ADR-0028: Curation costs roughly a day per document, and the measured figure is a floor

Status: Accepted — satisfies D-32 Date: 2026-09-02

Decisions

D-58 Stage 2 and stage 3 are sized on half a day to a full day of human effort per document

Not an hour. The dominant stage is review and reconciliation, not parsing and not translation drafting.

D-59 The measured figure is treated as a floor, not an estimate

The proxy corpus systematically understates what The Plant's binder will cost.

Context

R-35 answered D-32's question: roughly half a day to a full day per document, with review and reconciliation dominating at the order of 90-180 minutes. R-36 found parsing is not the cost — all three spike documents were born-digital with zero scanned pages and all 16 GHS sections present and correctly numbered.

R-45 records that the proxy sample understates the real cost: it was chosen for availability, and The Plant's binder will contain scanned sheets, older formats and formulated products whose documents are less tidy than a major manufacturer's Acetone SDS. D-56's page-image verification and D-44's cross-document reconciliation both land inside the dominant stage and were not separately budgeted in the spike.

Rejected options

• Treat the measured band as the planning figure — rejected: R-45 states plainly that the sample is not representative, and planning to a floor as though it were a mean guarantees a slip.
• Spike further before sizing — rejected as measuring precision the plan will not use; the hour-versus-day distinction was the one that mattered and it has been answered.

This ruling may not be re-decided

If a change contradicts this ADR: stop and raise it. Do not implement over it. Do not quote the spike's number as an expected cost without R-45's caveat attached.

Consequences

A corpus of 20-40 documents is weeks of human effort, not days. 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.

Coverage

Upstream	Landed in	Evidence	Note
R-35	D-58	the measured band and dominant stage
R-45	D-59	the proxy's systematic understatement