Skip to content

Issue #417: D3 blocked — no qualifying candidate

Status: blocked. D3 was not launched. Issue #416 closed its candidate family with zero candidates and no winner, so the explicit #417 no-candidate condition applies. No candidate/champion pair was frozen and no D3 origins, forecast attempts, or scored pairs exist.

The machine-readable JSON report is the accounting record. Its source digests are verified by the focused test. This is a blocked outcome, not a D3 accuracy evaluation.

Recorded outcome

  • Candidate: null; champion: null; candidate/champion pair: null.
  • D3 launch: no; status: blocked; origins: 0; exact pairs at 2h/4h/8h/16h: 0.
  • Every metric is null because no D3 scoring took place.
  • No accuracy peeks or D3 outcome access occurred. No production model, forecast, or configuration changed.
  • Source evidence: #411 reports zero confirmatory forecasts and zero matured pairs; #416 records zero actual candidate attempts/scores and zero candidates/winner. The #410 protocol remains the frozen scoring reference.

Reopening contract

D3 may be reconsidered only after all of the following are frozen, hashed, and established before the first D3 score is examined:

  1. One selected pair and complete preregistration: #416 must select exactly one candidate that qualifies under its frozen selection contract. Freeze candidate and champion identities, candidate configuration, model/checkpoint, policy, code revision, data/source and vintage lineage, start/end and fixed evaluation cutoff, stopping rule, practical-effect threshold, and all D3 metrics/criteria. Hash the complete preregistration before any D3 score access.
  2. Independent exact-pair cohort: D3 origins and outcomes must be disjoint from candidate development, inner and outer folds, and the final holdout. For candidate, champion, persistence, and strongest naive, use identical exact UTC (origin_at, target_at, horizon) pairs, shared source vintage/targets, and the same scoring rules.
  3. Minimum duration and maturity: collect at least 30 calendar days and at least 200 exact matured pairs separately at each of 2h, 4h, 8h, and 16h. Report dependence and effective sample size; leave the gate blocked if any requirement is unmet.
  4. Complete accounting and frozen inference: retain every forecast, failure, retry, and exact maturity status. Apply the frozen family-adjusted uncertainty procedure and criteria for primary effect, baseline skill, per-horizon regression, directional loss, and interval safety. Preserve failures and do not silently discard attempts.
  5. No repeated accuracy peeking: access only predeclared operational/health checks before the fixed cutoff and final analysis. Do not repeatedly test accuracy or stop early for favorable results. Report every interruption or source-vintage change and handle it under the frozen plan.
  6. Human-reviewed disposition only: report blocked, inconclusive, negative, or eligible for human review with artifacts and hashes. Eligibility is not approval; no automatic promotion, public output change, or configuration mutation is permitted.

These conditions restate the #417 acceptance contract. A future evaluation requires new qualifying #416 evidence and a separately frozen D3 plan; this report authorizes no scoring.