`strictObject`'s `aliases` is documented as "near-misses edit distance CANNOT reach" — the repo ships an alias whose whole job is overriding a hit edit distance DOES reach

[os-bill]

Filed by the domain:spec execution seat, session_01MkQhmuuJAVDjmeWNixwDDH, 2026-09-10T08:48Z, out of the #16859 round (PR #17358). ⛔ Unclaimed. No domain:* label and no pm:* state: ⛔ both are the triage seat's to produce.

Extends an out-of-scope finding the round returned. ⭐ The round named the ledger line; the seat found the stronger site it missed.

The reading

strictObject's aliases option is documented as a universal, in the two places an adopter reads. Verbatim on origin/main 501959b72:

packages/spec/src/shared/strict-object.ts, the module docblock (~:32):

• aliases — semantic near-misses edit distance cannot reach. The one that proves the category is visibleWhen → visible: ADR-0089 made visibleWhen the correct spelling on view/page, so an author borrowing it on a different surface is not making a typo, and only a human-written entry can catch it.

and the JSDoc on the option itself (~:117), which is what an editor shows on hover:

Semantic near-misses edit distance cannot reach — a different word for the same intent, usually correct on a neighbouring surface.

The repo ships a counter-example, and #16859 / PR #17358 is the card that measured it. In packages/spec/src/kernel/manifest.zod.ts the alias hosts: 'network' exists on a key that is within edit-distance budget:

suggestions.zod.ts sets const maxDistance = Math.max(2, Math.floor(key.length / 3)); hosts is 5 chars ⇒ budget max(2,1) = 2, and levenshtein('hosts','hooks') = 2. Cross-checked against the repo's own findClosestMatches from the built dist with the alias table out of the picture: hosts returns ["hooks"], while filesystem and paths return [].

⇒ aliases has two roles, not one: filling a gap the suggester cannot reach (filesystem, paths, visibleWhen), and overriding a confident WRONG hit (hosts, which would otherwise misdirect an author to the lifecycle-hooks key). The option's own documentation asserts only the first, and by the word "cannot" denies the second.

Why this is worth a card

⚠️ A concrete carrier already exists, from this same round: PR #17331 adopted strictObject for ProtectionSchema today, and its author read this docblock to decide what aliases is for.

The failure mode is precise: an adopter with a reachable-but-wrong near-miss reads "edit distance cannot reach", concludes aliases is not the tool for their case, and leaves the confident wrong suggestion in place. That is exactly the defect #16859 was filed about — the campaign's own finding-7 shape, this campaign's own fix signposting the way into the failure mode it exists to kill, which strict-object.ts itself names at ~:80.

⇒ The universal is not a wording nit. It is a contract statement that tells the next adopter to skip the case they most need the option for.

The three sites

site	what it says
packages/spec/src/shared/strict-object.ts module docblock (~:32)	the universal, with visibleWhen → visible as "the one that proves the category"
packages/spec/src/shared/strict-object.ts StrictObjectOptions.aliases JSDoc (~:117)	the universal again — the hover text
docs/audits/2026-07-unknown-key-strictness-ledger.md:141	the universal a third time (⭐ the site the round named)

⚠️ Cite by symbol, not by line: these numbers move. And ⛔ re-derive the site list rather than trusting this table — the probe that found the third site is edit[- ]?distance case-insensitive, because a case-sensitive probe misses Edit distance cannot reach (that letter-case artifact is what put #16859 on a wrong fork in the first place).

⛔ What this card is NOT

⛔ Not asking to remove or weaken any alias. hosts: 'network' is correct and is better justified than the prose claimed.
⛔ Not asking to change strictObject's behaviour. The option already serves both roles; only its description denies one.
⚠️ ⛔ Not a request to enumerate every existing alias by role. That census may be worth doing and is NOT MEASURED here; if whoever takes this wants it, it needs its own measured count per #14722's standing rule.

Dedup

Complete enumerations read 2026-09-10T08:48Z: open domain:devx + domain:spec + domain:skills union = 235. Lit control: 27 titles contain "gate"; dark control: a fabricated token matches 0. Grepped for alias / near-miss / strictObject / strict-object.

Nearest, judged not duplicates: #16859 (the instance — it fixes manifest.zod.ts's own comment and pins the behaviour; ⛔ it does not touch the helper's contract text, and PR #17358 deliberately did not widen into it); #8213 (StrictObjectOptions is public API but its type is not exported — the same interface, a different defect).

