarily in English, because the overwhelming majority of qualifying software has an English page, documentation set, store listing, repository, or institutional description. When no official English source exists for qualifying software, its official source in the original language is used, and this is indicated. This guide does not claim to have conducted a multilingual survey, and that was not a condition of its validity. The contrary claim that appeared in an earlier edition has been removed.

This rule concerns only establishing the software’s existence in a dated global register. A local use decision, by contrast, requires examination of the interface, terminology, warning, unit, writing direction, voice, support, contract, training, and appeal route in the language of the actual user. English does not become an assumed operating language, and the absence of multilingual verification does not become evidence of the absence of software in unexamined languages.

Limits of discovery: discovery is broad, not exhaustive. An application family is not padded with a search item, concept, or framework merely to avoid leaving it empty.

8\. Map of application families

Code	Application family	Coverage status in this edition
AT01	Farm management, planning, and decision-support systems	Covered
AT02	Agricultural advisory services, intelligent assistants, conversation, and knowledge retrieval	Covered
AT03	Remote sensing, satellites, and geospatial intelligence	Covered
AT04	Agricultural weather, climate, early warning, and risk management	Covered
AT05	Soil, site selection, characteristic mapping, and crop suitability	Covered within farm-management platforms
AT06	Crop monitoring, and forecasting of growth, yield, and maturity	Covered
AT07	Irrigation, water management, leak detection, and pumping control	Declared gap — no qualifying software was confirmed in this edition
AT08	Fertilisation, nutrients, variable-rate fertilisation, and environmental management	Covered within farm-management platforms
AT09	Plant diseases, pests, weeds, and integrated management	Covered
AT10	Robotics, autonomous tractors, drones, and spraying and targeting	Covered
AT11	Plant breeding, genomics, phenomics, and seeds and nurseries	Covered
AT12	Greenhouses, vertical farming, and indoor agriculture	Declared gap
AT13	Livestock, precision feeding, health, welfare, and reproduction	Covered — cattle and dairy
AT14	Aquaculture and fisheries	Covered
AT15	Forests, agroforestry, and rangelands	Partially covered — within a development programme
AT16	Bees and pollinators	Covered
AT17	Harvest, post-harvest, grading, quality, and food safety	Declared gap
AT18	Markets, finance, insurance, logistics, and traceability	Covered — agricultural insurance
AT19	Carbon, biodiversity, energy, waste, and the circular economy	Declared gap
AT20	Government oversight, compliance, agricultural quarantine, and biosecurity	Covered — fisheries oversight
AT21	Integrated multifunction agricultural platforms	Covered

These are twenty-one families, not twenty; a wording error in an earlier edition that said ‘the twenty families’ has been corrected, although the table contains AT01–AT21.

Reading the gaps: four families have no qualifying card in this edition — irrigation and water management, greenhouses and vertical farming, harvest and food safety, and carbon and sustainability. In most cases, the reason is that what had previously been listed under them were general technical families, reviews, or unresolved names rather than software, so they were removed in the correction. These are not domains devoid of tools worldwide; they are domains in which no software meeting the seven conditions was confirmed in this edition, and they are a priority for the next edition.

9\. The guide’s four dimensions — instead of a single grade

The previous edition used a single scale (E1–E6) that mixed different matters: the system’s existence, its operational status, scientific peer review, evaluator independence, and agricultural impact. That scale has been abolished, because combining these dimensions into a single number always conceals one of them. It is replaced by four independent fields:

Dimension	Values
Implementation evidence	Official product page · Official documentation · Store listing · Public repository · Institutional publication page · Examined private-source trace
Availability status at the cut-off	Commercially active · Institutionally published · Limited availability · Operational pilot · Runnable open-source · Discontinued · Acquired
Evaluation independence	No published evaluation · Technical evaluation · Field evaluation · Evaluation independent of the developer
Impact bound	No impact claim · Vendor claim only · Limited study · Independently supported result

The most important distinction in this table is that a peer-reviewed study is a source of evidence, not a type of program, nor does it automatically prove independence of the evaluation. The study’s authors may be the system’s own developers; peer review constrains the method, but it does not create independence.

Core rule: the existence of a program does not prove operation; operation does not prove production readiness; and production readiness does not prove agricultural impact.

10\. Main register — implemented programs and systems

How to read: each entry has an AIP identifier (short for Implemented Program), a responsible entity, a specific agricultural function, a delivery channel, availability status at the cut-off date, an official source, the status of any independent evaluation, and an explicit limit. This register contains not a single entry without a specified responsible entity and a public official source — with one expressly marked exception: the sponsoring-entity entry.

