finding(cli): an EIGHTH lucide icon-name resolver is emitted inside a template literal in `app-generator.ts`, with zero normalisation — structurally invisible to the record-reading census three ways over

[os-project-manager]

Filed by the domain:ui execution seat (session session_01EMrWaQw3XS5DxTHxp4yRyC) on behalf of the measurement-only run dispatched for #5935, which measured this but was forbidden from creating issues.

⛔ Unlabelled beyond finding — routing, type and priority are the triage seat's to produce. packages/cli could plausibly land in more than one stream and this seat is not its router.

The finding

packages/cli/src/utils/app-generator.ts:842 emits, inside a template literal, a DynamicIcon that resolves an authored icon name by indexing lucide's named-export namespace:

lucideIcons[name]

with zero normalisation — no tokeniser, no toPascalCase, no alias map. That is a third icon-name vocabulary, more forgiving than either of the two the repo censuses, and it ships into generated applications.

Why the existing gate cannot see it

scripts/check-lucide-icon-record-names.mjs discovers record-reading resolvers and reconciles them against DECLARED_RECORD_READERS, failing loudly in both directions. It does not and cannot see this site, for three independent reasons, any one of which would be enough:

1. it lives inside a string literal (emitted code, not executed code in this file);
2. it uses a namespace import rather than the named icons import the discovery keys on;
3. its base is a property access rather than a local binding.

⭐ Its absence from DECLARED_RECORD_READERS is therefore correct, not an omission — it reads neither censused vocabulary. This is not a gate defect to file against the gate; it is a surface outside the census by construction. That distinction matters, because "the gate did not catch it ⇒ the gate is blind" was measured false once already in this family (#6940, where the ratchet had in fact ledgered the drift with the exact question).

Why it is worth tracking anyway

#5935's ruling (2026-08-31, 总监席第 5 场决裁批 #1, verbatim 「同意」, 选项 B) converges the tokeniser and iconNameMap to one copy behind a shared seam, and adds: 「本裁定后新容器 ⛔ 不得再自带解析器,一律走 seam」.

This site is a resolver that:

• does not go through the seam,
• accepts names the seam would reject (no normalisation at all — a raw PascalCase export name works here and nowhere else),
• and is generated into user applications, so its vocabulary becomes an authoring contract nobody declared.

⇒ After #5935's consolidation lands, this would be the one surviving divergence, and the only one invisible to the census that is supposed to prevent exactly that.

Not in this card

• The consolidation itself (Consolidate the eight lucide record-reading icon resolvers into one seam — three tokenisers and a 5-of-8 alias map mean the same authored name renders on one surface and not another #5935 — currently needs-user-decision on a different question: the ruled onUnresolvable domain does not cover the tree's four unresolvable behaviours).
• The gate's discovery predicate. ⛔ Widening it to see inside template literals is a real design question with a real false-positive cost, and is not proposed here.

Provenance and bound

Measured at c1fe272ad during #5935's pre-dispatch enumeration, whose census was re-derived across all 6,229 git-tracked files with the pathspec bound declared. That run's own repo-gate reading was OK lucide icon names: 182 authored/declared names reaching 7 record-reading resolvers are live icons keys (exit 0) — this site is the eighth, and it is outside that 7 by construction rather than by miss.

⚠️ Dedupe bound, stated honestly: the enumeration run's REST search channel answered 403 in its session (its known-hit control was 403 too, so no reading was taken from it) and it deduped on the zero-quota web issues page with a same-run control that returned #5935. Candidates #6837 and #6771 were read and excluded. ⇒ That is a weaker channel than the semantic search; a cheap re-check at triage is warranted.

Refs: #5935 (the consolidation and its ruling) · #5631 (the placeholder-and-warn ruling) · #5633 / #5993 / PR #6350 (earlier members of the same set) · #6721 (header-bar joined RECORD_READING_TYPES) · #5992 (the census blind-spot class)