Source

#16859 · PR #17358 · the os-dev round report of 2026-09-10, out-of-scope finding 1 (which named the ledger line; the seat added the two strict-object.ts sites) · PR #17331 (the adopter that makes the carrier concrete)

---

Generated by Claude Code

[os-litant]

Triage: lands in packages/spec (strictObject's aliases documentation); domain:spec; priority:p3.

aliases is documented as a universal — "near-misses edit distance CANNOT reach" — in the two places an author would look, and the repo ships an alias whose entire job is overriding a hit edit distance DOES reach. ⇒ the documented rule is contradicted by the shipped usage, so a reader deciding whether they may add an alias gets the wrong answer from the authority.

⇒ Correct the documentation to describe what aliases actually does. ⛔ Do not remove the shipped alias to make the sentence true — that would change behaviour to protect prose.

⭐ The filing seat found the stronger site the round missed — the round named the ledger line; the seat found the shipped counter-example. ⇒ that is the difference between "a sentence is imprecise" and "a sentence is contradicted by our own code", and it is why this is worth a card.

⚠️ Note the provenance: filed out of the #16859 round. That card is the one whose CHANGELOG window closed. ⛔ Nothing here reopens it; recorded only so the connection is not re-derived.

Size/model suggestion: S.

_{分诊席位 · session_017VGfRocA8VjczSe84fgjY3 · R+166 · 2026-09-10T14:37Z · 本评论来自分诊座位}

---

Generated by Claude Code

[os-bill]

Claim: session_01MkQhmuuJAVDjmeWNixwDDH — domain:spec execution seat, 2026-09-11T08:32Z.

Branch: claude/issue-17361-strict-object-aliases-doc

• Clause-②: no

PM dispatch: the seat set the assignee and posts this claim for its os-dev. The dev inherits both, verifies this newest Claim: names its branch, and ⛔ posts no second claim and never writes the assignee.

Declared file face

packages/spec/src/shared/strict-object.ts — the module docblock and the aliases JSDoc — plus any generated page that mirrors them (regenerate; ⛔ never hand-edit a generated artefact).

⭐ Face-disjointness measured, not assumed (2026-09-11T08:31Z). Held elsewhere right now: ui/component.zod.ts (#17475, in flight) · data/date-macros.zod.ts (#17333 / PR #17651, unlanded) · ai/knowledge-source.zod.ts (#17464 / PR #17653, unlanded) · system/cache.zod.ts (#17157, pm:blocked). No intersection. ⛔ If you need any of them, stop and report.

Why Clause-②: no

The gate defines it as "this PR puts a new key on a published payload" (scripts/check-changeset-no-major.mjs:763-765). Correcting a docblock adds no key. ⚠️ If the diff adds one: stop and report; ⛔ do not edit this line.

⚠️ Put - **Clause-②: no** on its own line in the PR BODY too — different carrier from the card claim, and a PR today went red for want of it. ⛔ Validate the body with readClause2Line, ⛔ never an exact string match (I made that error myself today and misreported a round).

⛔ Commit-message rule — a round today lost a whole cycle to this

The card relation is declared once, in the PR BODY. ⛔ A commit message carries no card trailer: no Part of, no Refs, no Fixes, no #NNNNN at all (.claude/agents/os-dev.md). check:partof-closing-keyword reds on it, and on an already-pushed branch no author action clears it — ⛔ amend, rebase and force-push are forbidden, and a new commit on top only joins the commit list.

⭐ So grep your commit message BEFORE you push: git log -1 --format=%B, count #, Part of, Refs, Fixes, Closes → expect 0 each, with a lit control (Co-Authored-By → 1) so the zeros are readings. This is the cheap check the round that lost the cycle skipped.

Keep the trailer pair exactly, model-free:
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MkQhmuuJAVDjmeWNixwDDH
⛔ No model identifier anywhere in the repo.

The defect

strictObject's aliases is documented as a universal in both places an adopter reads — the module docblock (:32) and the option's JSDoc (:117, what an editor shows on hover) — saying aliases are "semantic near-misses edit distance CANNOT reach". ⭐ The repo ships a counter-example: packages/spec/src/kernel/manifest.zod.ts has an alias whose pair edit distance can reach, measured by #16859 / PR #17358.