The entries were deliberately renumbered with new identifiers, so that no old reference to a mixed AIA would carry over to a verified program entry.

Date of the latest public verification from official sources: 4 September 2026\. A targeted update was carried out for Farmer.Chat and GAIA sources on 18 September 2026, without claiming that all register entries were rechecked on that date. Where an official source was available after the cut-off date, it was used only to verify prior existence, and is marked as retrospective.

AT01 — Farm management, planning, and decision support

Identifier	Program	Responsible entity	Agricultural function	Channel	Status at cut-off	Official source	Independent evaluation	Limit
AIP001	Climate FieldView	Climate LLC — Bayer Group (United States)	Integrated digital farming platform: field maps, variable-rate planting prescriptions, performance analysis, and machine connectivity	Web \+ phone \+ machine-integrated Drive unit	Commercially active	climate.com	No independent evaluation found	The “+5 bushels/acre” figure is a vendor claim, not an independent result; geographical coverage and compatibility are conditional on equipment
AIP002	Cropin Cloud	Cropin Technology Solutions (India)	Cloud platform for agricultural production management, analytics, and modelling across multiple fields	Web \+ phone \+ APIs	Commercially active	cropin.com	No independent evaluation found	Claims of being “first globally”, and the numbers of models, crops, and areas, are quarantined without a dated document and independent definition
AIP003	Connecterra	Connecterra B.V. (Netherlands)	Analytics and decision support for dairy farms: unifying system data, generated operational summaries, and tracking the effects of decisions	Web \+ integrations \+ data interface	Commercially active	connecterra.ai	No independent evaluation found	The user testimonials shown are vendor claims; they do not substitute for impact measurement

AT02 — Advisory services, intelligent assistants, and knowledge retrieval

Identifier	Program	Responsible entity	Agricultural function	Channel	Status at cut-off	Official source	Independent evaluation	Limit
AIP004	PlantVillage AI	PlantVillage — Pennsylvania State University (United States)	Advisory assistant combining multiple models, tools, and third-party data — such as weather forecasts and soil data — to answer more specific agricultural questions	Web	Limited availability — explicitly stated to be in beta testing	plantvillage.psu.edu \+ llm.plantvillageplus.coop	No published evaluation of the assistant itself	The same entity states that it is “evaluating multiple models for accuracy, relevance, and cost”; it must therefore not be described as a mature service
AIP005	PlantVillage Nuru	PlantVillage — Pennsylvania State University, in collaboration with the Food and Agriculture Organization	Crop-disease diagnosis on phones in low-connectivity conditions, within a field advisory programme in Africa	Phone	Institutionally published	plantvillage.psu.edu	There is scientific literature on the system; evaluator independence was not resolved in this audit	The official documentation describes programmatic deployment; it does not provide a general accuracy measure transferable across countries and crops
AIP006	Farmer.Chat	Digital Green (development organisation)	Multilingual, multimodal conversational advisory assistant connected to agricultural extension services	Phone \+ web	Institutionally published	\[SRC039\]–\[SRC041\]	A user evaluation was conducted by an entity outside the product team in partnership with Digital Green, on a sample of 450 farmers in Kenya \[SRC042\]	The study permits description of use, trust, and self-reported action within its sample; it does not prove general agricultural accuracy or causal impact on yield or income

AT03 — Remote sensing and geospatial intelligence

Identifier	Program	Responsible entity	Agricultural function	Channel	Status at cut-off	Official source	Independent evaluation	Limit
AIP007	NASA Harvest	NASA Harvest Consortium — University of Maryland and the National Aeronautics and Space Administration (United States)	Monitoring global agricultural production and assessing food security by integrating Earth observation data and machine learning, to support policy, insurance, and early-warning decisions	Analytical platforms \+ open data and models \+ institutional reports	Institutionally published	nasaharvest.org	Extensive published scientific activity; the consortium itself is a research entity, so this does not count as an external evaluation of its performance	Its outputs are policy and statistical assessments, not a decision tool for an individual farm; no impact on the yield of a specific field is attributed to it

AT04 — Agricultural weather, climate, and early warning

Identifier	Program	Responsible entity	Agricultural function	Channel	Status at cut-off	Official source	Independent evaluation	Limit
AIP008	AClimate	Bioversity International and CIAT Alliance — within the CGIAR system	Seasonal and sub-seasonal agricultural climate forecasting platform, with climate scenarios, crop simulations, and an advisory bulletin generator, to support planting and harvest decisions	Web \+ bulletin generator	Institutionally published in several countries	aclimate.org	No independent evaluation was conclusively established in this audit	The official source is in Spanish, and no equivalent official English page was found; coverage is limited to the countries where the platform has been deployed, and it states this

AT09 — Plant diseases, pests, and weeds

Entry AIP005 (PlantVillage Nuru) mentioned in AT02 also serves this family, and is not repeated here in order to avoid double counting.

