ADR-0017: Functional acceptance now; performance targets only against a measured baseline

Status: Accepted Date: 2026-09-02

Decisions

D-30 No performance target is set until the current state has been measured

Stage 2 acceptance criteria are functional and checkable ("the record displays PPE with its SDS Section reference"). Performance targets are added after the needs study measures how long a lookup takes today.

D-31 Baseline measurement is part of the needs study

The surviving study objective (R-24, objective 2) is fieldwork with the four roles. Timing how long a Safety Officer currently takes to answer "which glove for this drum" is part of it.

D-32 Curation cost is measured by a two-document spike before /to-spec

Two SDS from the proxy corpus, curated end to end — parse, structure, translate, review — and timed. Two is enough to distinguish an hour per document from a day per document, which is the only distinction stage 2 sizing needs.

Context

R-19 and R-33 both list expected benefits and neither quantifies anything. Neither poster contains a single number. Stage 2 needs acceptance criteria, and "improves safety" is not one.

The obvious fix — invent targets now — is worse than having none: the final report would compare the result against a number somebody made up. R-32 says the current state is scattered and hard to search, and nobody has timed that either, so there is nothing to improve on yet.

D-11 separately made human translation review the corpus bottleneck, and nobody has timed a single document, so "20-40 SDS" could be two days or three weeks.

Rejected options

• Set targets now with the owner — rejected: hard criteria immediately, but measured against guesses.
• Qualitative evaluation only — rejected as leaving stage 2 with no objective bar and making "done" a matter of opinion.
• Skip evaluation — rejected: it leaves the surviving study objective largely unanswered.
• Spike the full corpus rather than two documents — rejected as front-loading the most tedious work before anyone has agreed what is being built.

This ruling may not be re-decided

If a change contradicts this ADR: stop and raise it. Do not implement over it. Do not put an invented number in the spec because a ticket wanted a threshold.

Correction to an earlier rationale

ADR-0009's Consequences claims curation cost "is itself a result for study objective R-04.2". R-24 deleted that objective. The measurement is still worth doing — for sizing, per D-32 — but not as a research finding. ADR-0009's ruling stands; only that sentence's justification is void. See ADR-0018.

Coverage

Upstream	Landed in	Evidence	Note
R-19	D-30	unquantified benefits are why no target is set yet