ADR-0043 — ADR-number and standards-rule collision adjudication

ADRsUpdated 2026-05-16 12:00 EDT6 min readEdit on GitHub ↗

ADR-0043 — ADR-number and standards-rule collision adjudication #

Status: Accepted
Date: 2026-05-16
Deciders: Natan
Source: 2026-05-16 documentation drift sync (see

CHANGELOG.md ↗ [Unreleased] §"documentation drift sync"). The drift audit surfaced four reservation collisions across in-flight PRDs that — until this ADR — were tracked only as side-by-side rows in docs/adr/README.md §"Reserved / planned" and docs/prds/00-tracker.md.

Context #

Four ADR numbers and one code-standards.md rule number were reserved by two different PRDs each. None of the reserved slots had files yet, so the conflicts were latent — but two of the PRDs (PRD-13 and PRD-16) are scheduled for the next development wave, and the first one to author its ADR would have silently forced the second to renumber post-hoc, breaking back-references.

Collisions surfaced by the drift sync:

Slot	Claimant A	Claimant B
ADR-0030	PRD-13 §13.1.1 — "Import-direction rules: `lib ← components ← systems`"	PRD-16 §16.1.1 — "Component-type registry: closed enum in code"
ADR-0033	PRD-09 §09.1.2 — "LLM outputs evaluated against external rubric, not fixed strings"	PRD-16 §16.1.2 — "Secret reference coordinate format"
ADR-0041	PRD-23 §23.2.1 — "`sqa_runs` vs `sqa_sweep_summary` table split"	PRD-22 §22.1.1 — "N-claim type shape: parallel `NegativeClaim<I,E>` vs reuse `Claim<O,E>`"
Rule 47 (`code-standards.md`)	PRD-13 §13.1.2 — "Concurrency code lives in named modules" (warn)	PRD-16 §16.2.1 — "Probes accept config, never read env directly" (fail)

This ADR adjudicates the four collisions and pins a procedural rule so future PRDs can detect and resolve reservation conflicts at scoping time rather than at authoring time.

Decision #

Procedural rule #

First-merged keeps the number. When two PRDs reserve the same ADR or code-standards.md rule number, the PRD whose authoring sub-item lands first (commit on master) keeps the slot. The later PRD renumbers to the next free slot at the bottom of the respective sequence, updates its body and tracker row in the same commit, and adds a "Reserved / planned" entry pointing at the new slot.

Tie-breakers (in order):

Master synthesis ship order. Where

docs/research/2026-05-09-master-synthesis.md pins a sequence (e.g. PRD-13 before PRD-10 at lines 371–395), the earlier-shipped PRD wins by default.

PRD scoping date. The earlier createdAt in the PRD's

frontmatter wins.

Smaller surface. The PRD whose ADR has fewer sub-items

referencing it wins (less downstream renumbering).

This is recorded as Rule 44 in docs/concepts/code-standards.md ↗ so PRD reviewers see the constraint at scoping time. (Rule 44 is not the same kind of rule as 1–43 / 46 — it constrains PRD authoring, not code. The rule still belongs in code-standards.md because it is part of the "how we write things down" discipline covered by ADR-0018.)

Concrete adjudication for the four open collisions #

Decisions follow the procedural rule above. Tie-breakers cited in parentheses.

Slot	Winner	Loser → new slot
ADR-0030	PRD-13 §13.1.1 (master-synthesis order; PRD-13 lands first per Wave A)	PRD-16 §16.1.1 → ADR-0044
ADR-0033	PRD-09 §09.1.2 (PRD scoping date 2026-05-09 vs PRD-16 2026-05-11)	PRD-16 §16.1.2 → ADR-0045
ADR-0041	PRD-22 §22.1.1 (PRD-23 §23.2.1 already shipped without authoring an ADR — the table-split decision is documented inline at `src/runners/sweep/sink.ts` and PRD-23 §23.2 Open Question 1 — so PRD-23's reservation is released)	n/a — PRD-23 releases; PRD-22 keeps ADR-0041
Rule 47	PRD-13 §13.1.2 (master-synthesis order; smaller surface — one check vs three)	PRD-16 §16.2.1 → Rule 48, §16.2.2 → Rule 49, §16.2.3 → Rule 50

After this adjudication, the post-state reservation table is:

#	Reserved for	Reserved by
0030	Import-direction rules (`lib ← components ← systems`)	PRD-13 §13.1.1
0033	LLM outputs evaluated against external rubric	PRD-09 §09.1.2
0035	System binding: roles, not system names	PRD-16 §16.1.3
0036	Environment as a first-class scope, not a tag	PRD-16 §16.1.4
0038	Route body-undelivered to unresolved	PRD-24 — referenced as Accepted, file backfill pending
0041	N-claim type shape (parallel `NegativeClaim<I,E>` vs reuse `Claim<O,E>`)	PRD-22 §22.1.1
0042	L4 transport beyond `fetch()` (Bright Data / Playwright)	PRD-24 anchors §24
0044	Component-type registry: closed enum in code	PRD-16 §16.1.1 (renumbered from 0030 by this ADR)
0045	Secret reference coordinate format	PRD-16 §16.1.2 (renumbered from 0033 by this ADR)

What this ADR is, what it is not #

Is: a one-time adjudication of four specific live collisions and a procedural rule to prevent future ones.

Is not: a renaming of an already-merged ADR. Every winner above holds an unmerged reservation; the renumbering is cheap because no ADR file or downstream code reference exists yet.

Consequences #

Positive.

PRD-09, PRD-13, PRD-16 can author their ADRs in parallel without

number conflicts.

Future PRD scoping has a rule to follow at reservation time.
docs/adr/README.md becomes the canonical reservation registry

(was already, but now with adjudication codified).

Negative.

PRD-16's PRD body and tracker rows need editing in the same

commit as this ADR. Touches 4 ADR-number references and 3 rule- number references in PRD-16. Mechanical, low risk.

PRD-23 needs a small edit removing its ADR-0041 reservation

(it never authored it; that reservation was speculative per PRD-23 §23.2 Open Question 1).

Renumbering rule favors landing-order over scoping-order in step
1. This is a deliberate choice — landing order is the

observable fact; scoping order is intent. A PRD that has been scoped for months but hasn't shipped should not block one actively shipping.

Falsifiability #

This ADR is falsified — and supersedable — if any of the following becomes true:

Procedural rule produces a worse outcome twice. If, in two

distinct future collisions, "first-merged keeps the slot" forces a much larger PRD to renumber more downstream references than its smaller competitor, the rule's heuristic is wrong. The tie-breakers in §"Procedural rule" exist to mitigate this, but if they aren't enough, the rule should flip to "fewer downstream refs wins" with landing-order as the tie-breaker.

A merged ADR's number ever needs to change. This would mean

the "first-merged keeps" rule isn't load-bearing — at which point the convention should add a numbering-rewrite tool and accept that ADR numbers are not stable.

**The reservation table at docs/adr/README.md is wrong twice

in a row** (someone authored an ADR with a number that was reserved). Means the table is not getting read at scoping time. Either tooling (CI check on PR adding an ADR file) or convention change.