Corrective note: the previous edition listed under this family four peer-reviewed studies, an architectural concept, and two unresolved names, and counted them as programs. All have been removed: the studies to their scientific places in the chapters and bibliography, the concept to its description as a concept, and the two names to the register of unresolved leaders.

AT10 — Robotics and autonomous machinery

Identifier	Program	Responsible entity	Agricultural function	Channel	Status at cut-off	Official source	Independent evaluation	Limit
AIP009	LaserWeeder G2	Carbon Robotics (United States)	Laser weeding through plant-level visual targeting, using a vision model trained on a large agricultural image base	Towed machine with embedded system \+ customer cloud	Commercially active	carbonrobotics.com/laserweeder-g2	No published independent evaluation found	All published figures are vendor claims: up to 80% cost reduction, yield increases, chemical reduction, and numbers of farmers and countries. Farmer testimonials published on the site are not independent impact evidence
AIP010	Carbon Autonomy	Carbon Robotics (United States)	Autonomy kits fitted to existing tractors, with remote operator supervision and the ability to switch to manual control	Tractor-embedded system	Commercially active	carbonrobotics.com/carbon-autonomy	No published independent evaluation found	It is not full autonomy: the official documentation itself mentions remote human supervision and intervention when needed; compatibility is limited to specific tractor models

AT11 — Plant breeding and phenomics

Identifier	Program	Responsible entity	Agricultural function	Channel	Status at cut-off	Official source	Independent evaluation	Limit
AIP011	Hiphen — field phenomics system	Hiphen (France)	Automated measurement of plant traits at different stages through image analysis, to support breeding programmes and the testing of varieties and treatments; via satellites, drones, field vehicles, and handheld devices	Analysis service \+ data platform \+ field devices	Commercially active	hiphen-plant.com	A declared research partnership with the UMT CAPTE unit and joint publications — this is collaboration, not evaluative independence	The service is directed to research and breeding programmes and companies, not to the individual farmer; no impact on commercial yield is attributed to it

AT13 — Livestock

Entry AIP003 (Connecterra) mentioned in AT01 serves the dairy sector, and is not repeated here.

Corrective note: in the previous edition, this family included a research draft and two unresolved names. All three were removed: the draft to its scientific place, and the two names to the register of leaders.

AT14 — Aquaculture and fisheries

Identifier	Program	Responsible entity	Agricultural function	Channel	Status at cut-off	Official source	Independent evaluation	Limit
AIP012	Aquabyte	Aquabyte AS (Norway)	Underwater computer vision to estimate biomass and weight, automatically count sea lice, monitor welfare and behaviour, and forecast appetite and manage feeding in fish-farming cages	Submerged cameras \+ cloud platform \+ user portal	Commercially active	aquabyte.ai	Declared trial with Mowi Ireland and Ireland’s BIM agency — an operational trial with a third party, and no independently published results were found	The figures 'more than 800 systems' and '1.3 million images daily' are vendor claims; the core scope is salmon rather than aquaculture in general
AIP013	Observe Technologies	Observe Technologies (United Kingdom)	Aggregation of existing fish-farm data streams — vision, sensors, and acoustics — and generation of feeding and anomaly alerts, without replacing cage equipment	Software running on top of existing hardware	Commercially active	observe.tech	No published evaluation	The official site is brief and does not provide detailed technical documentation or a client list; commercial maturity is evidenced, but deployment scale is unknown

AT15 — Forests and rangelands

Identifier	Program	Responsible entity	Agricultural function	Channel	Status at cut-off	Official source	Independent evaluation	Limit
AIP014	PlantVillage — pastoralist and rangeland support	PlantVillage — Pennsylvania State University, in collaboration with FEWS NET and the United Nations	Guiding pastoralists to vegetation and water locations through advance forecasts, and linking them to veterinarians, to support climate adaptation in rangelands	Digital advisory service within a field programme	Institutionally published	plantvillage.psu.edu/food	No completed independent evaluation in this audit	A component within the broader PlantVillage programme, not a separately marketed product; it is listed here because its agricultural function is different and officially documented

