RED pre-registration — claims-checked-against-disk

verdict: PENDING evidence_class: pending

RED pre-registration — claims-checked-against-disk

• Gate name (ledger name): claims-checked-against-disk
• Registry id: claims — viewer/gate_registry.json:183-188
• Phase: Phase 9 (viewer/verify_claims.cjs:3: "(Phase 9. gate_row: claims-checked-against-disk.)")
• Pre-registered: 2026-08-01
• Runner: viewer/verify_claims.cjs
• CI: ci: true
• Related: viewer/state_blocks.cjs, viewer/generate_state_blocks.cjs

What this document is, and what it is not

This is a transcription, not a judgement. The gate is registered in viewer/gate_registry.json and has no row in evidence/gates.ndjson; the blocker is the empty receipt_path the schema requires (viewer/lab/desk.cjs:317-322, production/schemas/gate_row.schema.json:8).

Every condition below is quoted verbatim, with path:line locators. Appending the ledger row is S4 — the operator's alone.

Motivation (verbatim from the runner)

viewer/verify_claims.cjs:13-27

// EVERY SINGLE DEFECT WAS IN HAND-WRITTEN PROSE:
//
//   - five documents at six locations named "build L6" as the next act, six hours after L6 shipped
//     at 6234f3d, while the plan said the next act was Checkpoint E. AGENT-CALIBRATION-PROMPT.md
//     tells every fresh agent to obey the next act BEFORE verifying anything, so a fresh agent
//     would have rebuilt a finished build;
//   - one banner said 25 registered gates in one paragraph and 23 in another (28 on disk);
//   - one file said 23 at one line and 25 at another;
//   - a review marked those counts CORRECTED while the documents still carried them wrong;
//   - a section named the five panels of a page nobody had fetched, on a port held by a different
//     binary than the source file that was read.
//
// Of 28 registered gates at that moment, the three that touched documents pinned a GENERATED file,
// resolved JSON pointers, or checked a JSON vocabulary. NOTHING CHECKED A CLAIM IN PROSE. The
// failure was not random: it landed in the one region the instrument did not cover.

PASS condition (verbatim)

The runner states no PASS — sentence. It states what it checks, at viewer/verify_claims.cjs:31-32:

// It checks that a citation RESOLVES, that a generated block is FRESH, that an absence claim
// carries a DECLARED SCOPE that still returns nothing, and that no document restates a next act.

and its verdict is conjoined with a time budget, which is easy to miss and belongs in the row. viewer/verify_claims.cjs:437-443:

console.log(`\n  elapsed ${ms} ms (budget 5000)`);
console.log(
  `\nGATE: ${failed.length === 0 && ms <= 5000 ? "PASS" : "FAIL"} - claims, ${results.length - failed.length}/${results.length} checks`
);
console.log("  (Citations, freshness and scoped absences. Whether a SENTENCE IS TRUE is not checkable");
console.log("   here, and this gate never claims it is.)");
process.exit(failed.length === 0 && ms <= 5000 ? 0 : 1);

For the ledger row's pass_condition field:

It checks that a citation RESOLVES, that a generated block is FRESH, that an absence claim carries a DECLARED SCOPE that still returns nothing, and that no document restates a next act. — every check passes AND elapsed <= 5000 ms.

FALSIFIES condition

NOT STATED IN THE RUNNER.

File read: viewer/verify_claims.cjs, in full. The string falsif does not occur anywhere in it. There is no step in evidence/remediation/phase9_plan.json that declares one for this gate either — this gate post-dates the plan's step list.

The nearest thing the runner does state, and it is narrower than a falsifier

viewer/verify_claims.cjs:48-50:

// So: a document NAMES a claim with an annotation; it never DEFINES one. The measures and the
// search scopes live here, in code, in closed registries. An unknown id is a FAIL, not a skip —
// otherwise a typo becomes a way to go green.

That is a rule about one check, not a falsifier for the gate. It is recorded as such and not promoted.

The disclaimer the runner makes about itself — carry it into the row

viewer/verify_claims.cjs:29-36, verbatim:

// WHAT THIS GATE DOES AND DOES NOT CLAIM
// --------------------------------------
// It checks that a citation RESOLVES, that a generated block is FRESH, that an absence claim
// carries a DECLARED SCOPE that still returns nothing, and that no document restates a next act.
// **It cannot check whether a sentence is true.** viewer/verify_plan_consistency.cjs:22-25 says the
// same of status — "a matter of fact about the world, and no scan can settle it" — and that is the
// most honest line in that file. This gate makes the same disclaimer out loud rather than implying
// a completeness it does not have.

A row whose notes omits "It cannot check whether a sentence is true" reads as prose validation. It is not that.

Protocol

1. Run node viewer/verify_claims.cjs from the repository root.
2. --json emits a machine-readable result for the existing HUD pusher (viewer/verify_claims.cjs:427-433). The runner never opens a socket itself (viewer/verify_claims.cjs:429: "THIS FILE NEVER OPENS A SOCKET"); do not add one.
3. Record elapsed ms alongside the verdict. A run that fails only the budget is a different fact from a run that fails a check, and the single GATE: word does not distinguish them.

Ship-gate discipline

• The 5000 ms budget is part of the pass condition. A green obtained on a fast box and a red obtained on a loaded one can be the same tree; note the machine.
• Evidence class C on a first local run.

Non-goals

This gate does not verify that any document is truthful. It verifies citations, generated-block freshness, declared-scope absences, and that no document restates a next act.