[finding] ADR-0082's chain diagram labels the near-miss prop typo "hard" — decision 5 of the same ADR, and `os validate` itself, both say it is a non-fatal advisory

[os-steve]

Found while repairing ADR-0082 line 99 for #10808 (PR #11912). That card's scope was deliberately narrowed by the PM to line 99 only, with adjacent citations to be filed rather than fixed — so this is filed, not fixed.

The site

docs/adr/0082-react-component-contract-governance.md, decision 6's chain diagram (the last box, four lines above the line #10808 repairs):

 prop gate (os validate — decision 5)
 (hard: missing-required / typo)

The label puts missing-required and typo on the same side of one word, hard.

Why it is wrong

The same ADR says otherwise, fifteen lines earlier. Decision 5's own body splits the two by severity and is correct:

• missing a required binding (e.g. ObjectForm with no objectName) → error (fails os build).
• a near-miss of a known prop (edit distance ≤ 2, e.g. onSucces → onSuccess) → warning.

And the CLI agrees with decision 5, not with the diagram. Measured on origin/main at fd50e59e7, CLI and the showcase app's dependency closure built in a worktree, mistakes injected into examples/app-showcase/src/ui/pages/renewals-pipeline.page.ts under a trap restore (the same runs written up in PR #11912, whose report carries the full output):

injected	exit	verdict line
missing required objectName	1	✗ Author-time rules failed (1 issue)
the onSucces typo alone	0	✓ Validation passed plus one ⚠ advisory

A typo alone exits 0. Nothing about it is hard.

Why it is worth an edit

The diagram is the compressed form a reader takes away — it is what gets quoted into a skill, a doc page or a PR description, and it is the half of decision 6 a reader reaches first. Read as it stands, it advertises a gate that fails a build on a near-miss prop name, which the platform does not deliver (declared ≠ enforced, Prime Directive #10 — pointed the other way: the doc claims more enforcement than exists).

Note the near-miss tolerance is a deliberate, load-bearing design choice, not an accident: decision 5 records that arbitrary unknown props are not flagged at all because the contract's data props are a curated subset. A diagram that flattens the split erases the reasoning.

Suggested fix

One line in the diagram box, e.g. (missing-required: error / near-miss prop: advisory). Governed file (docs/adr/**), human-merge-only, so the edit is a draft PR with review requested — but no pre-ruling looks necessary here either: decision 5 already records the correct severities, so this is a factual repair of an illustration, not a decision change.

---

Generated by Claude Code

[claude]

Triage: lands in docs/adr/0082-react-component-contract-governance.md, decision-6 chain diagram (line 96); domain:devx (docs quality). → pm:queue, Task — premise re-verified on origin/main: line 96 still reads (hard: missing-required / typo) while decision 5 of the same ADR and the measured CLI both make the near-miss typo a non-fatal advisory. Governed face (docs/adr/**): the PR stays draft, human merge only — ACCEPT ends at request-review/assign per seat discipline. Size/model suggestion: S mechanical, sonnet floor (cite the dispatch-gates tier output at dispatch time).

---

Generated by Claude Code

[os-steve]

Claim: PM dispatch (devx seat)

• Session: session_015ahemw8RcTgqtxrj15PEZx
• Branch: claude/issue-11914-adr-0082-chain-diagram-severity

This card came out of #10808's dev as an out-of-scope finding — it sits four lines above the line that card repaired, and was correctly filed rather than fixed, because my scope on #10808 was line 99 only. That fence was right and this is the follow-through.

⛔ GOVERNED SURFACE

docs/adr/**. The PR stays draft: ⛔ never flipped ready, never armed, never queued. ACCEPT ends at request-review from os-zhuang; the hand-merge is the review record.

The defect, and why it is not cosmetic

ADR-0082's decision-6 chain diagram (line 96) labels the prop gate (hard: missing-required / typo) — flattening the severity split that decision 5 of the same ADR spells out fifteen lines earlier, and that the CLI measurably keeps.

That split was just re-measured on #10808's PR (#11912), on current main, with the CLI built and both mistakes injected: a missing required objectName exits 1; an onSucces typo alone exits 0 with a ⚠ advisory and the advisory count moving 31 → 32. So the diagram is wrong about behaviour that has a live measurement behind it, and it contradicts its own document.

⚠️ A reader who consults the diagram rather than decision 5 will believe a near-miss prop name is fatal. #11912 repaired the prose one paragraph below and deliberately left this; the two should agree.

Tier: sonnet — triage suggested it and I agree. Derived live, no path-derived mandate on docs/adr/**, so sonnet is the floor and this is a deliberate floor call: the correct behaviour is already measured and quoted on #11912, so what remains is making one diagram label say what decision 5 and the CLI both already say.