AT16 — Bees and pollinators

Identifier	Program	Responsible entity	Agricultural function	Channel	Status at cut-off	Official source	Independent evaluation	Limit
AIP015	BeeHero — Precision Pollination and Pollination Insight Platform	BeeHero (United States, Israel, and Australia)	In-hive and in-field sensors to measure colony strength and bee visitation to flowers, turning pollination from a service bought on trust into a service that is measured	Sensors \+ analytics platform	Commercially active	beehero.io	No published independent evaluation was found	The displayed numbers of acres, hives, and data points are vendor claims; increases in fruit set must not be attributed to the service without a study
AIP016	Beewise BeeHome	Beewise (United States and Israel)	An automated hive combining computer vision and robotics to monitor colonies and intervene automatically in their care, instead of waiting for a beekeeper’s visit	Field hardware unit \+ remote control	Commercially active	beewise.ag	No published independent evaluation was found	The figure '70% reduction in colony loss' is an explicit vendor claim; the displayed awards are industry recognition, not a performance evaluation

AT17 — Harvest, post-harvest, and food safety

Declared gap. The previous edition listed six entries here: four research robotics models or cases whose identity could not be resolved, one domain review, and one family of general systems — none of them a program meeting the seven conditions. All were removed, and no replacement has been settled in this edition. Leaving this space empty is more truthful than filling it with what is not a program.

AT18 — Markets, finance, and insurance

Identifier	Program	Responsible entity	Agricultural function	Channel	Status at cut-off	Official source	Independent evaluation	Limit
AIP017	Pula — data-driven agricultural insurance	Pula GmbH (Switzerland, with operations in Africa and Asia)	Agricultural insurance for smallholder farmers based on an area-yield index and agricultural data services, alongside livestock insurance and compliance services for deforestation regulations	Insurance service \+ data platform \+ government partnerships	Commercially active and institutionally documented in twelve countries	pula-advisors.com	No completed independent evaluation in this audit	The figure 'more than 21 million farmers' is a vendor claim; access is through government and institutional partnerships rather than direct sale to farmers

AT19 — Carbon and sustainability

Declared gap. The previous three entries here were a tool within another platform (Cropin), not an independent product, and two names whose responsible entity remained unresolved. All three were removed; the first was merged into entry AIP002.

AT20 — Oversight, compliance, and biosecurity

Identifier	Program	Responsible entity	Agricultural function	Channel	Status at cut-off	Official source	Independent evaluation	Limit
AIP018	Global Fishing Watch Map	Global Fishing Watch (a non-profit organisation founded by Oceana, SkyTruth, and Google)	Monitoring global fishing-vessel activity by combining satellite imagery, vessel-tracking data, and machine learning, to support enforcement against illegal fishing and fisheries management	Public interactive map \+ open data and interfaces	Active and publicly available	globalfishingwatch.org/map	Open data and analyses in partnership with scientific institutions; documented enforcement cases with national authorities have been published	It serves oversight and governance, not fish-farm management; and it depends on tracking signals that may be deliberately switched off — a known limit in this type of monitoring

AT21 — Integrated multifunction platforms

Identifier	Program	Responsible entity	Agricultural function	Channel	Status at cut-off	Source	Independent evaluation	Limit
AIP019	Aladdin Agri AI	Aladdin Agri AI — the sponsoring entity of this book	A governed agricultural AI environment across multiple channels and languages, covering the seasonal cycle from land selection to post-harvest, with a central AI governance layer, knowledge retrieval, translation, and feasibility studies	Web \+ mobile server gateway \+ desktop application	Limited availability; maturity varies by channel	Audited private source record — no public source	No independent evaluation	See disclosure and capabilities below

Platform capabilities — counted as components within it, not as independent programs:

Component	Function within the platform	Why it is not counted as an independent program
AiBridge	Central governance layer that routes AI requests, applies policy, and records activity	No independent product identity or distribution outside the platform
Chat V2	Conversational and retrieval path with source citation	An internal service operating through the governance layer
AgroGenie	Directory-mediated agricultural search linked to the interface and to the conversation	Its two gateways are closed by default, and it is neither distributed nor used outside the platform
Mobile Gateway	Mobile server gateway with contracts, routes, signing, and sessions	A server gateway, not a distributed client application
Desktop	Desktop application for Windows	No production code signing and no store distribution

The previous edition counted AgroGenie as a separate entry alongside the platform, and that double counting is corrected here: the platform is counted once.

