Same evidence. Independent predictions. Results tracked before and after kickoff.
Consensus spread across 9 sportsbooks and how far each line has moved this week. Five-model picks post about two hours before each kickoff.
| Kickoff (ET) | Matchup | Spread | Moved | Total | Five-model picks |
|---|---|---|---|---|---|
| Sun 9:30 AM | Eagles at Jaguars | Jaguars -7.5 (books -7.5 to -8) | 1 toward Jaguars | 42 | About 2 h before kickoff |
| Sun 1:00 PM | Bears at Packers | Bears -1.5 (books 0 to -2) | 1.5 toward Packers | 45.5 | About 2 h before kickoff |
| Sun 1:00 PM | Texans at Titans | Texans -7.5 | 0.5 toward Texans | 38 | About 2 h before kickoff |
| Sun 1:00 PM | Bengals at Dolphins | Bengals -6.5 (books -6.5 to -7) | 0.5 toward Dolphins | 42.5 | About 2 h before kickoff |
| Sun 1:00 PM | Raiders at Patriots | Patriots -3.5 | — | 45 | About 2 h before kickoff |
| Sun 1:00 PM | Vikings at Saints | Vikings -2.5 (books -1 to -2.5) | 1 toward Vikings | 42 | About 2 h before kickoff |
| Sun 1:00 PM | Browns at Jets | Jets -2 (books -1 to -2.5) | — | 39.5 | About 2 h before kickoff |
| Sun 1:00 PM | Colts at Steelers | Steelers -2.5 (books -1 to -3) | — | 44 | About 2 h before kickoff |
| Sun 1:00 PM | Giants at Commanders | Commanders -4 (books -3.5 to -4) | 0.5 toward Commanders | 42 | About 2 h before kickoff |
| Sun 4:05 PM | Broncos at Chargers | Broncos -3.5 (books -3 to -3.5) | — | 42 | About 2 h before kickoff |
| Sun 4:25 PM | Lions at Cardinals | Lions -5.5 | 1 toward Lions | 54.5 | About 2 h before kickoff |
| Sun 4:25 PM | 49ers at Seahawks | Seahawks -3 (books -3 to -3.5) | — | 46 | About 2 h before kickoff |
| Sun 8:20 PM | Ravens at Falcons | Falcons -3 (books -3 to -3.5) | 6 toward Falcons | 43.5 | About 2 h before kickoff |
| Mon 8:15 PM | Bills at Rams | Rams -3 (books -3 to -3.5) | — | 54.5 | About 2 h before kickoff |
Grok 4.6 led against the contest line at 11–5. Favorites won 9–7 outright but went 4–10 against the sportsbook spread.
| Model | Contest line | Sportsbook spread | Outright winners |
|---|---|---|---|
| Grok 4.6 | 11–5 | 9–5 (2 push) | 12–4 |
| Claude Fable 5.1 | 11–5 | 7–7 (2 push) | 12–4 |
| GPT-6 Astra | 10–6 | 7–7 (2 push) | 10–6 |
| DeepSeek V4 Pro 0813 | 8–8 | 6–8 (2 push) | 9–7 |
| Gemini 3.8 Flash | 8–8 | 5–9 (2 push) | 10–6 |
Records use each model's first pregame pick at the saved line. One week is a small sample.
Winner ≠ spread. Each has its own rank and record. Cards ordered by outright winners · Season through Week 5. Live records show where picks stand at the current score; season records count final results only.