---

Generated by Claude Code

[os-steve]

os-dev-report

{
 "issue": 11914,
 "status": "done",
 "branch": "claude/issue-11914-adr-0082-chain-diagram-severity",
 "pr": "https://github.com/objectstack-ai/objectstack/pull/11981",
 "premise_still_valid": true,
 "summary": "ADR-0082's decision-6 chain diagram at docs/adr/0082-react-component-contract-governance.md:96 read `(hard: missing-required / typo)`, flattening the error/warning split decision 5 records fifteen lines earlier (missing required binding = error, near-miss prop typo = warning). Reworded the label to `(missing-required → error / typo → warning)`, matching decision 5's own vocabulary and arrow notation. Only that one line changed; decision 5's prose and the adjacent line 99 (in-flight repair under PR #11912, still unmerged) were left untouched.",
 "tests": "Verified on current origin/main (ef786a51b) before editing: line 96 still read `(hard: missing-required / typo)` and line 99 still carried the un-repaired old-path/old-wording text, confirming PR #11912's repair had not landed — premise valid, proceeded. `node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack` derived 5 local gate families for the diff (docs/adr/0082-react-component-contract-governance.md), re-derived identically at final commit c93e1844d. Built dependency closures first (pnpm --filter '@objectstack/lint^...' build, pnpm --filter @objectstack/lint build) after hitting PREREQUISITE-NOT-MET on a fresh worktree's missing dist/. All 5 gates run under scripts/pm/os-verify-lock.sh, each read via its own EXIT capture (not piped) — verdict lines quoted, all command-exit 0: `check-adr-anchors: OK (52 anchored file(s) ... 28016 citation(s) across 3518 file(s) resolve)`; `✓ doc authoring guard: 389 files clean — no bare metadata literals`; `✓ check:doc-formula-expressions: 22 record-scoped formula example(s) across 421 files / 1448 TS blocks judged clean`; `✓ check-governed-merges --self-test: 129 assertions`; `✅ check-adr-links: 551 relative link destination(s) under docs/adr/ resolve`. Also ran check:nul-bytes (not path-derived, run anyway since a docs file was hand-edited): `check-nul-bytes: OK (scanned 6658 text file(s) ... no raw ASCII control bytes)`. git diff confirms exactly one line changed in the file; no ablation applicable (docs-only single-line label edit).",
 "open_questions": [],
 "out_of_scope_findings": []
}

Generated by Claude Code

---

Generated by Claude Code

[yinlianghui]

ACCEPT — PR #11981 · reviewed against GitHub, not against the report

Reviewer of record: devx seat, session_01UjM2ia8Av1v5NqfqQEQmC6 (seat #6023). This card was dispatched by the previous session and its report sat uncollected; this is the collection.

Checklist result

criterion	reading
PR shape	draft ✓ · base main ✓ · first line Fixes #11914 ✓
scope	1 file, +1/−1 — docs/adr/0082-react-component-contract-governance.md only. Verified via get_files, not the report's self-description.
the edit itself	patch read directly: (hard: missing-required / typo) → (missing-required → error / typo → warning). Decision 5's own vocabulary is "error" / "warning", so the new label lands the reader on the same distinction rather than a third phrasing.
changeset	skip-changeset, correct — docs-only, publishes nothing.
CI	26 raw runs → grouped by name, latest per name: 0 failure, 0 cancelled, 0 in_progress. Both governance checks (No other open PR may claim the same single-writer path, Part-of PR must not also close its card) green.

⚠️ CI counted the way this seat's own instrument rule requires — raw aggregation over check_runs double-counts superseded attempts and manufactures false reds.

⛔ Governed surface — terminal handling, not a landing

docs/adr/** is governed. The four-part ending applies and this seat will not merge, flip ready, queue, or arm auto-merge:

1. Review verdict recorded here. ✓
2. PR stays draft, visibly pending. ✓
3. Review already requested from os-zhuang by the dispatching session — verified present in requested_reviewers, so no second request is needed. ✓
4. Listed under "awaiting a human merge" in the round report. ✓

⭐ Serial constraint the maintainer needs before merging — these two PRs collide

PR #11912 (#10808) edits the same file three lines away, at line 99, and is also an open draft awaiting the same reviewer.

• docs(adr): un-flatten ADR-0082's decision-6 chain diagram severity label #11981 changes line 96; its diff hunk spans lines 93–99 and carries line 99 as context.
• docs(adr): repair ADR-0082's golden-page citation — the moved path, and an os validate output that never renders #11912 changes line 99; its hunk spans roughly 96–102.

The two hunks overlap, so whichever merges second will conflict and needs main merged into it before it can go in. They are not independent despite touching different lines.

⇒ Suggested order: #11912 first (it carries the measured CLI evidence that makes both edits correct), then #11981 rebased/merged onto the result. Either order works; what does not work is merging both without touching the second one afterwards. Recorded on both PRs and in the seat's hot-file serial queue.

No rework requested — the change is correct as it stands.

---

Generated by Claude Code