What Happens When a Cocoa Pesticide Question Has No Reviewed Answer?

A reviewed-answer library lets an extension officer spend time on uncertain cocoa cases instead of repeating settled guidance. It works only when the library is reviewed, the limits are clear, and every uncertain or safety-sensitive question reaches a named person.

In April 1970, Apollo 13’s carbon dioxide problem became urgent after the command module’s square filters could not fit the lunar module’s round openings. Engineers in Houston worked from the limited materials available on the spacecraft and sent the crew a procedure to build an adapter. The outcome was uncertain until the crew assembled it and the air improved.

NASA’s account of Apollo 13 records the value of that moment accurately: the ground team did not replace the crew’s judgment. It gave them a checked response for a known, immediate problem, so attention could stay on everything else required to bring Jim Lovell, Jack Swigert, and Fred Haise home.

An extension officer faces a smaller version of that allocation problem. A farmer may ask a familiar question about cocoa pod disease, shade, pruning, or where to seek help. Another may mention an unclear pesticide product, an unfamiliar symptom, a recent spraying, or a possible risk to people in the field. Those questions should not receive the same treatment.

A safe answer begins with a boundary

A useful reviewed-answer library contains advice that an agriculture reviewer has already approved for a narrow situation. The system identifies the matching answer and delivers it in language the farmer can understand. It does not create new agronomy guidance from a plausible-sounding prompt.

That boundary matters because fluent language can hide a serious error. Neuralis has already seen machine translation turn the Twi word for cocoa, `kokoo`, into “chicken.” In agricultural advice, that kind of substitution can send a conversation in the wrong direction before anyone notices.

The same caution applies to chemical questions. If a library does not contain verified dosage, re-entry, and pre-harvest guidance for a product and situation, the right response is to withhold the instruction and escalate. A polite refusal with a clear handoff protects the farmer better than an invented answer. What Happens When a Pesticide Question Has No Reviewed Answer? explains why that pause is part of the service, not a gap to hide.

The officer receives the questions that need judgment

Picture the extension officer opening a queue late in the afternoon. A farmer has asked a common question that matches a reviewed block. The answer can go out promptly in spoken Asante Twi, without requiring the officer to repeat the same explanation for the tenth time that week.

Then another question arrives: the farmer gives a pesticide name that does not match the reviewed list, or asks when people can safely return to a sprayed field. The system should mark it uncertain, preserve the original question, and route it to the named officer. It should not try to fill the silence with confidence.

That is how reviewed answers reserve human time. They reduce repeat work where the answer is already known and reviewed. They make the uncertain cases visible instead of burying them among routine messages. The officer can investigate the farm context, check the product label or local guidance, and decide what applies.

For the farmer, the difference is practical. A fast answer helps when it is genuinely safe. A clear escalation helps when the answer depends on details the system does not have.

Build the library around real field questions

A library should start small and stay accountable. Neuralis’s AgriVoice pilot plan centres on 37 owner-approved blocks, reviewed spoken Asante Twi, and a Ghanaian agronomist’s decision on held chemical content. Before farmers use it, the pilot also requires a cocoa-sector partner and a named extension officer who receives escalations.

The content needs to reflect the questions farmers actually ask, including ambiguous speech, code-switched language, and questions distorted by speech recognition. A library built from neat written examples can fail when a voice note arrives with background noise, a local product name, or a symptom described in everyday terms.

Each answer also needs a clear trigger for escalation. That could be a missing product match, a question outside the reviewed topics, low confidence in the selected block, or any request for chemical instructions that lack verified safety details. The escalation owner needs a response-time commitment and a queue they can act on.

Measure whether the handoff protects people

Apollo 13’s improvised adapter mattered because it worked under real constraints. A reviewed-answer library deserves the same standard: evidence from real use, rather than confidence from a demo.

For AgriVoice, success means at least 70% of farmer questions are answered or correctly escalated, with no unsafe pesticide answer reaching a farmer. It also means farmers understand the response, use the service again, and can reach a person when the system cannot safely help.

The library does not make the extension officer less important. It makes their expertise more available at the point where it carries the most weight. Like the Houston team in 1970, the goal is to prepare checked procedures for known problems while keeping skilled people focused on the decisions no procedure can safely make.