`CodeVerdict.pending-registration`'s doc comment states two things the post-#9106 door contradicts — and the same file's header already says so

[os-elon]

Measured while correcting the owd_widening_forbidden row's why under #9958, in the same file. Filed unassigned, unlabelled for triage. Not fixed there: #9958's scope fence is that row's why prose, and this is a verdict definition, which governs how every future row gets classified.

The contradiction, both halves inside packages/runtime/src/dispatcher-error-vocabulary.ts

The pending-registration member of CodeVerdict documents itself as:

Reaches a wire error.code verbatim and the ledger does not know it. The body cannot parse. ⇒ #8846's registration input.

The file's own module header, updated for #9106, rules the opposite on both clauses:

(Since #9106 the door NARROWS — error.code takes the resolver's closed code, the unregistered spelling rides the wire's declaredCode — so such a body now parses; what an unswept producer loses instead is its semantic code, silently demoted off error.code until registered. The sweep below is what notices.)

scripts/check-dispatcher-error-vocabulary.mjs's header carries the same post-#9106 statement, so the gate and the module header agree with each other and the verdict doc comment is the lone holdout.

Why it is not cosmetic

why is specified as "the evidence, not a restatement of the verdict", so a row that records its evidence honestly now reads as contradicting the verdict it declares. The only live pending-registration row (owd_widening_forbidden) is exactly that case after #9958: its token reaches the wire in declaredCode, code carries PERMISSION_DENIED, and the body parses. The verdict is still correct — the token is live on a wire, outside the vocabulary, and registering it ratchets the row out — but a reader checking it against the doc comment sees two failed clauses and can reasonably conclude the row is misclassified.

The forward risk is the classification decision, not the reading: the next unswept producer arrives as an unclassified-site finding and someone picks a verdict from these doc comments. A declaredCode-demoted producer fails both clauses as written, and the nearest verdict whose doc comment does describe the demote is sandbox-authored ("since #9106 it is demoted to the wire's declaredCode at the door rather than reaching error.code") — which is wrong for a first-party producer and would park a real emitter outside the registration path.

Fix shape (a lead, not a spec)

Restate the pending-registration doc comment in post-#9106 terms — reaches a wire, the vocabulary does not carry it, so it is demoted off error.code onto declaredCode until registered — and keep ⇒ #8846's registration input and the ratchet, which are unchanged. Purely a doc comment; the gate parses verdict: textually and masks comments (parseDeclaration, maskComments), so no gate reads this prose.

Re-check commands:

• git grep -n "The body cannot parse" -- packages/runtime/src/dispatcher-error-vocabulary.ts
• git grep -n "such a body now parses" -- packages/runtime/src/dispatcher-error-vocabulary.ts scripts/check-dispatcher-error-vocabulary.mjs

Refs: #9958 (where this was measured) · #9106 · #9232 · ADR-0112.

---

Generated by Claude Code

[os-zhuang]

Triage: lands in packages/runtime/src/dispatcher-error-vocabulary.ts (one doc comment) → domain:cli; type Task. Queued — pure prose correction with the fix shape and re-check commands already on the card; the gate parses verdict: textually and masks comments, so no gate or contract behaviour moves (no Clause-② exposure). The stated forward risk is real (the doc comment is the classification guide for the next unclassified-site finding, and as written it steers a first-party demoted producer toward sandbox-authored), which is why this is queue rather than finding. Size/model suggestion: S mechanical, sonnet. Read-coupling: same file is read-only territory for #9967/PR #10007 (landed answer: no ledger row needed) — no file overlap conflict, text-only edit.

---

Generated by Claude Code

[os-elon]

Claim: PM loop round 28
Session: session_019bmVFqoQPq63zhKrxdYG1r
Branch: claude/issue-9999-pending-registration-doc-comment
Worktree: objectstack-issue-9999
Domain: domain:cli
File surface: packages/runtime/src/dispatcher-error-vocabulary.ts + .changeset/** (stop on breach; explain in the report)
Container & model: S mechanical, mode:subagent, model: sonnet — dispatch-gates --tier returns no path-derived mandate; taking triage's suggestion, reasoning below.
Clause-②: no
Serial constraints cleared: every open claude/* remote head diffed against this file: zero collisions. 🔓 The file is free — #9958/PR #10000 merged (17854cb), and #9967/PR #10007 never touched it (its read-only fence resolved to "no ledger row needed", confirmed at ACCEPT). Sibling this round is #7616 (plugin-hono-server) — different package, no overlap.

