NFL Week 6: Commanders vs 49ers — Model Analysis

Game: Commanders vs 49ers Posted: October 17, 2026 · VAR Research Week: 6 · 2026-27 season Kickoff: Tue Oct 20 2026, 00:15 UTC Venue: Levi’s Stadium Tier signal: PRIME spread (|edge| ≥ 6, pre-registered) Model notes (not a filter): internal label FADE; flags qb_injury_warn, high_mc_var_boost, high_meta_confidence, high_meta_ultra Pre-registration: /methodology/2026-27-predictions

PRIME-tier picks

Model context

Win-probability projection: 49ers 90.5% vs Commanders 9.5%. Projected total: 45.9 points.

How to read this

Projections come from VAR’s production NFL ensemble. The PRIME-tier filter is the model’s pre-registered high-confidence threshold: PRIME spread fires at |edge| >= 6 points (the model’s spread differs from the market spread by at least 6 points; nothing else gates it). The PRIME_TOT totals tier (|edge| >= 7) was retired on 2026-09-24; picks published before that date remain in the record.

Filter correction 2026-10-02: publication filter corrected to the pre-registered |edge| ≥ 6; before this date only the internal ‘road favorite’ subset was posted. Games that qualified and were not posted are counted as unpublished signals on /performance, not added to the record after the fact.

Validation status: the backtest figures originally cited for these tiers (April 2026) were withdrawn on 2026-05-17 after an audit found feature lookahead in the harness. The leak-free recomputation grades them at break-even or below (spread |edge| >= 6: 51.1% against the close, 95% CI [45.2%, 57.0%]; PRIME_TOT 44.3% straight-up). These posts are a pre-registered forward test of that signal, not a claim of edge. Live forward tracking: /performance. Full correction notice: /insights/nfl-2026-27-pre-registration-correction.

The 2026-27 NFL season is VAR’s first publicly-verifiable forward test of the PRIME spread tier (and, until 2026-09-24, PRIME_TOT). Success and failure criteria for the season are pre-registered at /methodology/2026-27-predictions.

Reminders

• Lines are recorded at the time of publication. Closing-line edge may differ.
• This is research output, not financial advice. VAR does not place bets on behalf of readers.
• Postmortems for any model miss are published within 24 hours under /insights.