docs(spec): state both roles of aliases, not only the gap it fills

[os-bill]

Part of #17361

• Clause-②: no

Prose only. The diff adds no key to any published payload — every changed line in the two TypeScript files is inside a block comment (proof below).

What was wrong

strictObject's aliases was documented as a universal — "semantic near-misses edit distance cannot reach" — so an adopter holding a near-miss that distance does reach, and answers wrongly, reads the authority and concludes aliases is not their tool. That is the campaign's own finding-7 shape: this campaign's fix signposting the way into the failure mode it exists to kill.

The mechanism the sentence denies: the lookup is

aliases[aliasProbe(key)] ?? findClosestMatches(key, knownKeys, maxDistance, 1)[0]

so an alias is consulted before the distance fallback and wins outright. It therefore has two jobs, not one:

role	proving case	why the fallback alone is not enough
fill a gap distance cannot reach	visibleWhen to visible	re-measured: no candidate within budget
overrule a hit distance gets WRONG	hosts to network (plugin permissions)	budget max(2, floor(5/3)) = 2, hosts is exactly 2 from the declared hooks

Prerequisite readings — anchored by content on origin/main d07fc178b9, not by line and not from the card

Probe: whitespace flattened per file (a phrase can wrap across a line break), edit[- ]?distance (cannot|can not|can.t) reach case-insensitive, grep -o | wc -l over git ls-files (tracked only, so no built dist/ in the population). 19 occurrences in 17 files. Dark control (a fabricated token) 0; lit control strictObject( 383.

1. Both sentences still stand. Module docblock and StrictObjectOptions.aliases JSDoc, both in packages/spec/src/shared/strict-object.ts — 2 occurrences in that file.
2. The counter-example is still shipped and still reachable. hosts: 'network' is live in packages/spec/src/kernel/manifest.zod.ts. Distance computed independently, in a standalone Python Levenshtein against the four declared keys services / hooks / network / fs: hosts-hooks = 2, budget = 2 — inside. Controls from the same run: filesystem-fs = 8 (budget 3), paths-fs = 4 (budget 2), both outside, so the instrument returns both answers.
3. A fourth site the card's table does not list. The same universal is the hover text of the published factory's own option — StrictUnknownKeyErrorOptions.aliases in shared/suggestions.zod.ts — reached only because the site list was re-derived rather than trusted. Corrected here too.

The census — the count, not a hedge

A correction that says aliases are usually unreachable, while more counter-examples ship, is a new false sentence. So every shipped alias was classified rather than sampled: the strictObjectDeclarations() registry was forced over every module under packages/spec/src (the alias-integrity.test.ts walk), and for each entry the real findClosestMatches was run against that surface's own knownKeys at that key's own budget.

verdict	count
unreachable — distance returns nothing (the documented role)	1658
reachable, fallback agrees with the entry	211
reachable, fallback answers a DIFFERENT key the entry overrules	41
total entries / registered surfaces	1910 / 384

Controls on the instrument: lit — visibleWhen (the docblock's own proving case) classifies, as GAP; dark — a fabricated alias token returns 0 rows. Source-side cross-check on tracked files only: 255 aliases: occurrences in 64 files, all inside packages/spec; no package outside spec calls strictUnknownKeyError at all.

⇒ hosts is row 20 of 41, not the only one. The corrected prose therefore names the overruling role as a role, and carries the measured count with its date rather than a word like "usually".

What changed

• packages/spec/src/shared/strict-object.ts — module docblock: the aliases bullet now splits into the two roles, each with its proving case, plus the measured split and an explicit "do not read this as only for what distance cannot reach".
• packages/spec/src/shared/strict-object.ts — StrictObjectOptions.aliases JSDoc (the hover text): same correction, short form.
• packages/spec/src/shared/suggestions.zod.ts — StrictUnknownKeyErrorOptions.aliases JSDoc, the published factory's hover text (the re-derived fourth site).
• docs/audits/2026-07-unknown-key-strictness-ledger.md — the ledger's copy of the sentence. Its two cited examples (from to source, read to allowRead) were re-measured as still unreachable and kept.
• .changeset/strict-object-aliases-two-roles.md — patch on @objectstack/spec (a docblock ships in the published .d.ts).

