Detection verification

Run 2026-08-14 with rungs doctor against the four source repositories at the working-tree states surveyed in docs/research/. This is ADR-0004 §5’s acceptance criterion — a module whose [detect] block misclassifies any of the four is not finished — pulled forward from Phase 6 because detection was the riskiest unverified claim in the design.

Reproduce: node src/cli.ts doctor <path>

1. Result

A present · **P** n present with n artifacts adoptable as command gates or a prompt library · ¶ different paradigm, reported not acted on · — absent

Module	Rung	rift-forge	hexguard	hexguard-templates	axiom-mesh
`instructions`	0	A	A	A	A
`adr`	1	A	A	—	A
`backlog`	1	A	A	A	¶
`ci`	1	A	A	A	A
`findings`	1	A	—	—	A
`gates`	1	P 16	—	—	P 7
`session`	1	—	—	—	A
`audit`	2	—	A	A	—
`skills`	2	A	P 3	—	P 22
`specs`	2	A	—	A	¶
`workflows`	2	—	A	A	—
`design-sync`	3	A	¶	—	—
`release`	3	A	A	A	—
`doc-authority`	4	A	—	—	P 1
`concurrency`	5	A	—	—	—

The cells the whole product rests on came out right:

gates on rift-forge → P 16. It has more gate machinery than anyone and no ledger, and detection finds its validators adoptable as command gates without proposing to touch one.
gates on axiom-mesh → P 7. The PowerShell validators, in a repo with no root package.json, adopt without a Node dependency — which is what ADR-0002 was for.
doc-authority on axiom-mesh → P 1, matching its ownership registry as registry-without-gates: the hard part already done, the enforcement missing.
backlog on axiom-mesh → ¶, correctly refusing to act on a milestone system.
skills on axiom-mesh → P 22. Better than the catalog predicted, which said create. The 21 playbooks are adoptable now that the format is an open standard.
concurrency discriminates cleanly: 21 custom merge-driver declarations in rift-forge, zero in the other three.

2. Seven bugs, all found by running

None of these were visible from reading the manifests. Each is fixed and re-verified.

#	Bug	Consequence	Fix
1	`adopt_as` matches did not set state	*`gates` reported absent* on rift-forge** — the catalogue’s single headline case, invisible	An `adopt_as` match is ADR-0004 state 4: the function exists in a shape we can map
2	Paradigm evaluated even when paths matched	rift-forge’s pulled design mirror reported as both an external authority and an in-repo design system, on a `theme.ts` the pattern never meant to reach	Paradigm is consulted only when nothing else matched
3	`audit` matched `-audit-.md` anywhere	Hit `WI-108-…-audit-epic.md` and a launch-readiness epic — work items, not audits	Narrowed to a dedicated audits directory
4	`findings` inferred by frequency	Proposed `WI` — the backlog’s prefix — because a register cites work items more than it defines its own ids, 787 to 722	An `anchor` on the register’s own `NEXT-*` marker wins outright
5	`\b([A-Z]{1,6})-\d` matched inside compound ids	Proposed `US` on hexguard-templates from `FOUND-US-110` — exactly the false positive ADR-0004 was written about, arriving through inference instead of presence	Negative lookbehind `(?<![A-Za-z0-9-])`
6	Encodings share the id shape	Proposed `UTF` (32 matches, from `UTF-8`) as hexguard’s backlog prefix	`exclude_values` stoplist
7	`.gitattributes` as a `concurrency` signature	Reported axiom-mesh — 29 branches, no land protocol — as having concurrency tooling. A signature matching nearly every repo is not a signature	Scan the file’s content for a custom merge driver via new `marker_paths`

Two module signatures were also widened after the run: adr missed hexguard’s four decisions in docs/.ai/decisions/ (glob was docs/decisions/**), and specs lost an inference rule that proposed a prefix for a parameter whose value is a shape.

3. What the run does not establish

The A-versus-P distinction is not measured. Detection reports presence and adoptable artifacts; whether an existing system is a superset, a partial, or a near-miss is a per-artifact judgement, and per-artifact detection is not implemented. The catalogue’s expectation matrix predicts completeness this run cannot confirm or refute.
workflows reads absent on rift-forge and axiom-mesh, where the catalogue expected partial. Both have decision procedures — in skills and in 21 prompt playbooks — but neither has a plan- document tree, which is what the signature looks for. Left as-is: under-detection is the chosen bias, and widening this one would start matching any directory of markdown.
Nothing was installed. add, render and check do not exist yet, so the claim that adopting rift-forge’s gates yields a ledger without touching its files remains a design intent.
Detection says nothing about quality, only that files are where a module’s files would be.

4. Verdict against the acceptance criterion

Met for presence and paradigm; not yet testable for completeness. Every paradigm case is correctly refused, every adoptable case is correctly surfaced, and no module claims a repo it does not have. The A/P refinement waits on per-artifact detection.