finding: three `edit distance cannot reach` citations are measurably false — and ~12 more of the 19 occurrences are unclassified

[os-bill]

Filed by the domain:spec execution seat from the #17361 round, which measured these but ⛔ left them alone — they sit outside its declared file face. ⛔ No domain:* and no priority from me; grading and routing are triage's. ⚠️ One of the three sites is in packages/objectql, which is not this seat's lane.

Three citations that are measurably false — class (b)

Each says edit distance cannot reach a case where, measured, it does:

site	cites	measured
packages/objectql/src/validation/record-validator.ts	"edit distance cannot reach it (latitude → lat)"	FALSE — latitude has budget 2 and is distance 2 from the declared altitude, so the fallback reaches it and answers altitude
packages/spec/src/data/default-value-shape.ts	same citation	FALSE, same measurement
packages/spec/src/data/driver/turso.zod.ts	"Semantic near-misses only — the spellings edit distance cannot reach", above a table containing uri → url	FALSE — the fallback reaches uri → url at distance 1 and answers identically

⭐ The first two are worse than merely wrong: the example is cited as proof of the rule and proves the opposite. latitude → lat is an overrule — an alias whose job is to beat a hit the fallback reaches and gets wrong — which is the exact category the sentence claims does not exist.

The measurement this rests on, and why it is a reading

PR #17662 (card #17361) re-derived the real behaviour rather than reasoning about it: the lookup is aliases[aliasProbe(key)] ?? findClosestMatches(...) — alias first, winning outright — so an alias has two jobs, not one. Its census forced the strictObjectDeclarations() registry over every module under packages/spec/src and ran the real findClosestMatches against each surface's own known keys at that key's own budget:

384    registered surfaces
1910   alias entries total
1658   unreachable gap        ← the only role the prose described
 252   reachable
 211     …fallback agrees
  41     …fallback answers a DIFFERENT key — the alias OVERRULES it

Lit control: visibleWhen (the docblock's own proving case) classifies as GAP/unreachable, so the instrument does classify and the surviving example is verified. Dark control: a fabricated alias token matches 0 rows. The edit distances were computed with a Levenshtein written in that round — ⛔ not the repo's function, ⛔ not inherited — whose own controls return unreachable too (filesystem → fs = 8 vs budget 3; paths → fs = 4 vs budget 2), so it is not an instrument stuck on one answer.

⇒ 41 overruling aliases ship today. The false sentences above are not an edge case; they misdescribe a category with 41 members.

⚠️ The population is only partly classified — do not treat this card as the whole list

The phrase pattern (edit distance cannot/can not/can't reach, case-insensitive, whitespace-flattened) occurs 19 times across 17 tracked files. PR #17662 corrected 4 of them. This card names 3 more that are measurably false. ⭐ That leaves roughly a dozen unclassified — ⛔ do not close this line of work on the three rows above.

Already re-measured and TRUE — ⛔ do not re-check these: ui/chart.zod.ts (both tables), data/object.zod.ts, and the hedged "most of these" in data/authoring-key-lint.ts.

What a taker should do

Correct each false citation to describe both roles, and ⭐ carry the measured number rather than a hedge — "usually", "mostly" and "rarely" are how the original sentence survived being wrong. Re-measure before writing: anchor by content, not by line; ⚠️ flatten whitespace first (these phrases wrap across line breaks, and for a JSDoc block strip the leading * per line, or a flatten still reads a mistyped-anchor 0); give every zero a lit control.

⚠️ packages/objectql is a different lane from packages/spec. If triage routes this as one card, the round that takes it will need the cross-domain in-flight check; splitting it is also reasonable.

Duplicate check — completed to the limit the session allows

⛔ Full-population dedup is not available: REST /search/* is refused for this session and MCP search_issues is rate-limited. What I completed at 08:29Z: enumeration of 569 open issues, grepped by TITLE — edit distance 1 (#17361, the card being fixed), aliases 1 (same), strictObject 2 (#17361 and #8213, a different defect), near-miss 2 (#17361, #16303 — a different defect), strict-object 0, suggestion 0. Lit control spec = 75; fabricated control = 0. ⚠️ Titles only; bodies were not searched. Close as duplicate without hesitation if one surfaces.

Related: #17361 / PR #17662 (the four sites it fixed, and the census) · #16303 (near-miss spellings costing the reading, different mechanism) · #17655 / #17656 (the same shape of defect — one false sentence copied across files — on a different subject)

---

Generated by Claude Code

[os-bill]

The unclassified residue, measured — and the card (mine) is right on the verdict, imprecise on the reason, and missing a carrier

Follow-up by the filing seat, 2026-09-12T01:43Z, on origin/main @ 7ef2965047. ⛔ No grading, no routing, no label touched — this card carries finding only and first grading is triage's.

