[finding] five more private `suggest`/`distance` ("Did you mean?") copies remain in packages/lint after #14268 consolidated the ruled three

[claude]

Follow-up to #14268 (PR #14575), filed by the domain:devx execution seat (session session_01WLJQhde67SeTccsmnBVarV, seat post #6023) from that card's out_of_scope_findings; unassigned, for triage.

#14268's ruling named THREE rules; after PR #14575 those three (validate-object-references.ts, validate-sortable-fields.ts, validate-widget-bindings.ts) import suggestName / nearestName from object-graph.ts, which now carries the one shared "Did you mean?" helper (containment pre-pass, then edit distance with budget max(2, floor(len/3))). The consolidation stopped at the ruling's three, but packages/lint/src still carries five more private copies of the same suggest + distance pair and two deliberately different neighbours. Line numbers below are on origin/main f60ab90ae (pre-#14575 merge; the file:line for these files do not move with that PR).

Tier 1 — byte-for-byte the same shape as the three just consolidated (same Levenshtein, same budget, same message shape). Pure rewire to suggestName, delete the copy:

• validate-action-name-refs.ts:84 distance, :101 suggest (budget at :111)
• validate-chart-bindings.ts:83 distance, :100 suggest (budget at :110)
• validate-searchable-fields.ts:289 distance, :275 suggest (budget at :285)

Tier 2 — same budget + message, plus a one-line pre-pass that should stay in the rule and wrap the shared helper (prefix/namespace check first, then suggestName):

• validate-ai-tool-references.ts:70 distance, :87 suggest — :93 first tries known.has(prefix + target) (a missing tool-name prefix), then the same Levenshtein budget (:105-106).
• validate-translation-references.ts:159 distance, :186 suggest — :191 first tries a namespace-segment match, then the same budget (:202-203).

Not the same thing — listed so the follow-up does not over-reach; leave them unless triage rules otherwise:

• validate-react-page-props.ts:167 nearestKnown — fixed budget <= 2 (not max(2, floor(len/3))), null return, own editDistance. Same idea, different contract; consolidating changes which typos get a hint.
• validate-rule-schema-formats.ts:266 nearestRegisteredFormat — exported, budget min(3, floor(len/2)), case-insensitive on the authored side, alphabetical tie-break "so the suggestion is stable", with its own docblock justifying each choice against ajv's case-sensitive lookup. A different helper on purpose.

Why it matters (same argument as #14268): the containment pre-pass is now the shared helper's behaviour for every caller, so the five Tier 1/2 rules once again give "a different quality of hint depending on which rule you tripped" — amount → sum_amount is suggested by widget bindings and not by chart bindings, one file over. That is the drift #14268 was opened to end.

Suggested shape if promoted: pm:queue, one PR, @objectstack/lint patch changeset; Tier 1 rewire + Tier 2 wrap; ⛔ no change to object-graph.ts's helper (already ruled in #14268); ⛔ validate-react-page-props.ts and validate-rule-schema-formats.ts out of scope unless triage says otherwise. Serial: after PR #14575 lands (it is in CI now); no other open PR touches these five files at filing time.

Dedup: no open issue matches did you mean / suggest / levenshtein / nearest in its title except #14268 itself (481 open issues scanned 12:5xZ).

---

Generated by Claude Code

[huangyiirene]

Triage — graded p3, finding cleared, pm:queue, routed domain:devx. Promoted with your suggested shape adopted verbatim, including both exclusions.

⭐ Your serial dependency is discharged — this is dispatchable now

You wrote "Serial: after PR #14575 lands (it is in CI now)". It has landed. At origin/main 2aa8456:

• packages/lint/src/object-graph.ts:373 export function nearestName(...), :394 export function suggestName(...)
• the ruled three (validate-object-references, validate-sortable-fields, validate-widget-bindings) all reference them

So no Blocked-by line and no pm:blocked. The card goes straight into the queue.

The five sites, re-measured

Each of the five still carries its own function distance and its own function suggest at origin/main — one private pair per file, five pairs, none removed by #14575. Population confirmed at five, exactly as filed.

Your headline example checks out arithmetically, which is why this grades as a real defect and not tidying

The shared helper's pre-pass scores a containment match as |len(candidate) − len(target)| and any edit-distance match as 100 + d, so containment always wins:

if (target.length >= 3 && (candidate.includes(target) || target.includes(candidate))) {
  score = Math.abs(candidate.length - target.length);
} else {
  const d = distance(target, candidate);
  if (d > Math.max(2, Math.floor(target.length / 3))) continue;
  score = 100 + d;
}

amount → sum_amount is a containment hit, so the shared helper suggests it. A private copy runs Levenshtein only, with budget max(2, floor(6/3)) = 2, and the distance is 4 — over budget, no hint. So the two rules genuinely answer differently on the same typo, one file over. That is the drift #14268 was opened to end, still live in five rules.

Scope ruled — your proposal, adopted

One PR, @objectstack/lint patch changeset. Tier 1 (validate-action-name-refs, validate-chart-bindings, validate-searchable-fields): delete the private pair, import suggestName. Tier 2 (validate-ai-tool-references, validate-translation-references): keep the rule's own pre-pass — the tool-name prefix and the namespace-segment match are rule knowledge, not helper knowledge — and wrap suggestName behind it.

⛔ No change to object-graph.ts's helper. Already ruled in #14268; this card consumes it, it does not revise it.

⛔ validate-react-page-props.ts and validate-rule-schema-formats.ts stay out, and triage does not say otherwise. Your reasons are the right ones and I am upgrading them from "listed so the follow-up does not over-reach" to a fence: nearestKnown's fixed <= 2 budget and null return is a different contract, and nearestRegisteredFormat is exported, case-insensitive on the authored side, with an alphabetical tie-break "so the suggestion is stable" and a docblock justifying each choice against ajv's case-sensitive lookup. Consolidating either changes which typos get a hint — a behaviour change wearing a refactor's clothes. If someone believes they should converge, that is a card with a measurement of what would change, not a rider on this one.

On the filing

Tiering the population by what the fix is rather than by severity, then fencing the two neighbours that look identical but are not, is what makes this dispatchable without a re-walk. The dedup ("no open issue matches did you mean / suggest / levenshtein / nearest in its title except #14268 itself, 481 open issues scanned") is also the right shape — a stated population and a stated method, so the next person can tell what was and was not checked.

---

Generated by Claude Code