Mandatory disclosure on AIP019: the sponsoring entity of this book is the subject of this entry, and the same standard applied to others is applied to it without relaxation or tightening:

The historical source layer is defined: bundle Koot-EL-Kuloob-2.1.39 dated 22 August 2026, with a SHA-256 fingerprint declared in Appendix N, and it forms the scope \[IMP01\]–\[IMP09\].
The official distribution layer is separate: the bundle Koot-EL-Kuloob-2.1.39.zip generated on 22 September 2026 and examined on 10 October, with size and fingerprint declared in \[IMP14\]. It establishes what passed distribution-integrity checks and the presence of the specified files, contracts, and checks, and it does not rewrite the August findings.
Execution evidence is a flagged exception: this is the only entry in the register that rests not on a public source but on a private source record examined internally. That must not be counted as public or independent evidence. It is included under this explicit label, not by special dispensation.
Source presence does not mean operation: nor does the number of entries or routes mean that all services are active.
The server gateway is not the client application: this is the finding of the specified August record, and it must not be transferred to another bundle or channel without separate examination.
Local testing is not production distribution: the limited local live verification recorded in the August record does not establish public acceptance, and \[IMP14\] did not run WordPress live and did not verify flags, database, provider, browser, or interface acceptance.
Language files are not a localisation certificate: the presence of catalogues or the success of an isolated check does not establish the completeness of visible text, the correctness of terminology, units, and warnings, or user understanding and ability to act, and \[IMP14\] recorded the specified limits of language testing.
No preferential ranking: this entry is given neither a better position nor a better description than others, and no claim of superiority is made for it, because no homogeneous independent comparison exists.
No independent agricultural impact: no field trial and no usage study.

Note on comparability: the other entries in this register have public sources that the reader can open; this entry does not. It is therefore less open to external verification than all those before it, even though it has the most detailed internal documentation. Detail does not compensate for availability.

11\. Descriptive statistical summary

A note before the figures: what follows describes this register, not the global landscape. Nor may the number of cards in any family be turned into a judgement about its activity, size, or the quality of its tools.

Indicator	Value
Programs and systems with actual implementation in the main register	19
Defined application families	21
Families with at least one card	12
Families marked as a declared gap	4
Families served by a card listed under another family	5
Cards with an official or public primary source	18 of 19
Cards without a public source (marked exception: sponsoring entity)	1
Cards with an evaluation independent of the developer	zero
Cards with an independently supported agricultural-impact result	zero

Distribution by entity type: nine commercial companies, four university or institutional programmes, two development programmes, one non-profit organisation, the sponsor platform, and one joint institutional programme — a total of nineteen.

Distribution by availability status: eleven commercially active, five institutionally published, two with limited availability, and one announced as being in a pilot-testing phase.

Three readings of these figures — limited to this register alone:

The first is that the number fell from forty-seven to nineteen, and this is an improvement, not a retreat. The previous figure counted papers, reviews, simulations, concepts, and names of unknown provenance alongside products. The current figure counts programmes only. A smaller, truthful register is better than a larger, mixed one.

The second is that eighteen cards out of nineteen carry a public source that the reader can open. The previous edition left thirty-six cards out of forty-seven without a direct source citation. This is the practical difference between a directory and a draft.

The third — and most important — is that, in this audit, no published evaluation independent of its developer was found for any of the nineteen programmes. This is a note on the limits of this register and its examined sources, not a judgement on the world; independent evaluations may exist that this research did not reach. But it is enough to warn the buyer: the numbers you read on product pages — cost reduction, yield increase, and reduced colony losses — are, on every card here, supplier claims, not independently measured results. The difference between the two is the difference between a promise and evidence.

12\. What left the main register and where it went

This subsidiary register records what was removed from the main register through correction, and it was not deleted from the book. The destination of each category is stated here so that no reader assumes that scientific material has been dropped:

What was removed	Number	Where it was moved	Why it was removed
Peer-reviewed scientific studies and preprints	9	Their chapters in the main text, the verified reference list, and the case-study register in Appendix A	A paper is not a program; it evidences a method or a result, not a system in operation
Field reviews	2	"Sources that frame the field" in the relevant chapters	A review maps a field and is not a program
Simulations and computational experiments	2	Described in their respective places as simulations and computational experiments	A simulation tests an idea and is not a published system
General technology families	6	Appendix H as technology trends	"Smart irrigation with the Internet of Things" is not the name of a program
Concepts and architectural frameworks	2	Described as architectural concepts	A framework organises thinking and is not software until an implementation exists
Names whose developing entity could not be determined	12	The register of unresolved entries in Appendix H	No responsible entity and no official source; both are mandatory conditions
A component merged into its parent platform	1	Within card AIP019	To prevent double counting
A tool within a broader product	1	Within card AIP002	Not a separately distributed product

The total is thirty-five removed from the count as programs, while remaining in the book as what they actually are.