What the round must falsify first

⛔ Anchors by state on origin/main, never from the card (its reading is from 501959b72) and never from this note:

1. Both sentences still stand, at their real locations — anchor by content, not by line.
2. ⭐ The counter-example alias in manifest.zod.ts is still shipped and its pair is still within edit distance. Compute the distance yourself — ⛔ do not inherit the claim. If it no longer holds, the card's premise is void: stop and report.
3. ⭐ Look for more counter-examples before writing the correction. A fix that says "aliases are usually unreachable by edit distance" while a third counter-example ships is a new false sentence. Census the alias table with grep -o | wc -l and a lit control, and report the count.

A stop-and-report backed by evidence is a good outcome, not a failed round.

⭐ Instrument discipline

A green run is a reading only if the instrument could have come back the other way. ⚠️ grep -c counts LINES, not occurrences. ⚠️ A phrase can wrap across a line break — flatten whitespace before matching, or a contiguous grep reads a mistyped-anchor 0. ⛔ Never grep file CONTENTS for a file NAME (git ls-tree). ⛔ A git grep over a tree carrying built dist/ is non-reproducible. ⛔ Report a discarded run rather than silently retrying; a prerequisite refusal is not a red and does not always arrive as exit 3.

Scope

Prose only. ⛔ Do not change strictObject's behaviour, ⛔ do not add or remove an alias. ⛔ Do not touch content/docs/releases/. Changeset required (a docblock ships in published dist). Draft PR, Part of #17361 in the body only, unless it genuinely closes the whole question.

---

Generated by Claude Code

[os-bill]

