One click runs the corrected totals math chronologically from strict pre-kickoff receipts. DCR uses its true conversion-rate scale, EPA and every additive feature are centered on that week’s leakage-safe pregame population, and the scoring level uses only prior completed games with shrinkage. Legacy moneyline/spread outputs remain unchanged. The old +16 overlay is excluded, and promotion remains blocked until both locked accuracy and historical market-profitability evidence pass.

Phase 5A preflight has not been run yet.

Preflight checks immutable KV before any provider action. When all snapshots are cached, this lane rebuilds separate v3 D1 receipts and the audit with zero Odds API calls or credits; only genuinely missing snapshots can reach the historical provider. Requests remain locked to moneyline (h2h), spread and total at T−6h and T−5m. The clean audit excludes same-week outcomes, gives each bookmaker one consensus-line vote, settles only an actual bookmaker price, and explains every uncovered predicted game. Market data never enters totals math, and no result promotes automatically.

Phase 5E historical market preflight has not been run yet.

Runs a read-only truth test over the immutable Phase 5A replay and clean Phase 5E market audit. It measures model-vs-market accuracy, edge discrimination, probability calibration, cold-start bets, and each totals stage. It makes zero provider calls, spends zero Odds API credits, writes nothing, and cannot promote totals.

Phase 5F signal truth has not been run yet.

Evaluates one predeclared read-only totals candidate built directly from the Phase 5F diagnosis. Correlated EPA/YPP/PPD/DCR/red-zone inputs vote through one robust efficiency block instead of stacking across multiple stages; pace owns one volume adjustment; penalties own one discipline term. The old chaos mean shift, secondary total adjustment, duplicate DCR/pace paths, and legacy total calibration are excluded. Candidate probability comes from strictly earlier-week candidate residuals, and zero-history games are projected but cannot become bets. 2025 outcomes are evaluation-only and never fit candidate coefficients or thresholds.

Phase 5G bounded totals candidate has not been run yet.

Rebuilds the totals mean from the football scoring identity: expected drives × matchup points per drive. Each offense is paired symmetrically with the opponent defense using certified D1 team-game history from strictly earlier weeks. The existing four-game live-bridge reliability budget shrinks small samples toward the weekly league center. YPP, third-down rate, and expected plays remain diagnostic receipts only, so they cannot double-count the mean. Market prices and target outcomes never enter the projection.

Phase 5H drive matchup candidate has not been run yet.

Audits whether a semantically cleaner pace measure adds information beyond the real sportsbook total before we build another totals candidate. It derives strict-prior possession seconds per offensive play from certified D1 team-game truth, pairs each offense with opponent defensive pace faced, applies the existing four-game reliability prior, and compares that diagnostic against actual market residuals. Phase 5I creates no projected total, no bet, no probability, and no production mutation.

Phase 5I tempo orthogonality audit has not been run yet.

Tests semantically distinct strict-prior football families one at a time before another totals candidate is allowed. The audit measures tempo, EPA matchup, explosives, red-zone scoring, turnovers, sacks, penalties, pass tendency, and recent scoring form against the real sportsbook residual and again after controlling for the sportsbook signal. 2025 outcomes are evaluation-only: no coefficient, threshold, probability, bet, or ROI bucket is fit here.

Phase 5J orthogonal football signal audit has not been run yet.

Searches only the completed Phase 5A 2025 D1 replay for recurring winner-direction combinations. Weeks 2–12 discover candidate fingerprints; those exact definitions are then checked unchanged on locked Weeks 13–18. It ranks repeatability using sample size, Wilson bounds, weekly stability, and holdout confirmation. This reader remains winner-direction-only and makes no ROI claim; Phase 5E market-price evidence is audited separately, and totals remain excluded here.

Phase 5B fingerprint reader has not been run yet.

The additive +16 totals overlay is retired and cannot run or create new live snapshots. Its old report and forward record remain read-only evidence. Totals Core v2 must pass the locked 2025 resimulation and a separate historical market-profitability audit before promotion.

Phase 5D totals electrification has not been run yet.

Shows recurring 2025 winner-direction signal combinations that survived a discovery → locked-confirmation split. These are evidence filters for 2026 forward testing, not guarantees or historical ROI claims.

No winning-signal report loaded.

Tracks only canonical Phase 5C pregame snapshots. Final scores may grade an existing snapshot, but cannot create or redefine one after kickoff.

No 2026 Phase 5C forward-test record loaded.

Existing immutable Phase 5D snapshots remain readable for comparison, but the +16 overlay cannot generate new selections. It is not profitability evidence for Totals Core v2.

No legacy Phase 5D totals record loaded.

Loads the candidate produced by the autonomous 2025 resim. Platt ML calibration and affine margin calibration are fit only on Weeks 2–12 and judged on untouched Weeks 13–18 with aggregate + weekly-stability gates and reliability bins. This card never writes production calfit.

No Phase 5A calibration candidate loaded.

Applies the exact frozen 2025 REPEATABLE fingerprints to the current 2026 Brain winner direction. The first eligible pregame snapshot for each game becomes the canonical forward-test record; after kickoff it is never backfilled or replaced. No 2026 refit, no totals, and no market ROI claims.

Phase 5C has not been run yet.

Uses the frozen 2025 discovery-selected correction plus repeatable residual fingerprints, compares against the current market total, and freezes the first eligible pregame snapshot. Strong labels require at least a 2.5-point corrected edge plus agreeing residual-fingerprint evidence.

Phase 5D live totals has not been run yet.

The Worker never auto-claims human review. Build the proposal only after the v1.7 pipeline completes. Review the exact hash/count, explicitly approve that immutable proposal, then freeze it into the locked test set. Any later semantic or raw-text drift breaks certification.

No v1.7 gold proposal built yet.