41 correct · 17 incorrect now
0 tiedSeason final: 41W · 17L · 0T70.7% win rate
35 correct · 23 incorrect now
0 pushesSeason final: 35W · 23L · 0P60.3% win rate
30 correct · 25 incorrect now
3 pushesSeason final: 30W · 25L · 3P54.5% win rate
Awaiting pregame selection
Season final: 0W · 3L · 0 void0.0% win rate
41 correct · 17 incorrect now
0 tiedSeason final: 41W · 17L · 0T70.7% win rate
31 correct · 27 incorrect now
0 pushesSeason final: 31W · 27L · 0P53.4% win rate
30 correct · 25 incorrect now
3 pushesSeason final: 30W · 25L · 3P54.5% win rate
Awaiting pregame selection
Season final: 0W · 3L · 0 void0.0% win rate
39 correct · 19 incorrect now
0 tiedSeason final: 39W · 19L · 0T67.2% win rate
35 correct · 23 incorrect now
0 pushesSeason final: 35W · 23L · 0P60.3% win rate
29 correct · 26 incorrect now
3 pushesSeason final: 29W · 26L · 3P52.7% win rate
Awaiting pregame selection
Season final: 1W · 2L · 0 void33.3% win rate
37 correct · 21 incorrect now
0 tiedSeason final: 37W · 21L · 0T63.8% win rate
29 correct · 29 incorrect now
0 pushesSeason final: 29W · 29L · 0P50.0% win rate
27 correct · 28 incorrect now
3 pushesSeason final: 27W · 28L · 3P49.1% win rate
Awaiting pregame selection
Season final: 0W · 3L · 0 void0.0% win rate
37 correct · 21 incorrect now
0 tiedSeason final: 37W · 21L · 0T63.8% win rate
26 correct · 32 incorrect now
0 pushesSeason final: 26W · 32L · 0P44.8% win rate
26 correct · 29 incorrect now
3 pushesSeason final: 26W · 29L · 3P47.3% win rate
Awaiting pregame selection
Season final: 0W · 3L · 0 void0.0% win rate
Provisional comparison on 1 shared games this week. Season records stay within nfl-evidence-v1; W = wins, L = losses, T = ties, P = pushes. Scheduled games are not counted as winning or losing. A live lead is not a final win or a win probability.
Buccaneers: Baker Mayfield (QB) out
Oct 5, 3:07 PM ETPaste into Claude Web or Codex Web for winners, both spread choices, score totals and an optional three-game parlay. Includes every scheduled day. Independent web reviews are separate from the controlled model comparison.
Winner = who wins the game. Spread = who covers the handicap. Winners below come from each model’s saved predicted score, not a separate moneyline analysis. Week 5 spread agreement uses the contest line when supplied. Agreement is not a measured chance of winning.
| Game / kickoff / actual scoreSeason through Week 5 · Ordered by Outright winners | Claude Fable 5.1 41 correct · 17 incorrect now 0 tiedSeason final: 41W · 17L · 0T70.7% win rate 35 correct · 23 incorrect now 0 pushesSeason final: 35W · 23L · 0P60.3% win rate 30 correct · 25 incorrect now 3 pushesSeason final: 30W · 25L · 3P54.5% win rate Awaiting pregame selection Season final: 0W · 3L · 0 void0.0% win rate | Grok 4.6 41 correct · 17 incorrect now 0 tiedSeason final: 41W · 17L · 0T70.7% win rate 31 correct · 27 incorrect now 0 pushesSeason final: 31W · 27L · 0P53.4% win rate 30 correct · 25 incorrect now 3 pushesSeason final: 30W · 25L · 3P54.5% win rate Awaiting pregame selection Season final: 0W · 3L · 0 void0.0% win rate | GPT-6 Astra 39 correct · 19 incorrect now 0 tiedSeason final: 39W · 19L · 0T67.2% win rate 35 correct · 23 incorrect now 0 pushesSeason final: 35W · 23L · 0P60.3% win rate 29 correct · 26 incorrect now 3 pushesSeason final: 29W · 26L · 3P52.7% win rate Awaiting pregame selection Season final: 1W · 2L · 0 void33.3% win rate | DeepSeek V4 Pro 0813 37 correct · 21 incorrect now 0 tiedSeason final: 37W · 21L · 0T63.8% win rate 29 correct · 29 incorrect now 0 pushesSeason final: 29W · 29L · 0P50.0% win rate 27 correct · 28 incorrect now 3 pushesSeason final: 27W · 28L · 3P49.1% win rate Awaiting pregame selection Season final: 0W · 3L · 0 void0.0% win rate | Gemini 3.8 Flash 37 correct · 21 incorrect now 0 tiedSeason final: 37W · 21L · 0T63.8% win rate 26 correct · 32 incorrect now 0 pushesSeason final: 26W · 32L · 0P44.8% win rate 26 correct · 29 incorrect now 3 pushesSeason final: 26W · 29L · 3P47.3% win rate Awaiting pregame selection Season final: 0W · 3L · 0 void0.0% win rate |