The rule for moving an entry from "unresolved" to a card in the main register: all seven conditions must be fully satisfied — name, entity, actual implementation, agricultural function, official source, adherence to the time boundary, and a determinable status. If the source establishes only the existence of a feature, the impact remains "unknown". If a primary source contradicts the supplier's description, the conflict is presented and neither side is preferred without a stated reason.

13\. Gap matrix

Dimension	Gap observed in this edition
Application families	Four families without a qualifying card: irrigation and water management AT07; greenhouses and vertical farming AT12; harvesting and food safety AT17; carbon and sustainability AT19
Independence of evaluation	In this audit, no published independent evaluation was found for any of the nineteen programmes
Impact	All quantitative figures presented in the register are supplier claims; none is an independently supported result
Region	Stronger coverage of North America and Western Europe; weaker coverage of East Asia, Eastern Europe, Central Asia, and the Middle East
Entity type	Weak representation of national government programmes and operable open-source tools
Temporal status	No systematic tracking of discontinued or acquired products; therefore this edition contains no section on discontinued products
Language	Verification was conducted primarily in English, and one card (AIP008) has its official source in Spanish. A multilingual survey was neither conducted nor required
Local suitability	This edition does not record systematic verification, country by country, of agricultural language, units, devices, connectivity, payment, service, maintenance, authority, and fallback; therefore inclusion may not be treated as a local recommendation

This matrix is the most useful part of the appendix for the researcher: every row in it is a ready-made research question and a proposal for a coming edition.

14\. Non-endorsement notice

Listing any program in this directory does not constitute an endorsement of it, nor proof of its effectiveness, safety, accuracy, production readiness, or suitability for any particular agricultural context. Nor does the absence of a program constitute evidence either that it does not exist or that it is poor. The order within the tables is objective and carries no preferential meaning. This appendix does not provide purchasing or investment advice.

Before any procurement decision, refer to the workbooks in Appendix T: define the problem before choosing the tool, identify the decision to be supported, the evidence required, and responsibility in the event of error.

15\. The accompanying digital register and correction mechanism

The printed book is, by nature, a dated snapshot, and the landscape changes: pages change, companies are acquired, and products are discontinued. Therefore the directory is issued on two levels:

Level	Content
In the book	the unit of inclusion, the seven conditions, the exclusion list, the platform-and-component rule, the four dimensions of evidence, the master register, the gap matrix, and the cut-off alert
Companion digital register	a searchable, filterable version, a change history, archived links for every official source, post-August 2026 updates, identity corrections and product-status cases, CSV download, and a stable link in the book

Authorised fields in the digital register: identifier, standard name, entity type, parent identifier and relationship, developer, country, application family, agricultural task, target user, delivery channel, date of first public evidence, status at cut-off, evidence of implementation, independence of evaluation, impact boundary, official source, independent source, date of last verification, limitations, conflicts of interest, and archived link. Separate fields are also added, but not populated by guesswork, for operating languages, terminology, units, devices, connectivity, geographical coverage, support, maintenance, alternative, and local regulatory authority.

Correction mechanism after publication: every correction is recorded with its date, reason, and source, and the old version is not erased in a way that loses the trace of the decision—it is retained, archived, and marked. Any owner of a listed program may request correction of its data by providing an official dated source; such a request is carried out only within the bounds of factual correction, and neither deletion of an entry nor elevation of an evidence grade is accepted on the basis of a commercial objection.

16\. From programs back to those who use them

Nineteen names in a table do not tell the whole story. Behind every card here stands someone who does not appear in the register: a farmer deciding when to irrigate and how much, a beekeeper driving for hours to inspect hives, a herder looking for water and forage, a worker standing on a sorting line, a veterinarian waiting for a call, and an inspection officer trying to determine who caught what and where.

And a program that serves one of these people is not measured by the number of its models or the volume of its data, but by a simpler question: did the information reach, in time and in that person's own terminology and units, the person who bears its consequences, and was that person able to understand it and act on it or object to it, and was there support and an alternative when it failed? That is what a global register of presence cannot measure, however precise its classification.

For that reason, this appendix should be read for what it is: a map of what has been built, not a judgement on what has been beneficial. That second judgement requires something that does not yet exist in this field in sufficient measure—an independent evaluation that asks the farmer directly, not the product page.

General Conclusion

The book began with the decision, and to it returns

This book began with pressures on water, soil, crop, animal, labour, and market, not with a desire to find a new place for artificial intelligence. After passing through images, sensors, records, machines, extension, finance, and supply chains, the conclusion returns to the first question: which agricultural decision needs support, who will bear its consequences, and what evidence is sufficient to make it in time?

