Skip to content

Issue #412 — TimesFM context/target evaluation blocked

Outcome: blocked; no scoring or hypothesis

The prerequisite in #412 is an adequately powered, measured context/horizon failure from #411. The frozen #411 readiness report passes protocol validation, but it does not identify such a failure: its historical corpus and prospective evidence gates are blocked, all reported metrics are null, and it has zero confirmatory forecasts, zero expected pairs, and zero matured pairs. Its last verified prospective snapshot had 48 timestamp-eligible forecasts, all 48 excluded as legacy/unversioned; the newest provenance-complete origin at 2026-09-27 11:00 UTC was not mature. Therefore no TimesFM context or target policy was scored, and this record does not invent a candidate hypothesis.

No candidates or folds were preregistered or run. All metrics and the winner are null; no holdout was accessed, and no production code, configuration, or forecast policy was changed. This is a blocked research disposition, not a negative result about any context or target policy.

Machine-readable disposition: JSON.

Evidence snapshot and hashes

The source is the checked-in #411 report at ISSUE_411_PARITY_BASELINES_BLOCKED.json, SHA-256 c1b3471495c0c32bcb8a5a90ae7fe5a4e20a6bdcb5d4d617e8d83e8be18202a7. It records the #410 protocol gate as passed, while corpus and eligible mature production-policy evidence remain blocked. The #397 canonical corpus audit records 0 of 32,136 target hours and 0 of 4,320 warm-up hours; its SHA-256 is e2bf0c671a5a4327c6852d02614a7d2e403ed12f8fe7c4362336c5e87d44a0c7.

The frozen protocol is ISSUE_410_SCORING_PROTOCOL.json, version v11-prospective-1.0.1; file SHA-256 1b8c6705d53882a66c0ab1277cf2a63c845d00c2c7d0cc978c0c75d8121efd18, protocol digest 4da936d505df9a15cbd073ee85ee0e4ed2db483e6058f6b81ce163f4724bb4f0. The report's last-verified prospective snapshot supplies code SHA-256 79a95fc3debd4134ab68cfc55795007a340f15a6eff2ef7212a875a9846c28fc, database SHA-256 b102f14db535264ecc2474a3b338c929abea4c39616f233a5701611c9300f57f, and cohort SHA-256 462287eb433a27c16e4f25360d9ee40e5dda2fabdced13e5390859a055f7bbb8. These are the hashes and counts as reported by the frozen #411 artifact, not a new evaluation performed for #412.

Exact reopening gates and next action

  1. Resolve the #411 corpus and eligible mature production-policy evidence blockers, then rerun its frozen readiness protocol and retain the report and source hashes.
  2. Require that eligible, mature, exact-key #411 diagnosis name a concrete context/horizon failure and meet #410's minimum of 200 paired origins per horizon and 8 effective independent blocks per comparison. A missing or underpowered diagnosis remains blocked/inconclusive; do not infer one.
  3. Before viewing any #412 candidate scores, preregister the diagnosis-linked bounded candidate family, model/checkpoint identity, outcomes/comparisons, fixed folds, all exclusions and failures, stopping rule, practical effect, and complete multiplicity family.
  4. Evaluate on nested earlier purged walk-forward folds with exact shared UTC (origin_at, target_at, horizon) and input-vintage keys. Lock a chronological final holdout disjoint from candidate selection and all prior evaluations; preserve it untouched until the preregistered final analysis.

Next action: do not score or propose context/target candidates now. First resolve #411 and obtain its adequately powered measured diagnosis. Only then preregister the bounded #412 design and disjoint holdout before any scoring. Any later research recommendation remains separate from production changes.