|---|---|---|---|---|---|
Buccaneers @ Cowboys3/5 models agreeGroup contest:  Cowboys -9.5Oct 8, 8:15 PM ETScores loading… |
Tied for most wins: Claude Fable 5.1, DeepSeek V4 Pro 0813, Gemini 3.8 Flash, GPT-6 Astra, Grok 4.60 wins · 1 losses · 0.0% win rate
| Rank | Model | Wins | Losses | Ties | Win rate | Final | Pending | No pick |
|---|---|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.1 | 0 | 1 | 0 | 0.0% | 1 | 0 | 0 |
| 1 | DeepSeek V4 Pro 0813 | 0 | 1 | 0 | 0.0% | 1 | 0 | 0 |
| 1 | Gemini 3.8 Flash | 0 | 1 | 0 | 0.0% | 1 | 0 | 0 |
| 1 | GPT-6 Astra | 0 | 1 | 0 | 0.0% | 1 | 0 | 0 |
| 1 | Grok 4.6 | 0 | 1 | 0 | 0.0% | 1 | 0 | 0 |
Same games, same prompt: 1 shared forecasts per model. Ranked by wins, then win rate; equal scores share a rank. Win rate excludes ties and pushes.
Outright choices come from saved score estimates. A predicted tie is “No pick”; no moneyline profit is inferred. Week 1 is a small sample, not evidence of consistent skill.
Buccaneers @ CowboysOct 8, 8:15 PM ET
Claude Fable 5.1Winner: CowboysFrom predicted score · LOSS
Spread: Buccaneers +9.5
Contest: Buccaneers +9.5
Predicted score: TB 23 – DAL 31
Spread result: win
Quote: DraftKings -112
Oct 8, 6:14 PM ETNews impact: Confirmed/reported: Mayfield out, Daniels starting (report-1, report-2); Bucs rule out Winfield Jr., Dennis and Morrison (injuries); McLaughlin expected to play per Bowles. Dallas activated Tyler Smith from IR (report-3) though he was listed questionable and will wear a cast — his first game action since surgery, so effectiveness is uncertain; Dallas rules out Shelton, Durant and Overshown. Net effect: Tampa's secondary is thinner against the league's No. 2 scoring offense, which supports Dallas scoring plenty; Dallas's defense is thinner at LB/CB, which supports Tampa sustaining drives via Irving and Daniels' legs and keeping the game within the number. Weather is irrelevant at AT&T Stadium (roof status unverified). All of this news is public and likely already in the -9.5 line; the packet includes only a single odds snapshot, so no movement can be inferred. Historical results are not opponent-adjusted and the preseason entries were ignored.
Reasoning & counterargument: Dallas (2-2) hosts an 0-4 Tampa Bay team starting undrafted rookie Jalon Daniels for a second straight game with Baker Mayfield out (dislocated thumb). The packet shows Dallas's offense averaging 31.8 ppg (second in NFL) but a defense that has struggled, with every Dallas game this season decided within one score except the 37-20 Washington win, and Dallas allowing 30+ in two of four. Tampa Bay has lost all four games by 3-7 points and has allowed the fourth-fewest yards per game, with a run game (Bucky Irving) that faces a Dallas defense giving up 97 rush yards per game to RBs, now without LB Overshown and CB Durant. That profile — a high-variance Dallas defense versus a conservative, run-first, close-game Bucs team — makes 9.5 a lot of points, so the forced-choice spread lean is Tampa Bay +9.5, and the same team at the identical contest handicap. Strongest counterargument: the 0-4 team is starting a rookie who threw two INTs last week, loses its best safety (Winfield) and a corner (Morrison) against Dak Prescott, CeeDee Lamb (17 catches last week) and George Pickens, and Dallas gets Pro Bowl LG Tyler Smith back; a double-digit blowout is entirely plausible and the market already prices Mayfield's absence. No edge is demonstrated at the sportsbook price, so this is a pass rather than a betting lean.
