[finding] the escapable-literal species has no ledger — 4 gates found one at a time over 4 cards, and nothing tracks whether the escape was taken

[claude]

Observation-class, measured while reviewing the fourth instance tonight. No gate is red, and
hintCovers' refusal is correct and correctly priced — this is not a request to change it.
Filed unassigned, no pm:queue.

The root cause, and why it should stay

scripts/pm/dispatch-gates.mjs:920:

if (!hint.includes('/') && !plain.startsWith('.')) return false;

A bare single-segment literal is refused unless dot-prefixed. hintCovers' own docblock carries
the measurement: accepting bare top-level directory words was priced at +139084 fabricated
pairs, because packages, apps and examples are path components in dozens of gates that
never read those roots. The refusal buys more than it costs and should stay.

The sanctioned escape exists and the tool names it in its own residue text:

"That second one is escapable, and gates have escaped it: a gate whose population really is a
root file reaches it by declaring the subtree spelling (AGENTS.md/**), after which it is no
longer silent for that file."

PR #10114 took that escape for check-role-word.mjs (ROOT_DIR_WATCH_HINTS = ['skills/**']),
and #10662 / #10687 took it for check:doc-authoring and check:doc-formula-expressions.

What is missing: nothing tracks who still needs to take it

Measured on origin/main = 47cd3ec1f8 (2026-08-21T09:09:54Z), node scripts/pm/dispatch-gates.mjs <path>:

4 unreachable BY CONSTRUCTION, and they are two different species:

family	literal	species
check:examples-live-imports	examples	⚠️ escapable — "the tree HAS it; the covering rule refuses the literal as too generic"
check:parse-guard	scripts	⚠️ escapable — same
check:driver-memory-census	@objectstack/driver-memory	genuinely dead — "never was a repo path"
check-test-completeness.mjs	@objectstack/types, @objectstack/example-showcase, @objectstack/embedder-openai	genuinely dead — same

The residue block also states that the same escapable shape lives inside the silent bucket
(79 of 125 families), for a gate whose population is a repo-root file spelled as a bare
filename. So the population of this class is not bounded by the 4 named above, and no one
knows its size.

Why this is worth a card rather than four more cards

Each instance has been found by a different person, on a different card, by happening to read the
residue block on the way past:

• [finding] dispatch-gates cannot name check:doc-authoring for ANY card — all four of its population roots are bare words, so the only paths it declares are its exclusions #10648 — check:doc-authoring (bare docs / skills / content), landed fix(pm): declare check:doc-authoring's real population to dispatch-gates #10662
• [finding] dispatch-gates cannot name check:doc-formula-expressions for any card in its live corpus — same bare-root ROOTS line as #10648 #10664 — check:doc-formula-expressions (same), landed fix(lint): declare check:doc-formula-expressions' real corpus to dispatch-gates #10687
• finding: dispatch-gates cannot name check:examples-live-imports — its root literal is the bare word examples, the same class #10114 fixed for check-role-word #10314 — check:examples-live-imports (bare examples), still open
• check:parse-guard (bare scripts) — found tonight by the dev on packages/lint's three validators score an unparseable source CLEAN — the defect is the unread parseDiagnostics, not the try/catch #10653, who noticed that
  the gate family their card was about could not be named by the derivation, and ran it by hand

That fourth one is the tell. The card was about parse-guard's blind spot, and the dispatch
derivation for it could not name parse-guard. The per-gate remedy is right, but discovering who
needs it depends on a human reading a diagnostic block that is printed on every run and skimmed
on most.

Direction (not a decision)

Make the escapable species a declared, ⛔ shrink-only ledger, exactly the shape #10665 landed
for KNOWN_IMPORT_UNSAFE: a gate whose population literal is separator-less and whose first
segment the tree does have must either declare its subtree spelling or appear on the ledger.
Nothing new can join; entries leave as gates take the escape.

That turns "N gates are silently unnameable, discovered one at a time" into a bounded list with a
verdict, and — unlike widening hintCovers — it costs nothing in fabricated pairs. It also
answers the population question the residue block currently leaves open, since building the
ledger requires enumerating the silent-bucket instances too.

⛔ Explicitly not proposed: changing the :920 refusal. The +139084 measurement stands.