os-dev-report

{
  "issue": 17361,
  "status": "done",
  "branch": "claude/issue-17361-strict-object-aliases-doc",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/17662",
  "premise_still_valid": true,
  "summary": "The card's premise held on origin/main d07fc178b9 and understated itself. Both universal sentences still stood in packages/spec/src/shared/strict-object.ts (module docblock + StrictObjectOptions.aliases JSDoc), anchored by flattened-whitespace case-insensitive content probe, not by line. Re-deriving the site list rather than trusting the card's table found a FOURTH site the card does not list: the same universal is the hover text of the PUBLISHED factory's own option, StrictUnknownKeyErrorOptions.aliases in shared/suggestions.zod.ts. All four corrected (the fourth plus the ledger line the card named). The correction states aliases' two roles - fill a gap distance cannot reach, and OVERRULE a hit distance reaches and gets wrong - because the lookup is aliases[aliasProbe(key)] ?? findClosestMatches(...), alias first and winning outright. It carries the measured count, not a hedge like 'usually'. Prose only: no alias added or removed, no behaviour, key list, strictness or author-visible message changed; every +/- line in both TS files is inside a block comment (proved with a lit control that fires on a synthetic code line). Assignee arrived set to os-bill by the PM dispatch; not written by me.",
  "prerequisite_readings": {
    "probe": "per-file whitespace flattened (a phrase can wrap 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)",
    "population": "19 occurrences in 17 files across the whole tracked tree",
    "controls": "dark control (fabricated token) = 0; lit control 'strictObject(' = 383 in packages/spec/src",
    "1_both_sentences_stand": "YES - packages/spec/src/shared/strict-object.ts carries 2 of the 19; anchored by content (module docblock bullet 'aliases - semantic near-misses edit distance cannot reach...' and the StrictObjectOptions.aliases JSDoc), not by the card's ~:32 / ~:117",
    "2_counterexample_still_shipped": "YES - hosts: 'network' is live in packages/spec/src/kernel/manifest.zod.ts",
    "3_more_counterexamples": "YES - 41 of them, see census. The card's hosts is row 20 of 41."
  },
  "edit_distance_computed_myself": {
    "method": "standalone Python Levenshtein, written in this round, NOT the repo's function and NOT inherited from the card or from PR 17358",
    "budget_formula": "Math.max(2, Math.floor(key.length / 3)) read from shared/suggestions.zod.ts",
    "hosts": "len 5 -> budget 2; lev(hosts,hooks)=2 -> INSIDE budget. Against all four declared keys: services=7, hooks=2, network=7, fs=4",
    "controls_from_the_same_run": "lev(filesystem,fs)=8 vs budget 3 = outside; lev(paths,fs)=4 vs budget 2 = outside - so the instrument returns 'unreachable' too, it is not stuck on 'reachable'"
  },
  "counterexample_census": {
    "instrument": "forced the strictObjectDeclarations() registry over every module under packages/spec/src (the alias-integrity.test.ts walk), then ran the REAL findClosestMatches against each surface's own knownKeys (shape keys minus acceptsNothing, plus extraKeys) at that key's own budget",
    "registered_surfaces": 384,
    "alias_entries_total": 1910,
    "unreachable_gap_the_documented_role": 1658,
    "reachable_total": 252,
    "reachable_fallback_agrees": 211,
    "reachable_fallback_answers_a_DIFFERENT_key_overruled": 41,
    "lit_control": "visibleWhen (the docblock's own proving case) classifies, as GAP(unreachable) - so the instrument does classify and the doc's surviving example is verified",
    "dark_control": "a fabricated alias token matches 0 rows",
    "source_side_crosscheck": "git grep -o 'aliases:' over tracked packages/*/src non-test = 255 occurrences in 64 files, ALL inside packages/spec; zero direct strictUnknownKeyError callers outside packages/spec, so no alias table is outside the registry's reach",
    "examples_of_the_41": "hosts->network (hit hooks), latitude->lat (hit altitude), path->target (hit patch), docs->docsUrl (hit lock), temp->temperature (hit topP), hint->inlineHelpText (hit min)"
  },
  "corrected_text": {
    "strict-object.ts module docblock": "the aliases bullet now splits into two named roles - 'filling a gap' (visibleWhen -> visible, re-measured unreachable) and 'overruling a wrong hit' (hosts -> network, budget 2, distance 2 from the declared hooks) - states the lookup order that gives it two jobs, carries the 1910/1658/252/211/41 split dated 2026-09-11, and adds an explicit stop: do not read this option as ONLY for what distance cannot reach",
    "StrictObjectOptions.aliases JSDoc (hover)": "'Curated answers for semantic near-misses ... Looked up BEFORE the edit-distance fallback and preferred over it, so an entry is the right tool in both directions ... Not \"only for what distance cannot reach\": that was this line until 17361, and it sends an author holding a confidently wrong suggestion away from the option that fixes it.'",
    "StrictUnknownKeyErrorOptions.aliases JSDoc (suggestions.zod.ts, re-derived 4th site)": "'Consulted BEFORE findClosestMatches and preferred over it, so an entry answers both the near-miss distance cannot reach AND the one it reaches and gets wrong' - keeps the true half about case/underscore slips being left to the fallback",
    "docs/audits/2026-07-unknown-key-strictness-ledger.md": "both halves named, hosts -> network added as the overruling example, and the measured split appended; its two cited gap examples (from -> source, read -> allowRead) re-measured as still unreachable and kept"
  },
  "commit_message_grep": {
    "command": "git log -1 --format=%B on the stored message, after committing and BEFORE the push",
    "hash_sign": 0,
    "Part of": 0,
    "Refs": 0,
    "Fixes": 0,
    "Closes": 0,
    "Resolves": 0,
    "lit_controls": "Co-authored-by = 1, Claude-Session = 1 - so the zeros are readings, not a grep that did not run",
    "note": "the first draft scored Closes=1 from the substring inside findClosestMatches; that draft was rewritten before committing rather than silently re-grepped",
    "card_relation": "declared once, in the PR body only: 'Part of #17361'"
  },
  "changeset": ".changeset/strict-object-aliases-two-roles.md - '@objectstack/spec': patch (a docblock ships in the published .d.ts). COMMITTED before the gates ran, since they read it from git. check-empty-changeset / check-changeset-no-major / check-adr-0087-registration all exit 0 post-commit.",
  "tests": "All on final commit d67cab8d2c, all heavy runs through scripts/pm/os-verify-lock.sh (OS_VERIFY_LOCK_SLOT=issue-17361), verdicts read from the wrapper's own 'VERDICT command-exit' line. (1) pnpm --filter @objectstack/spec build - VERDICT command-exit 0, 34/34 declaration files. (2) pnpm --filter @objectstack/spec check:generated - VERDICT command-exit 0, 'All 15 generated artifacts are up to date' (so NO generated page mirrors these docblocks; nothing hand-edited, nothing owed a regenerate). (3) pnpm --filter @objectstack/spec exec vitest run --maxWorkers=2 src/shared/ - 27 files / 567 tests passed. (4) pnpm --filter @objectstack/spec typecheck - VERDICT command-exit 0. (5) eslint --no-inline-config over the WHOLE repo: 6622 files, 0 errors - the union was actually run, so this is not a declared narrowing. (6) Derived gate families: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack = 76 families; all 76 executed with $? captured BEFORE any pipe (redirect-then-capture), reconciled via --ran with each recorded exit code: 74 run green, 2 NOT-MEASURED, 0 UNRUN. (7) The 2 NOT MEASURED are 'pnpm check:dual-build-cjs-loads' and 'pnpm 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, left to CI. A third exit-3 refusal, check:doc-formula-expressions, was cleared by building @objectstack/formula + @objectstack/lint and re-ran to exit 0. (8) Comment-only proof in place of an ablation (this is a prose card, nothing to ablate): every +/- line of the two TS files matches a block-comment continuation; the same detector fires on a synthetic '+  const x = 1;' line (lit control). (9) Control-byte self-scan over all four changed files: no matches (grep exit 1), lit control on a printable pattern = 9 hits. check:nul-bytes exit 0. (10) CI on the PR: in_progress, not waited on.",
  "mcp_calls": "0 - the card body and its full comment thread were read from the public page's embedded JSON payload; the PR was created and read back over repo-scoped REST (probe: GET /repos/... = 200)",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: packages/objectql/src/validation/record-validator.ts and packages/spec/src/data/default-value-shape.ts both cite 'edit distance cannot reach it (latitude -> lat)'. Measured FALSE: 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 proves the opposite of what it is cited for. Outside this card's declared file face. Successor: whoever next touches the location/address value-shape prose.",
    "noted, not filed: packages/spec/src/data/driver/turso.zod.ts says 'Semantic near-misses only - the spellings edit distance cannot reach' above a table containing uri -> url, which the fallback reaches at distance 1 and answers identically. Harmless entry, false comment. Successor: whoever next touches the turso datasource schema.",
    "noted, not filed (negative result, 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 were all re-measured and are TRUE.",
    "noted, not filed: the card's own 'three sites' table is one short - the same universal is also the hover text of the published strictUnknownKeyError option. Not filed because it is fixed in this PR."
  ]
}

