ADR-0014: poster.pdf is the authoritative source; the first poster is history

Status: Accepted Date: 2026-09-02

Decisions

D-20 Where the two posters disagree, poster.pdf wins

chemical-safety-assisant-docs.pdf is retained as history, not deleted. Its findings keep their ids and remain accountable downstream; supersession is recorded per finding in docs/knowledge/research/poster-v2.md.

D-21 The scope is one case study — an automotive parts manufacturing plant — with four named user roles

Roles: the statutory workplace Safety Officer (จป.), supervisor, chemical store officer, and workers who handle chemicals. Generalisation to other plant types is future work, achieved by swapping the corpus rather than changing the product.

D-22 Per-role behaviour is out of prototype scope

One interface serves all four roles. The roles inform what content matters and who is interviewed for objective 2; they do not fork the UI.

Context

R-21, R-28, R-29, R-30 and R-34 each confirm a decision that was previously reached by inference: the subtitle says ต้นแบบ (prototype) confirming D-01; offline is a stated differentiator confirming D-06; the scan caption reads "place the label or QR code in the frame" confirming D-05; First Aid is a top-level tile confirming D-10; and the QR block is absent from the revision, supporting D-13. Recording this matters — a decision resting on the source is stronger than the same decision resting on an argument, and a later reader should not have to re-derive it.

R-22 and R-23 supersede R-03's four venues. R-24 supersedes R-04, and in doing so deletes the objective ADR-0001 cited — "study the benefits of using AI" is no longer in the source. ADR-0001's conclusion survives on R-21's direct evidence, which is stronger than the reasoning it replaces. R-33 supersedes R-19's benefit list; both remain unquantified.

R-31 records that screen 7 is still missing in the revision. Two independent revisions carry the same hole, while First Aid now appears as its own tile — so D-10's reconstruction looks right, and the absence looks like a layout artifact rather than a scoping signal.

Rejected options

• Delete the first poster and its findings — rejected: the R-ids are referenced by six accepted ADRs, and deleting them would break the traceability chain the repo exists to maintain. Superseded is not the same as wrong.
• Per-role interfaces — rejected for prototype scope. A จป. with statutory duties genuinely needs different things from a line worker, but building three interfaces before one is validated inverts the order of learning.

This ruling may not be re-decided

If a change contradicts this ADR: stop and raise it. Do not implement over it. Do not delete or renumber superseded findings.

Coverage

Upstream	Landed in	Evidence	Note
R-21	D-27	subtitle ต้นแบบ confirms D-01 against the source
R-23	D-21, D-22	four roles adopted for content and interviews, not for UI forking
R-24	D-20	supersedes R-04; ADR-0001's cited objective is gone, conclusion stands on R-21
R-28	D-27	offline confirmed as a stated differentiator, D-06 unchanged
R-30	D-27	First Aid tile confirms D-10
R-31	D-27	screen 7 gap reproduced in both revisions; treated as a layout artifact
R-33	D-30	expected results superseded; measurement deferred to a real baseline
R-34	D-27	QR absent from the revision, supporting D-13