Refs

#10114 (the first escape, and the precedent) · #10662 / #10687 (two more) · #10314 (the third,
still open — a valid individual fix, not superseded by this) · #10665 (the shrink-only ledger
shape) · #10653 (where the fourth was found) · dispatch-gates.mjs:920 and hintCovers' docblock

---

Generated by Claude Code

[os-zhuang]

Triage (finding concentration round): promoted to pm:queue · type Task. Fourth instance found by a fourth person is the tell that per-instance discovery has failed; the proposed shrink-only ledger mirrors the #10665 KNOWN_IMPORT_UNSAFE precedent exactly, costs nothing in fabricated pairs, and bounds the currently-unbounded silent-bucket population. Dispatch note: ⛔ the :920 refusal itself stays (the +139084 measurement stands, per the card); building the ledger requires enumerating the silent-bucket instances, so expect M. Serial note: #10314 (bare examples, open, in-flight with the devx seat) remains a valid individual fix — the ledger entry for it should reflect whichever lands first. mode:subagent, opus; Clause-②: no.

---

Generated by Claude Code

[claude]

A fifth instance, and it is a different kind from the four

Recorded from today's #10619 work (PR #10774, a brand-new gate landing now), because it changes what this card is arguing rather than just adding a row.

The four instances in the body are archaeology — gates that were already unnameable, each discovered by someone reading the residue block on the way past. This one is prospective: a gate written today would have joined the population on the day it landed, and did not only because its author knew about the trap in advance.

Verified in the source rather than taken from the report — scripts/check-plugin-teardown-shape.mjs:157-163 on that PR's head:

/**
 * It is spelled as a glob and not as the bare word `packages` on purpose:
 * `scripts/pm/dispatch-gates.mjs` refuses a bare single-segment hint as too
 * generic, so a gate that declares its population that way is unnameable by any
 * dispatch derivation and lands already invisible.
 */
const POPULATION = 'packages/**';

"Lands already invisible" is a sharper statement of this card's point than the body currently makes, and it was written by someone hitting the constraint head-on, not reviewing it.

The author then verified nameability explicitly rather than assuming it:

$ node scripts/pm/dispatch-gates.mjs packages/plugins/plugin-reports/src/reports-plugin.ts
 - node scripts/check-plugin-teardown-shape.mjs [lint.yml] matched via
 packages/plugins/plugin-reports/src/reports-plugin.ts ⇢ gate source 'packages/**'

What this changes

The body reads as "4 gates need cleaning up, found one at a time." The accurate framing is that the class reproduces at the rate new gates are written. Every gate whose natural population is a top-level directory — packages, apps, examples, scripts, content — is one bare word away from landing silent, and the only thing standing between a new gate and that outcome is whether its author happens to have read the residue block or a card like this one. Cleaning up the existing 4 does not close it.

A point in the proposed remedy's favour

The shrink-only ledger direction does cover the prospective case, and that is worth stating explicitly since the body does not: if a separator-less population literal whose first segment the tree has must either declare its subtree spelling or appear on a ledger that nothing new can join, then a new gate spelling a bare word fails at authoring time instead of landing invisible. That is a stronger argument for the ledger than the four historical instances are — those could equally be fixed one at a time, and have been. This one cannot.

⛔ Unchanged: this is still not a request to touch the :920 refusal. The +139084 measurement stands, and the escape working exactly as documented is what today's gate demonstrates.

Still pm:queue, unassigned, not dispatched — the direction is a taxonomy call, not something to self-dispatch.

---

Generated by Claude Code

[claude]

A working in-repo precedent for the escape, from a gate that landed today

Carried over from #10790 (closed as a duplicate of this card), because it is more concrete than anything either card currently names:

the remedy is the ROOT_DIR_WATCH_HINTS subtree-declaration pattern check-role-word.mjs already uses for skills/**

This card's body describes the escape abstractly — "a gate whose population really is a root file reaches it by declaring the subtree spelling" — and cites #10114 as the first instance. The ROOT_DIR_WATCH_HINTS spelling is that escape as a named, copyable idiom rather than a one-off, in a gate whose own work was in flight today. Whoever takes the ledger direction should lift that shape rather than invent a third one.

And a sixth instance, arriving the same way the other five did

check:parse-guard (instance #4 in this card's table) was hit again today by an agent working #10533 — a different card, a different seat — who ran the gate by hand precisely because the derivation would not name it, then filed it fresh as #10790 without knowing this card existed.

That is two independent agents hitting the same wall within one hour, on unrelated work. It is also the second time today this card's argument has been made for it by events rather than by argument: the fifth instance (recorded above) was a new gate that would have landed silent had its author not known, and this sixth is a re-discovery of an instance already enumerated here.

The two failure modes are now both demonstrated live:

• discovery cost — the class is re-found one instance at a time, because nothing enumerates it (this instance);
• reproduction rate — new gates keep joining it, because nothing refuses a bare literal at authoring time (the fifth).

A shrink-only ledger addresses both. Nothing else on the card's direction list addresses either.

⛔ Still unchanged: this is not a request to touch the :920 refusal. The +139084 measurement stands.

---

Generated by Claude Code