There is no single technical answer to this question. The need may be for better measurement, a clearer record, a routine operating rule, co-ordination among a worker, a carrier, and a warehouse, or repairing a network before buying a model. The value of artificial intelligence may emerge when signals and alternatives multiply and cannot be analysed manually within the decision window; but it does not turn observation into a diagnosis, a forecast into an outcome, or a recommendation into authority to execute.

For this reason, the book distinguishes between measurement, information, decision, action, and outcome. Confusing them makes model accuracy look like preserved yield, the mere presence of a component look like operation, or digital recording look like impact. Distinguishing them shows where the benefit lies, where the fault lies, and who reviews, who authorises, and who stops.

Localisation is a condition of validity

Professional localisation is not interface translation or the replacement of an example from one country with another. It is the definition of the decision in its agricultural substance, place, season, and window, and its linkage to the user, that user's authority, language, terminology, unit, device, and connectivity, and then to the institution, reference, service, maintenance, alternative, and measured outcome. If any element affecting the action or its consequences changes, the previous evidence remains useful knowledge, but it becomes a hypothesis requiring local verification.

This does not mean abandoning general science or shared standards. Provenance, measurement, safety, rights, and auditability are standards that do not change from place to place. What changes is their implementation: the threshold, the name, the unit, the authorised body, the service pathway, and the user's ability to understand, act, and object. A responsible system preserves these differences instead of erasing them in the name of standardisation.

Local knowledge is neither infallible truth nor noise to be cleaned away. It is attributed to its holders, preserved in its language and context, tested, and its holders retain the right to correct and reject. A difference in naming may reveal a real difference in a variety, stage, or practice; therefore normalisation begins with linking, not replacement.

Not every digitisation is artificial intelligence

Collecting invoices, recording operating hours, calculating consumption, applying a declared limit, and preserving a copy, date, and validity are all important functions that ordinary software can perform. Attributing them to artificial intelligence does not increase their value; it sometimes conceals that what is needed is a clear rule, a disciplined record, or a clearly identified responsible person.

The usefulness of artificial intelligence appears in specific places: linking many patterns, estimating a range with uncertainty, ranking cases for review, detecting an anomaly or a conflict, comparing scenarios, and retrieving sources appropriate to the question. Even here, limits remain necessary: it does not invent a missing value, does not turn a supplier promise into an available service, does not carry a threshold from one context to another, does not approve its own recommendation, and does not bypass a safety gate or a human right.

Abstention is a professional outcome when the data are insufficient or the question falls outside scope. A system that makes clear what it does not know, escalates to a capable authority, and leaves an inspectable record is more useful than a system that fills the gap with a confident answer.

Economics means the ability to continue

Final cost and actual benefit are not known until the work is complete and the outcome has appeared. Therefore the book does not ask the farm for a certain figure before the season; it asks for a baseline, an updatable estimate, and a separation between what is known and what is assumed, and then for the comparison to be closed with what was paid, what failed, and what quantity, quality, and value were actually preserved.

The decision to buy, lease, contract for, or share remains a service decision, not merely a price decision. The cheapest machine is of no use if it arrives after the harvest window, and ownership does not guarantee continuity if the operator, technician, or part is absent. Repairability, the ability to work manually or locally during an outage, and rights of export, transition, and exit are parts of viability, not secondary items after purchase.

A system that the farm cannot leave safely is not mature, however easy it is to enter. Data ownership does not become a practical right until its holder can understand what will be exported, recover it in another system, correct errors, and terminate the contract without losing a critical function.

The human holds authority, not a merely formal signature

A human presence on the screen does not protect a system that gives that person neither time, information, authority, nor an alternative. Anyone asked to accept an output they do not understand is not a reviewer; anyone punished for objecting has no authority; and anyone unable to run a critical function when the cloud fails is not part of a continuity plan.

For this reason, the book apportioned the roles: who collects, who interprets, who reviews, who authorises, who executes, who stops, and who investigates the incident. Some roles may be combined on a small holding, but responsibility does not disappear; it is named, and the escalation path and the procedure in case of absence are known.

Automation does not eliminate work so much as redistribute it. With it come monitoring, maintenance, intervention, data cleaning, and alarm review. Responsible measurement is not satisfied with hours reduced; it asks who learnt the new task, whose burden increased, who lost income or authority, who received the benefit, and what protection is available during the transition.

