ADR-0034: The Corpus is sized to one production area, not to a document count

Status: Accepted — amends ADR-0011 and ADR-0028, both of which stand Date: 2026-09-08

Decisions

D-79 The Corpus covers every chemical handled in one Study Area, and its size is whatever that number turns out to be

The Study Area is the single production area where the needs study runs. Coverage inside it is complete; outside it the app says so, as it already does for any chemical not in the Corpus. The "twenty to forty documents" figure is replaced as a planning target — not because it was too many, but because a count was never the right unit.

D-80 The Study Area's chemical list is requested from The Plant ahead of the binder, and is a smaller ask than the binder

A list of what one area handles needs no document handover and can be answered by a supervisor. It sizes the work, it identifies which documents to request first, and it is a test of access that costs The Plant almost nothing to pass.

What each amendment touches

Stated precisely, because "amends" is a claim the status gate cannot check.

• ADR-0028 — its Consequences established twenty to forty documents as the planning figure. D-79 withdraws that figure as a target. D-58's per-document cost and D-59's floor are untouched.
• ADR-0011 — D-14 stands entirely: the Corpus is still The Plant's own holdings and nothing else. What narrows is its reading of R-16, "the chemical list is now whatever the binder holds", which becomes whatever the Study Area holds. The binder remains the source; it is no longer the scope.

Context

ADR-0028 measured Curation at half a day to a full day per document and ruled the figure a floor (D-58, D-59). Its Consequences named the only lever: "If that is unaffordable, the lever is corpus size — and cutting it means the app answers for fewer chemicals, which is a product decision for the owner, not something to absorb by rushing review." The spec repeated it: "corpus size is the schedule."

That framing left the decision as a number to be haggled down, and a number is the wrong unit.

D-01 makes this an academic deliverable whose study objectives are the point, and D-31's needs study is what answers them. Consider a Corpus of twelve documents chosen for availability, and a participant who handles thirty chemicals on their line. The study then measures Not Stated and Corpus Unavailable in most sessions, which tells nobody whether the product works — only that it is empty. The number was affordable and the study was worthless.

Now consider the same twelve documents chosen because they are every chemical on that participant's line. The identical effort produces a study in which the app can answer, and in which a failure to answer is a real finding about the product rather than an artefact of what happened to be curated.

R-22 already scopes the effort to one case study site. R-25 names the plant's actual process chemicals, and R-16 records that the chemical list is whatever the binder holds. What none of them supplies is a scoping unit smaller than the whole plant, and the whole plant was never affordable at D-58's rate.

ADR-0011 makes The Plant's binder the Corpus and access to it is still not secured. D-80 exists because that dependency has been treated as all-or-nothing when it is not: the list is separable from the documents, it is cheap to ask for, and it is worth more than any planning estimate this repository can produce on its own.

Decision

The Corpus is scoped to a Study Area rather than to a document count. The area's chemical list is requested first and sizes the Curation work; the documents for those chemicals are then requested and curated. Whatever number falls out is the number, and D-58's per-document cost and D-59's floor caveat apply to it unchanged.

Rejected options

• Keep twenty to forty documents — rejected. It is a plant-wide ambition costing ten to forty person-days at a rate ADR-0028 says understates the truth, against a binder nobody has yet been given, for a study that does not need plant-wide coverage to be valid.
• A smaller fixed number chosen by budget — eight, or twelve — rejected. It is affordable and it is the option that produces the worthless study described above. Choosing documents by what fits the budget optimises the one variable the study does not care about.
• Defer until binder access is secured — rejected for the reason ADR-0016 already gave: it leaves the schedule hostage to a third party who has promised nothing. D-80 is the smaller version of the same request and is a better test of access than continuing to wait for the larger one.
• Size the Corpus to the chemicals the source poster names — rejected. R-25's three named chemicals plus the parts-cleaning mixture are a proxy set (D-28) and a proof test (D-29), not a scope. They are what the pipeline is built against, not what the study is run against.

This ruling may not be re-decided

If a change contradicts this ADR: stop and raise it. Do not implement over it.

Specifically: do not quietly extend the Study Area to reach a rounder number of documents; do not curate a chemical from outside it because its SDS was easy to obtain, which is the availability bias ADR-0016 already warns about; and do not treat this ADR as a reduction in Curation cost. It changes which documents are curated, not what each one costs.

Consequences

What becomes easy. The "no coverage guarantee" gap becomes a bounded claim a report can defend: inside the Study Area coverage is complete and every miss is a finding. Recruitment for D-31 also sharpens, because the participants are the people who work in the area the Corpus covers.

What becomes hard. The Study Area must be chosen before Curation starts, and choosing it needs The Plant. This ADR converts a large blocked dependency into a small one; it does not remove it.

What this does not change. D-58's half-day-to-a-day per document and D-59's floor stand completely. If the area holds thirty-five chemicals, this ADR has bought no effort at all — but it will have said so before the work started rather than after twenty documents.

Coverage

Upstream	Landed in	Evidence	Note
R-16	D-79, D-80	the chemical list is the sizing artifact, requested separately from the documents	re-scoped from ADR-0011's plant-wide reading
R-22	D-79	one case study site narrowed to one area inside it
R-23	D-79	the four roles are recruited from the area the Corpus covers
R-25	D-79	the named chemicals are proxy and proof test, explicitly not the scope	ADR-0016 governs their use