UNI Universal Natural Intelligence

Wiki · Evidence & Verdicts

RED pre-registration — gaia-drift-wellformed

Evidence & Verdicts · docs/receipts/red_preregistration_drift_wellformed.md @ 44baf03d5041 (gen2-runtime) — opens the published snapshot ac338733bbba

How to read this page

Eighty-seven dated pages: receipts, pre-registrations, handoffs, validation records and review verdicts. A receipt is written at the moment a piece of work was checked. It names what was claimed, the commit and the seed, what was actually run, and the outcome in one of a small set of controlled words. Then it names what the work did not achieve. That last part is what makes it a receipt rather than an announcement. A pre-registration is the same discipline run in advance: the conditions that would count as a pass and the conditions that would falsify the claim are written down before the run, so neither can be adjusted once the numbers arrive.

That is why so many small dated stubs are an audit trail rather than noise. No one of them is meant to be a good read. The value is in the sequence and in the dates, because you can watch a prediction be registered, then the run happen, then the verdict land — sometimes against the prediction. Pages here record a falsified result, a rejected fix, a retracted overclaim, and a green receipt that turned out not to be reproducible from the commit that carried it. A record that carried only successes would be worth a good deal less than this one.

A gentle way in is to read a pre-registration first, so the shape becomes familiar, then a result page, then one of the corrections. This section sits off the main navigation on purpose: it is the record you check the rest of the site against, not the place to begin.

What it is not: documentation, and not a summary. Nothing here has been tidied in hindsight. Every entry reads as of its date, a later entry may overturn an earlier one, and the presence of a page is not a claim that its result stood.

A Plain and a Clear version of this page have not been written yet. What follows is the document itself.

Precise — the source document

This is the document. Rendered from the repository at the commit above, with nothing rewritten for the web. A gate re-renders it on every deploy and fails the build if a single byte differs.

verdict: PENDING evidence_class: pending

RED pre-registration — gaia-drift-wellformed

  • Gate name (ledger name): gaia-drift-wellformed
  • Registry id: drift-wellformed — viewer/gate_registry.json:36-40
  • Phase: Phase 9, step 1.5 (viewer/gaia/verify_drift_wellformed.cjs:1)
  • Pre-registered: 2026-08-01
  • Runner: viewer/gaia/verify_drift_wellformed.cjs
  • CI: ci: true
  • Related: ADR-0002 Amendment 1 (Decisions 5 and 8), viewer/gaia/collectors.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/gaia/verify_drift_wellformed.cjs:3-9

// Amendment 1, Decision 5: both sides of a drift signal MUST be the same kind under the same normalization —
// a comparison is well-formed only if `equal: true` is REACHABLE, i.e. some achievable state of the world
// makes the two byte-sets identical. Five signals failed that: prose against a path (fqdn_cjs,
// gate_row_schema_path), a label against an array (resolver_planned), a JSON blob against a 54 KB document
// (self_caps_doc_vs_served). They stayed red through a day of real corrections and would have stayed red had
// every correction been perfect. The cost is not the red pixel — an inequality nobody can act on stops being
// read, which is how drift.git_dirty_vs_clean sat unread while pointing at a live defect.

PASS condition (verbatim)

viewer/gaia/verify_drift_wellformed.cjs:3-5 — Decision 5, quoted by the runner as the property it holds every repaired signal to:

// Amendment 1, Decision 5: both sides of a drift signal MUST be the same kind under the same normalization —
// a comparison is well-formed only if `equal: true` is REACHABLE, i.e. some achievable state of the world
// makes the two byte-sets identical.

Mechanical form, viewer/gaia/verify_drift_wellformed.cjs:19:

// Usage: node viewer/gaia/verify_drift_wellformed.cjs      exit 0 = PASS, 1 = FAIL.

For the ledger row's pass_condition field:

both sides of a drift signal MUST be the same kind under the same normalization — a comparison is well-formed only if equal: true is REACHABLE, i.e. some achievable state of the world makes the two byte-sets identical.

FALSIFIES condition (verbatim)

The runner does not use the word "falsifier". It states its FAIL condition in its own words at viewer/gaia/verify_drift_wellformed.cjs:15-17:

// SO EVERY REPAIRED SIGNAL IS MUTATED HERE (M1). Each mutation runs against a REBUILT comparison using the
// same rule the collector uses, on FIXTURE INPUTS — the real repository is never edited, and no signal is
// re-pointed at anything on disk. A repair that cannot be shown to bite fails this gate.

and quotes Decision 8 at viewer/gaia/verify_drift_wellformed.cjs:11-13:

// Decision 8 is why this file is MANDATORY, not optional: "Every repaired comparison must be proved to still
// bite — point its declared side at a bad value and watch `equal` go false. A comparison repaired without
// that proof is indistinguishable from a comparison loosened."

For the ledger row's falsifies_condition field:

A repair that cannot be shown to bite fails this gate. — "A comparison repaired without that proof is indistinguishable from a comparison loosened." (ADR-0002 Amendment 1, Decision 8)

evidence/remediation/phase9_plan.json step 1.5 declares the same in the plan's own words: "a comparison repaired without a bite-proving mutation is indistinguishable from one loosened". The runner never labels either sentence "the falsifier" — that word does not appear anywhere in the file. The quotes above are its stated FAIL condition, not an agent's reading of one.

Protocol

  1. Run node viewer/gaia/verify_drift_wellformed.cjs from the repository root.
  2. Fixtures only — viewer/gaia/verify_drift_wellformed.cjs:16-17: "the real repository is never edited, and no signal is re-pointed at anything on disk."
  3. Record the exit code and the final DRIFT WELL-FORMEDNESS GATE: line (viewer/gaia/verify_drift_wellformed.cjs:326).

Ship-gate discipline

  • Every check must report both convergence and bite. A signal that converges but cannot be made to go unequal is the loosened comparison Decision 8 names; the runner already fails on that (viewer/gaia/verify_drift_wellformed.cjs:42). Record which arm failed, not just that one did.
  • This file must itself avoid writing literal ghost schema paths — verify_schema_pointers.cjs caught it twice on its first two runs, once for a fixture and once for the comment describing the fixture (viewer/gaia/verify_drift_wellformed.cjs:55-59). Any edit to this runner must be re-checked against that sibling gate.
  • Evidence class C on a first local run.

Non-goals

This gate does not assert that any drift signal currently reads equal: true. It establishes only that each repaired comparison is well-formed — that equality is reachable — and that each still bites when a side is pointed at a bad value.

sha256 a6dbb6ded54a78ac — of the original file, so what was ingested stays checkable.