Filing unassigned; observation-class. Found while writing the ADR-0001 tombstone (#7866 / PR #7960), which documents the discrepancy for a reader arriving at the number. This card is about the citing record's own wording, which that PR deliberately did not touch.
The observation
docs/adr/0002-environment-database-isolation.md:122, under Alternatives Considered:
- One global DB + tenant column. Was never on the table — already discarded in v3.4's ADR-0001.
That reference does not resolve to the record which actually held the number. Measured against full history (clone unshallowed — a shallow clone silently answers "never existed"):
docs/adr/0001-metadata-service-architecture.md — added 2026-02-10 (908d95c82), deleted 2026-02-11 (9da8e3e72). It decided how the metadataservice is registered (ObjectQL vs MetadataPlugin precedence) and says nothing about tenancy or database topology.git log --all --diff-filter=AD -- '*0001-*' returns exactly those two commits, on any path, on any branch. No other file ever claimed the number.
So "v3.4's ADR-0001" is a reference to a pre-registry document that has never existed in this repository. Today's ADR-0002 is itself dated 2026-04-19 and supersedes the v3.4/v4.0 per-organization database model, which is consistent with the reference being to a v3.4-era artifact that predates docs/adr/.
Worth noting the same mistaken attribution had propagated into tooling: the UNRESOLVED_ADR_CITATIONS entry in scripts/check-adr-anchors.mjs justified itself as "cited as history by ADR-0002". That half is corrected in PR #7960 (the entry is removed and the tombstone dissects the discrepancy).
Why this is not urgent
After PR #7960 the citation resolves — docs/adr/0001-withdrawn-metadata-service-architecture.md exists and has a "What ADR-0002 actually cites" section written for exactly this reader. No gate is red and no navigation dead-ends. The residual cost is that ADR-0002's sentence still asserts, in its own voice, a provenance that is not this repository's.
Why it was not fixed in PR #7960
ADR-0002 is an accepted record. Editing its prose is a decision about the decision log, not a rider on a tooling change, and docs/adr/ is maintainer-merge territory with no size exemption. Left for a maintainer call.
Possible dispositions
- Leave it. The tombstone already carries the explanation. Zero cost, mild residual inaccuracy in an accepted record.
- Qualify the sentence — e.g. name it as a pre-registry v3.4 document rather than a bare
ADR-0001, so it stops reading as a pointer into docs/adr/. Smallest edit that makes the record self-consistent. - Drop the citation, keeping the rejection ("was never on the table") without the provenance claim.
No recommendation offered — this is the maintainer's record to word.
Filing unassigned; observation-class. Found while writing the ADR-0001 tombstone (#7866 / PR #7960), which documents the discrepancy for a reader arriving at the number. This card is about the citing record's own wording, which that PR deliberately did not touch.
The observation
docs/adr/0002-environment-database-isolation.md:122, under Alternatives Considered:That reference does not resolve to the record which actually held the number. Measured against full history (clone unshallowed — a shallow clone silently answers "never existed"):
docs/adr/0001-metadata-service-architecture.md— added 2026-02-10 (908d95c82), deleted 2026-02-11 (9da8e3e72). It decided how themetadataservice is registered (ObjectQL vs MetadataPlugin precedence) and says nothing about tenancy or database topology.git log --all --diff-filter=AD -- '*0001-*'returns exactly those two commits, on any path, on any branch. No other file ever claimed the number.So "v3.4's ADR-0001" is a reference to a pre-registry document that has never existed in this repository. Today's ADR-0002 is itself dated 2026-04-19 and supersedes the v3.4/v4.0 per-organization database model, which is consistent with the reference being to a v3.4-era artifact that predates
docs/adr/.Worth noting the same mistaken attribution had propagated into tooling: the
UNRESOLVED_ADR_CITATIONSentry inscripts/check-adr-anchors.mjsjustified itself as "cited as history by ADR-0002". That half is corrected in PR #7960 (the entry is removed and the tombstone dissects the discrepancy).Why this is not urgent
After PR #7960 the citation resolves —
docs/adr/0001-withdrawn-metadata-service-architecture.mdexists and has a "What ADR-0002 actually cites" section written for exactly this reader. No gate is red and no navigation dead-ends. The residual cost is that ADR-0002's sentence still asserts, in its own voice, a provenance that is not this repository's.Why it was not fixed in PR #7960
ADR-0002 is an accepted record. Editing its prose is a decision about the decision log, not a rider on a tooling change, and
docs/adr/is maintainer-merge territory with no size exemption. Left for a maintainer call.Possible dispositions
ADR-0001, so it stops reading as a pointer intodocs/adr/. Smallest edit that makes the record self-consistent.No recommendation offered — this is the maintainer's record to word.