Sources: Week 5 TNF highlights: Best plays, moments, touchdowns, from Cowboys vs. Bucs Fantasy football buzz: Start Bucky Irving, sit Emeka Egbuka on Thursday night Cowboys activate Pro Bowl LG Tyler Smith (thumb) from IR injuries The Odds API matchup
Recorded: Oct 8, 6:15 PM ET
Original betting assessment: Pass — no wager recommended. This separate assessment does not cancel the model’s prediction.
Served by Anthropic. Cost: $0.2858.
Input SHA-256: a3ca5a78f0b1583160949d98d8423da5cd0794ae9b6500a6e2e26477502bf1e9
Each model chooses one locked ticket from the same fresh pregame slate once at least three shared games are eligible. Outright moneylines and sportsbook spreads may be mixed; contest lines are excluded. One lost leg loses the ticket. Pushes drop out; three pushes void it. Earlier weeks count final tickets only; live parlay ranks count a ticket as on track only after all three legs have started with no leg behind. Each ticket counts once, separately from its three legs.
Model selections for tracking. No wager is placed and no combined payout is quoted; leg prices may come from different books. Waiting for at least three shared, fresh pregame packets. No selections have been invented.
| Prompt | Week | Model | Winner W–L–T | Winner win rate | Spread W–L–P | Pending | ATS / 95% interval | Paper units | ROI | Contest W–L–P | Contest win rate | Total error | API errors |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| nfl-evidence-v1 | 1 | DeepSeek V4 Pro 0813 | 10–4–00 pending · 0 no pick | 71.4% | 6–8–0 | 0 | 42.9%21.4%–67.4% | -2.56 | -18.3% | 6–8–0 | 42.9% | 10.50 | 3 |
| nfl-evidence-v1 | Season | DeepSeek V4 Pro 0813 | 37–21–00 pending · 0 no pick | 63.8% | 27–28–3 | 0 | 49.1%36.4%–61.9% | -3.55 | -6.1% | 29–29–0 | 50.0% | 10.17 | 14 |
| nfl-evidence-v1 | 1 | GPT-6 Astra | 11–3–00 pending · 0 no pick | 78.6% | 6–8–0 | 0 | 42.9%21.4%–67.4% | -2.64 | -18.9% | 9–5–0 | 64.3% | 10.71 | 0 |
| nfl-evidence-v1 | Season | GPT-6 Astra | 41–22–00 pending · 0 no pick | 65.1% | 32–28–3 | 0 | 53.3%40.9%–65.4% | +0.58 | 0.9% | 38–25–0 | 60.3% | 10.10 | 0 |
| nfl-evidence-v1 | 1 | Claude Fable 5.1 | 11–3–00 pending · 0 no pick | 78.6% | 7–7–0 | 0 | 50.0%26.8%–73.2% | -0.67 | -4.8% | 9–5–0 | 64.3% | 10.64 | 8 |
| nfl-evidence-v1 | Season | Claude Fable 5.1 | 42–21–00 pending · 0 no pick | 66.7% | 34–26–3 | 0 | 56.7%44.1%–68.4% | +4.71 | 7.5% | 39–24–0 | 61.9% | 10.16 | 10 |