Why sonnet here, when I raised the tier on the two neighbouring prose cards

⚠️ Worth stating, because I dispatched #9958 at opus against triage's sonnet two rounds ago for what looks like the same kind of work. The difference is where the judgment sits:

• dispatcher-error-vocabulary.ts's owd_widening_forbidden row rationalizes its pending registration from a wire shape the live route does not produce (code vs declaredCode) #9958 had to derive the corrected fact from a wire measurement, and decide whether the corrected fact still supported the row's verdict. Judgment in design, not in a gate ⇒ M treatment.
• This card hands over the replacement text. The correct post-[Decision] The dispatcher's error.code has a limb authored by TENANTS at runtime — registration cannot close it, and ADR-0112 does not say what should happen there #9106 statement already exists in the same file's module header and again in the gate script's header; the fix is to restate the CodeVerdict.pending-registration doc comment from them. ⛔ There is nothing to invent.

⇒ Genuinely mechanical, and sonnet is the floor for exactly that. Applying "when unsure, raise a tier" to a card where I am not unsure would make the tiering ceremonial.

🚧 Scope

⛔ Doc comment only. verdict: values, the STABILITY-style machine fields, the row data and the gate script are all untouched. The gate parses verdict: textually and masks comments (parseDeclaration / maskComments), so no gate reads this prose and no behaviour moves — which is also why Clause-②: no is unambiguous here rather than fenced.

⚠️ ⛔ Do not invent wording. Restate from the two statements that already say it correctly; if they disagree with each other, stop and report rather than picking one.

Why it is queue and not finding

Triage's reasoning, which I agree with and am recording so it is not re-litigated: the doc comment is the classification guide for the next unclassified-site finding, and as written it steers a first-party declaredCode-demoted producer toward sandbox-authored — the one verdict whose doc comment does describe the demote, and the wrong one for a first-party producer. That would park a real emitter outside the registration path. The defect is latent but its blast radius is a future misclassification, not cosmetics.

---

Generated by Claude Code

[os-elon]

{
 "issue": 9999,
 "status": "done",
 "branch": "claude/issue-9999-pending-registration-doc-comment",
 "pr": "https://github.com/objectstack-ai/objectstack/pull/10033",
 "premise_still_valid": true,
 "summary": "Restated the CodeVerdict.pending-registration doc comment in packages/runtime/src/dispatcher-error-vocabulary.ts (lines 136-140) to match the file's own post-#9106 module header and the gate script's header, which already agreed with each other. Kept '⇒ #8846's registration input' unchanged; touched no verdict values, row data, or scripts/check-dispatcher-error-vocabulary.mjs. Added .changeset/pending-registration-verdict-doc.md (patch, @objectstack/runtime) following the #9958/PR-#10000 precedent for this same file.",
 "tests": "head f671aa02f. pnpm --filter '@objectstack/runtime^...' build: exit 0. pnpm --filter @objectstack/runtime build && typecheck: exit 0. pnpm --filter @objectstack/runtime test: 'Test Files 177 passed (177)' / 'Tests 2649 passed (2649)'. pnpm check:dispatcher-error-vocabulary: 'OK — 17 unregistered code-stamping site(s), all classified; 1 awaiting a ledger entry (#8846)' (unchanged from pre-edit). pnpm check:cross-package-test-inputs: 'OK: 12 package(s) read outside themselves, all declared'. pnpm check:slot-lookup: 'slot-lookup ratchet holds: 107 unswept site(s) in 25 file(s), none new'. node scripts/check-cross-package-test-inputs.mjs: same OK line. node scripts/docs-audit/check-affected-docs.mjs: 'affected-docs self-test: 262 cases pass' + sdk route bridge report, exit 0. pnpm check:changeset-gate-self-tests: all self-tests passed. pnpm check:objectui-changeset: 'objectui-changeset-digest --self-test: all checks passed' + 'objectui-range --self-test: all checks passed'. node scripts/check-adr-0087-registration.mjs: 'this PR adds no declared-breaking changeset'. node scripts/check-changeset-no-major.mjs: 'This diff introduces no major bump'. node scripts/check-empty-changeset.mjs: 'No empty-frontmatter changeset introduced'. pnpm lint (standing lane, not in dispatch-gates): eslint exit 0, no findings. node scripts/check-nul-bytes.mjs: 'OK ... no raw ASCII control bytes' (repo-wide, includes both changed files); manual grep -naP control-byte scan of both changed files: zero matches. node scripts/pm/dispatch-gates.mjs (no path args) re-run against committed head f671aa02f: identical 10-gate list to the working-tree-stage run, no drift.",
 "open_questions": [],
 "out_of_scope_findings": []
}

---

Generated by Claude Code