1. The population moved: 19 → 16

git grep -o -F 'edit distance cannot reach' origin/main | wc -l   → 16

⚠️ Not a correction of the original count — PR #17662 landed in between. ⛔ Occurrences, not lines (grep -c reads 14 here and would undercount by 2). Dark control edit distance cannot zzqq → 0.

2. All 16, classified

#	site	class
3	.changeset/strict-object-aliases-two-roles.md (×3)	✅ correct by construction — PR #17662's own changeset, quoting the old sentence as the thing it corrected
1	.changeset/permissions-alias-hosts-justification.md	✅ correct by construction — same, for the hosts fix
2	packages/spec/CHANGELOG.md:3407, packages/objectql/CHANGELOG.md:607	⚠️ frozen historical records. spec's says all three manifest aliases are unreachable, which the hosts fix disproved. A CHANGELOG is not rewritten; noting so nobody re-finds it
3	builtin-node-config.test.ts:67, state-machine.test.ts:234, object-strictness-batch20.test.ts:508	✅ true — measured: object→objectName dist 4 (budget 2), guard→cond dist 4 (budget 2). Both genuinely unreachable
1	kernel/manifest.zod.ts:51	⭐ already fixed, and it is the EXEMPLAR — see §4
3	ui/chart.zod.ts:774, :839, data/object.zod.ts:1607	⚠️ NOT MEASURED — each needs its own surface's declared key set. ⛔ I did not verify them and ⛔ they must not be read as clean
3	data/field-value.zod.ts:341, data/default-value-shape.ts:122, objectql/.../record-validator.ts:558	⛔ the live defect — §3

3. ⭐ The verdict holds, but the card states the reason wrongly — and misses the declaration site

The budget is Math.max(2, Math.floor(key.length / 3)) (shared/suggestions.zod.ts:461), and LocationValueSchema (data/field-value.zod.ts:334-:351) declares lat, lng, altitude, accuracy with aliases: { latitude: 'lat', longitude: 'lng' }. Measured Levenshtein:

probe 'latitude'   budget 2      probe 'longitude'  budget 3
  -> lat        5                  -> lat        7
  -> lng        7                  -> lng        6
  -> altitude   2  <= REACHED      -> altitude   4
  -> accuracy   8                  -> accuracy   9
  fallback answers: altitude        fallback answers: NOTHING

⇒ the clause "edit distance cannot reach it (latitude → lat)" is literally TRUE — 5 > 2. ⚠️ So my card's wording 「measurably false」 is too strong for these two sites, and I am correcting it rather than leaving it to be discovered. The defect is that the sentence is offered as the reason the alias exists, and the real reason is the stronger one: without the alias the fallback is not silent — it confidently answers altitude, a declared key the author never meant. The alias overrules a wrong hit; the prose describes it filling a silent gap.

⭐ And the two aliases on that one line are opposite cases: latitude is an overrule (fallback reaches altitude), longitude is a genuine gap (fallback reaches nothing). One comment covers both and is right about only one.

⭐ The card missed a carrier. It listed default-value-shape.ts and record-validator.ts. The third is data/field-value.zod.ts:341-:343, the alias declaration itself — the one place an adopter reading LocationValueSchema actually looks. ⇒ 3 live sites, not 2.

⚠️ Only uri → url (data/driver/turso.zod.ts:83) is false in the card's original sense: dist 1, budget 2 ⇒ the fallback reaches the intended key, so curating it changes nothing.

4. ⭐ The fix already exists in-tree — kernel/manifest.zod.ts:50-:63

// These two are the unreachable case: edit distance cannot reach a
// two-letter abbreviation from the word it abbreviates …
  filesystem: 'fs',  paths: 'fs',
// `hosts` is the opposite case, and the stronger reason to curate an
// entry: it IS within budget of `hooks`. … Without this line the fallback
// answers `hosts` -> `hooks` … This alias overrides a confident WRONG
// suggestion rather than filling a silent gap, and
// `manifest-unknown-keys.test.ts` pins that.

⇒ whoever takes this card should copy that shape, ⛔ not invent wording: it names both roles, gives the budget arithmetic, and points at the test that pins it. ⚠️ My own first attempt at this measurement computed hosts → network (dist 7) and nearly concluded the hosts fix was wrong; the reachability is against hooks, which only reading the source showed. ⇒ ⛔ do not compute a distance to the alias's target; compute it to every declared key.

_{domain:spec execution seat · session_01MkQhmuuJAVDjmeWNixwDDH · readings taken 2026-09-12T01:43Z on origin/main @ 7ef2965047}

---

Generated by Claude Code