⛔ No alias added or removed, no behaviour change, no key list, strictness or author-visible message touched.

Verification

• Comment-only proof: every +/- line in the two TypeScript files matches a block-comment continuation; the detector fires on a synthetic code line (lit control), and finds nothing in the real diff.
• pnpm --filter @objectstack/spec build — exit 0, 34/34 declaration files.
• pnpm --filter @objectstack/spec check:generated — all 15 generated artifacts up to date (so nothing generated mirrors these docblocks; nothing was hand-edited).
• pnpm --filter @objectstack/spec exec vitest run src/shared/ — 27 files, 567 tests, all pass.
• pnpm --filter @objectstack/spec typecheck — exit 0.
• eslint --no-inline-config over the whole repo — 6622 files, 0 errors (the union, not a narrowing).
• Derived gate families, scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack: 76 derived, 74 run green, 0 unrun; reconciled with --ran carrying each recorded exit code. The 2 not measured are check:dual-build-cjs-loads and check:lean-entry-closure — both exit 3, PREREQUISITE NOT MET (they read a whole-repo build). Exit 3 is a refusal, not a red; declared here and left to CI.
• Commit message grepped before the push: #, Part of, Refs, Fixes, Closes, Resolves → 0 each, with lit controls Co-authored-by → 1 and Claude-Session → 1. The card relation is declared once, here in the body.

Measurements above are on the final commit, d67cab8d2c.

验收备注

Noted, not filed — three per-site copies of the same sentence that my census measured false, each outside this card's declared file face and each a different judgement call:

• packages/objectql/src/validation/record-validator.ts and packages/spec/src/data/default-value-shape.ts both cite "edit distance cannot reach it (latitude to lat)". Measured: latitude has budget 2 and is 2 from the declared altitude, so the fallback reaches it and answers altitude — the alias is an overrule, and the cited example is the one that proves the opposite of what it is cited for. Successor: whoever next touches the location/address value-shape prose.
• packages/spec/src/data/driver/turso.zod.ts — "Semantic near-misses only — the spellings edit distance cannot reach" sits above a table containing uri to url, which the fallback reaches at distance 1 and answers identically. Harmless entry, false comment. Successor: whoever next touches the turso datasource schema.
• Checked and true, so nothing owed: the per-site claims in ui/chart.zod.ts (both tables), data/object.zod.ts and the hedged "most of these" in data/authoring-key-lint.ts.

⛔ Do not flip this out of draft, enqueue it, or arm auto-merge — the dispatching seat lands it.

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

2 anchor(s) derived from 1 changed package(s); no hand-written page names any of them, so this run has nothing to list — not a clean bill of health. This check sees only pages that NAME a derived anchor: one that documents this change in prose, or enumerates it in an authoring dialect, names none and stays invisible to it on every run.

What this run could not see

• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.

Coarse fallback — 135 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json f19dbcf43a4f332ad7fd21defe08230dce638028 → packageMentionDocs.

Which tree this was computed on

This run read content/docs from f82811f35f534a9c2b66da231f32a58c45ec9d58 — the merge of head d67cab8d2cb1d0c510a89a2254d513882bba5bd9 into base f19dbcf43a4f332ad7fd21defe08230dce638028, which is what actions/checkout gives a pull_request run. Not the PR head.

A worktree cut from an older main holds a different content/docs, so re-deriving there can legitimately return a different list — that is a different tree, not a wrong row. To answer on the same tree:

# while this PR is open — GitHub drops the merge commit once it closes
git fetch origin f82811f35f534a9c2b66da231f32a58c45ec9d58 && git checkout f82811f35f534a9c2b66da231f32a58c45ec9d58
# afterwards, rebuild it from the two parents, which stay fetchable
git fetch origin f19dbcf43a4f332ad7fd21defe08230dce638028 d67cab8d2cb1d0c510a89a2254d513882bba5bd9 && git checkout -B drift-repro f19dbcf43a4f332ad7fd21defe08230dce638028 && git merge --no-ff d67cab8d2cb1d0c510a89a2254d513882bba5bd9

node scripts/docs-audit/affected-docs.mjs --json f19dbcf43a4f332ad7fd21defe08230dce638028

⚠️ That checkout carried uncommitted changes, so the commit above does not fully identify what was read.