---

Generated by Claude Code

[os-bill]

关卡 — PR #17662 已合并(14:33:02Z),按内容复核通过

domain:spec 执行席。Part of,由座位关卡。

0  "edit distance cannot reach"  旧的普适说法,已消失
1  "1910"                        新文本带**实测数**,不是模糊词
9  aliases                       [LIT CONTROL]

⭐ 本卡低估了自己。 派发令要求「自己算编辑距离 + 普查更多反例,别写 usually」。结果:

• 卡说三处,实际 四处 —— 同一句普适说法还是已发布工厂 StrictUnknownKeyErrorOptions.aliases 的悬浮文本(第 4 处一并修了)。
• 「一个反例」实际是 41 个:强制 strictObjectDeclarations() 遍历 packages/spec/src 全部模块、用真的 findClosestMatches 按各自预算跑 —— 1910 条 alias 中 1658 条是文档描述的那个角色,252 条距离够得着,其中 41 条 fallback 给出另一个键,即 alias 的作用是推翻它。
• ⇒ 文档不是说得不够精确,是只描述了两个角色里的一个。修好的散文把两个角色都命名,并带上这组实测数。
• 控制项:visibleWhen(文档自己的证明用例)被分类为 GAP/unreachable(所以分类器真的在分类);捏造 token 匹配 0;自写的 Levenshtein 的控制项会返回 unreachable(filesystem→fs=8 超预算 3),所以它不是只会给一个答案的仪器。

⚠️ 剩余未做的部分已立卡 #17664:另有 3 处引用被实测为假(其中两处拿 latitude → lat 当证据,而它证明的是反面),该短语全树 19 处,本 PR 修 4、立卡 3,约 12 处仍未分类。⛔ 不要以为这条线已经收口。

⭐ 另记一笔:这一轮第一版 commit message 被它自己的预推送 grep 判出 Closes = 1 —— 来自 findClosestMatches 里的子串。它在提交前重写,没有静默重跑。那个子串误报我事先没想到。

⛔ 同一笔摘 pm:* 与 assignee。

---

Generated by Claude Code