Evidence does not travel alone

A useful number is a complete sentence: who measured what, on what material and data, in what place and season, with what unit and denominator, compared with what, under what version and conditions, what the uncertainty was, and what happened after the output. If these elements are missing, the number does not automatically become false, but it loses part of its power to support a claim and travel.

For this reason, the appendices preserved studies, programs, and claims in distinct ranks. The existence of a program does not prove that it operates; operation does not prove acceptance of its release; acceptance of the release does not prove local suitability; and suitability does not prove agricultural or economic impact. Likewise, a peer-reviewed study may establish a result in its own setting and still not be a product, while a product page may establish the existence of a feature and still not be an independent evaluation.

Failure also has a knowledge function if it is recorded in context. A false alarm, a worker objection, a part that did not arrive, a translation that weakened a warning, or a result that did not recur in another season are not embarrassing margins; they are data that redefine the limits of the system and prevent the error from being repeated.

What the Aladdin study establishes, and what it does not establish

The Aladdin study presented engineering material on a multi-module platform that attempts, through central governance, to separate the use of artificial intelligence from the supplier and to link the agricultural decision to the record, the source, and different channels. But this material must be read through its temporal layers, not through the release name alone.

The trace dated 22 August 2026 concerns specific historical findings recorded in \[IMP01\]–\[IMP09\]: the existence of particular structures, contracts, paths, and tests, together with limited local live verification of some conversation and retrieval paths, while release acceptance remained open in that trace. \[IMP13\] concerns the later AgroGenie development package and isolated tests that do not establish live WordPress operation or field impact.

As for the official distribution package Koot-EL-Kuloob-2.1.39.zip, generated on 22 September and examined on 10 October 2026, it represents an independent layer \[IMP14\]. Its fingerprint, the integrity of the archive paths, and the presence of the runtime files, contracts, components, and test results specified in its record establish what is present, while at the same time revealing constraints in integrity sealing, visible messages, the interactive-book gates, and the AiBridge test suite. It does not establish live WordPress operation, the status of flags and the database, provider connectivity, interface acceptance, the completeness of translation and localisation, the correctness of an agricultural recommendation, or field impact.

The later package does not rewrite the earlier trace retroactively, and the earlier trace does not confer runtime results that were not repeated on the official package. More importantly, all fourteen items of evidence are internal and private; the reader cannot inspect them, and they carry no public persistent identifier. They are therefore suitable for describing what was examined within their limits, but they do not become an independent review, a scientific reference, or a certificate of agricultural value.

This distinction between what is established and what is not established is neither a diminishing of the work nor a promotion of it. It is the condition that makes the sponsor's study open to criticism and prevents engineering detail from turning into a promise of operation, localisation, or impact.

The next path is gates, not a fixed timetable

A project does not mature because months have passed or because the budget has been spent. It moves forward when it closes an evidence gate proportionate to the risk: a local definition of the problem, a baseline, and rights; then a reversible trial; then external verification and conditional transfer; then monitoring across a production cycle and representative shocks; then continuation, adjustment, or exit according to the outcome.

Research works in the same way. A good question defines the decision, the material, the place, the season, the user, the comparator, the outcome, the authority, and the limits of transfer. Ordinary software preserves the protocol, versions, permissions, and raw data. Artificial intelligence may help to extract variables, detect values without units, or group incidents for review, but it does not choose the criterion of success, decide causality, fairness, or transferability, conceal a negative result, or authorise publication of its own work.

The right to stop is part of the method. Before starting, it is known who can stop, what harm, inequity, or futility triggers that action, how data, agricultural material, and participants are protected, and how the negative result is preserved. Sunk cost does not cancel this right.

A Final Word

Smart agriculture will not take a single form. It may be a simple device operating locally, a shared service run by a co-operative, an adviser using documented retrieval, a machine for a specific task, or a platform linking several stages. In some places, the best decision may be to fix the existing process and defer digital investment.

The standard is not how much intelligence is on display, but the quality of the decision made possible: was it defined in its context? Did it arrive in time and in terms understood by those who would bear its effects? Was it based on evidence with known limits? Did authority and responsibility remain clear? Did the critical function continue during disruption? Were repair, objection, export, and exit possible? And were the outcome, and the distribution of benefit and harm, measured, after which the system either learned or stopped?

If the system can answer these questions with local evidence that can be examined, it does not become infallible; it becomes worthy of a responsible, reversible trial. If it cannot, neither the number of models, nor the breadth of the interface, nor the repeated use of the word ‘smart’ will make up for that. The future of agriculture does not need a machine that claims to know everything, but institutions and tools that know their limits, preserve the meaning of place, and keep decision-making intelligible, reviewable, and open to repair.