NFL 2026 Week 7 — PROP RESEARCH CHALLENGER — October 19, 2026

Open Week7_Prop_Research.csv in Excel. The lines are MODEL-GENERATED reference thresholds, NOT sportsbook quotes. Fair odds describe the simulator, not proven betting value. No real overlay/underlay or recommended wager is asserted. All decisions remain HOLD.

Week7_Prop_Summaries.csv contains modeled means, medians, and central 80% simulation ranges. These ranges are NOT calibrated prediction intervals. Missing simulation coverage uses zero placeholders clearly flagged; exclude those rows from pricing.

Model: existing event challenger, deterministic seed 20261019; 5,000 shared game simulations per matchup. Drives and plays use Poisson counts, shared pace uses a gamma distribution, allocations use gamma/Dirichlet-style weights and binomial sampling, yardage uses gamma draws. Quarterback passing and receiver receiving yards/TDs are linked. Whole-yard rounding is applied for settlement comparisons. Scoring touchdowns EXCLUDES passing touchdowns. Kneeldown/negative-yard tails and efficiency tails remain approximate. No new neural-network model was trained.

Personnel is the existing base scenario, not a verified final Week 7 lineup. No claimed fresh weather, travel, or injury update. Return/limited/inactive scenario probabilities are not yet validated or mixed. Do NOT interpret an inactive placeholder as an under bet: book void/participation rules control settlement.

To evaluate actual prices, fill quotes.csv with one row per book/line and both offered American odds at the SAME line and timestamp. Use exact player_id and market from the research CSV. Include source_url and settlement_rules. With Python and numpy installed, run:

python price_quotes.py quotes.csv priced_quotes.csv

Keep the script beside prop_samples.npz. The script computes probabilities at the actual line, including pushes; EV is win probability times profit minus loss probability. No-vig market probabilities normalize over/under implied probabilities and are conditional on a decisive result. Edge is also conditional on no push. An EV above zero is only an UNVALIDATED potential overlay. Negative EV is an underlay for that side. Missing inputs produce an error, not a recommendation. No bankroll sizing or automatic bet placement is included.

Before promotion: walk-forward prop-specific probability calibration on held-out weeks, archived executable book quotes, baseline comparison, return after vig, closing-line comparison, uncertainty sensitivity, coverage/settlement audits, and correlated exposure limits. Current fantasy backtests do not validate prop betting performance. Simulation draws reduce numerical noise; they do not remove model error. Do not count 5,000 draws as 5,000 independent observed games.

Market audit: ScoresAndOdds Chris Olave page was checked October 19, 2026. Search-snippet receiving lines corresponded to historical games on the underlying page. Latest page had consensus touchdown quotes but no verified named-book two-sided quote. No quote was imported.

https://www.scoresandodds.com/prop-bets/3491988/chris-olave-3491988

Original fantasy forecasts and frozen backtest artifacts are unchanged. build_props.py is included as a workflow record; regeneration requires the original project's data and event_simulator.py. price_quotes.py works with the included samples.