ClyCites — Field Validation Kit¶

For: four cooperative visits + one lender conversation Timebox: two weeks Purpose: find out whether the fourteen entities describe anything real, and produce the conversion data the kernel cannot invent.

What this is for¶

Spec §13 says the schema is a hypothesis and should not be frozen from a desk. This is how you test it. Three outputs:

Photographs of the paper — every field on a form someone already fills in by hand is an input that cleared a much higher bar than one we imagined.
Measured conversion factors — real weights for local units, so UnitConversion rows can carry basis: "measured" instead of assumed_default.
Answers to four open decisions — D2 (duplicates), D5 (seasons), D7 (delegation scope), D8 (observation types).

You are not selling. You are not demoing. See §7.

1. Who to visit¶

Four cooperatives. The selection matters more than the count.

Criterion	Why
Four different districts	Conversion factors and season calendars vary regionally. Four coops in Wakiso tell you about Wakiso.
At least two different commodities	A bag of maize is not a bag of beans. Coffee, matoke, and grain have different handling entirely.
At least one badly run coop	Well-organised coops are unrepresentative. The messy one tells you what the schema has to tolerate.
At least one that already sells to a formal buyer	WFP, a processor, an exporter. That is where Agreements and two-sided confirmation actually exist on paper.

Use whatever warm intros you have — that beats cold-calling a better-fitting coop. District agricultural officers and existing farmer-group networks are the usual route in.

2. What to photograph¶

Ask for everything, photograph what they'll let you. The layout, the crossings-out, and the margin notes are data — a field people keep correcting is a field that matters.

Delivery / receiving book — the single most important document
Member register
Weighbridge or scale slips
Input loan or credit cards
Any contract or offtake agreement
Payment records — how they record that a farmer was paid
Store or stock cards
Anything taped to the wall. Price boards, conversion charts, grading posters. Wall artefacts are the coop's real reference data.

For each form, note: who fills it in, when, who else sees it, and what happens when it is wrong.

Privacy. These pages carry farmers' names, phone numbers, and sometimes NINs. Under the Data Protection and Privacy Act you need a lawful basis to hold that. Practical approach: ask the coop manager's permission explicitly, photograph blank or old forms where possible, blur names before the images leave your phone, and do not upload anything to a shared drive. If a form is only useful with real data on it, note the field names rather than keeping the image.

3. The weighing exercise — do not skip this¶

This is the part that directly unblocks the kernel.

Bring a hanging scale. A 200kg digital crane scale is inexpensive and is the single most valuable object in your bag.

At each site, for each local unit in use:

Weigh at least ten filled containers, not one
Record every weight, not the average — the variance is the finding
Note the commodity, the container type, and the local name for it
Note whether it was packed loose or tight, and whether it had been dried

Field	Example
District
Commodity	maize grain
Unit as spoken	kaveera
Unit type	bag
Weights (10+)	96, 103, 99, 101, …
Packed	tight / loose
Dried	yes / no / partly
Who says this is standard	coop manager / buyer / nobody

Ten weights per unit per commodity per district gives you UnitConversion rows with basis: "measured" and a defensible factor. Everything else in the registry stays assumed_default, and the metric on /metrics tells you how much of your tonnage rests on guesses.

Ask separately: has the bag weight changed in the last few years? If yes, you need valid_from / valid_to populated, not just a factor.

4. Interview guide¶

Eight questions. Ask them conversationally, not as a form. Follow tangents — the tangents are usually where the finding is.

1. Walk me through what happens when a farmer arrives with produce. Open, first, and do not interrupt. You are looking for steps that are not in the schema.

2. Who is allowed to sign or record for a farmer, and how does that person get that authority? (Open decision D7.) If the answer is "it's just understood" or "whoever is at the desk," then Delegation as specified invents a ceremony nobody performs, and the model needs rework.

3. What happens when two farmers have the same name? (D2.) Their existing practice tells you whether to merge or link. If they use a member number, ask to see how it is assigned and whether it is ever reused.

4. What do you call the growing seasons here, and when do they start? (D5.) Whether there is a shared vocabulary, and whether it is regional.

5. When the weights don't add up at the end of the season, what do you do? The realistic mass-balance discrepancy, and whether losses are recorded at all or just absorbed.

6. Show me a time this record was wrong. What happened next? The correction workflow. This tests supersession against reality.

7. What do you measure that isn't on any of these forms? (D8.) Moisture, colour, insect damage, dryness by bite test. Candidate observation_type entries.

8. If a bank asked you to vouch for a farmer's production, what would you show them? The most important question in the set. It tells you what a coop believes is credible evidence, which is the whole thesis in one answer.

5. The lender conversation¶

Separate visit. One credit officer at a SACCO, microfinance institution, or an agri-lending bank.

Bring a printed copy of the Appendix A delivery record from the spec. Ask exactly this:

Would you lend against this? If not, what is missing?

Then shut up and write down everything they say.

Follow-ups if the conversation allows:

What do you do today to assess a farmer?
What does it cost you when you get it wrong?
Would a guarantor still be required if you had two years of this?

If the answer is "we'd need a guarantor regardless of the data," the credit thesis needs rework, and finding that out in August is worth more than the entire kernel.

6. What to bring¶

Hanging scale (200kg) and a spare battery
Phone with plenty of storage, plus a power bank
Printed Appendix A record for the lender
Notebook — write down what people say, not your interpretation
Something for the coop's time: transport reimbursement, or lunch. Do not turn up empty-handed.

7. What not to do¶

Do not demo the software. You will get politeness instead of information, and they will start describing what they think you want.
Do not pitch. You are asking for help understanding their work.
Do not promise anything — not a pilot, not a payment, not a feature. A promise you cannot keep costs you the relationship and the district.
Do not correct them when their practice contradicts the schema. That contradiction is the finding. Write it down and move on.
Do not collect farmer personal data you have no lawful basis to hold. §2.

8. What to bring back¶

One document per coop, plus one summary:

Photographs, organised by coop and form type
The weighing table, raw weights included
Answers to the eight questions, quoted where possible
A field list: every field appearing on their forms, marked as appearing at 4, 3, 2, or 1 of the sites

Then apply the §13 test:

Appears at	Verdict
All four	Core
Two or three	Extension
One	Extension, low priority
None — but is in our spec	Justify or cut

That last row is the one to be honest about. A field in spec v0.2 that appears on no real form is a field we invented, and the burden is on us.

Then¶

Two more things only you can do, both cheap and both parallel:

One lawyer's hour on D3 (erasure versus append-only) and D9 (legal weight of Agreement). Both can force schema changes, and both are expensive to guess.
The resourcing question. Fourteen entities, offline sync, and seven applications is not a nights-and-weekends scope. Grant, raise, or cofounder — the funding shape decides whether you build for a pilot or for a standard, and it is a design input, not an afterthought.