| nfl-evidence-v1 | 1 | Gemini 3.8 Flash | 11–3–00 pending · 0 no pick | 78.6% | 7–7–0 | 0 | 50.0%26.8%–73.2% | -0.67 | -4.8% | 7–7–0 | 50.0% | 11.86 | 3 |
| nfl-evidence-v1 | Season | Gemini 3.8 Flash | 39–23–00 pending · 0 no pick | 62.9% | 29–30–3 | 0 | 49.2%36.8%–61.6% | -4.01 | -6.5% | 28–34–0 | 45.2% | 10.45 | 8 |
| nfl-evidence-v1 | 1 | Grok 4.6 | 11–3–00 pending · 0 no pick | 78.6% | 6–8–0 | 0 | 42.9%21.4%–67.4% | -2.6 | -18.6% | 7–7–0 | 50.0% | 12.86 | 0 |
| nfl-evidence-v1 | Season | Grok 4.6 | 43–20–00 pending · 0 no pick | 68.3% | 33–27–3 | 0 | 55.0%42.5%–66.9% | +2.78 | 4.4% | 35–28–0 | 55.6% | 10.51 | 0 |
| nfl-evidence-v1 | 2 | DeepSeek V4 Pro 0813 | 8–3–00 pending · 0 no pick | 72.7% | 4–7–0 | 0 | 36.4%15.2%–64.6% | -3.5 | -31.8% | 4–7–0 | 36.4% | 10.55 | 11 |
| nfl-evidence-v1 | 2 | GPT-6 Astra | 10–6–00 pending · 0 no pick | 62.5% | 8–8–0 | 0 | 50.0%28.0%–72.0% | -1 | -6.2% | 8–8–0 | 50.0% | 10.56 | 0 |
| nfl-evidence-v1 | 2 | Claude Fable 5.1 | 9–7–00 pending · 0 no pick | 56.3% | 7–9–0 | 0 | 43.8%23.1%–66.8% | -2.71 | -16.9% | 7–9–0 | 43.8% | 10.13 | 2 |
| nfl-evidence-v1 | 2 | Gemini 3.8 Flash | 8–7–00 pending · 0 no pick | 53.3% | 5–10–0 | 0 | 33.3%15.2%–58.3% | -5.61 | -37.4% | 4–11–0 | 26.7% | 9.47 | 5 |
| nfl-evidence-v1 | 2 | Grok 4.6 | 10–6–00 pending · 0 no pick | 62.5% | 6–10–0 | 0 | 37.5%18.5%–61.4% | -4.63 | -28.9% | 7–9–0 | 43.8% | 10.13 | 0 |
| nfl-evidence-v1 | 3 | DeepSeek V4 Pro 0813 | 10–6–00 pending · 0 no pick | 62.5% | 10–5–1 | 0 | 66.7%41.7%–84.8% | +4.03 | 25.2% | 10–6–0 | 62.5% | 10.44 | 0 |
| nfl-evidence-v1 | 3 | GPT-6 Astra | 10–6–00 pending · 0 no pick | 62.5% | 11–4–1 | 0 | 73.3%48.0%–89.1% | +5.88 | 36.8% | 11–5–0 | 68.8% | 10.75 | 0 |
| nfl-evidence-v1 | 3 | Claude Fable 5.1 | 10–6–00 pending · 0 no pick | 62.5% | 12–3–1 | 0 | 80.0%54.8%–93.0% | +7.75 | 48.4% | 11–5–0 | 68.8% | 11.06 | 0 |
| nfl-evidence-v1 | 3 | Gemini 3.8 Flash | 10–6–00 pending · 0 no pick | 62.5% | 12–3–1 | 0 | 80.0%54.8%–93.0% | +7.75 | 48.4% | 9–7–0 | 56.3% | 11.50 | 0 |
| nfl-evidence-v1 | 3 | Grok 4.6 | 10–6–00 pending · 0 no pick | 62.5% | 12–3–1 | 0 | 80.0%54.8%–93.0% | +7.79 | 48.7% | 10–6–0 | 62.5% | 10.44 | 0 |
| nfl-evidence-v1 | 4 | DeepSeek V4 Pro 0813 | 9–7–00 pending · 0 no pick | 56.3% | 6–8–2 | 0 | 42.9%21.4%–67.4% | -2.42 | -15.1% | 8–8–0 | 50.0% | 9.56 | 0 |
| nfl-evidence-v1 | 4 | GPT-6 Astra | 10–6–00 pending · 0 no pick | 62.5% | 7–7–2 | 0 | 50.0%26.8%–73.2% | -0.66 | -4.1% | 10–6–0 | 62.5% | 8.44 | 0 |
| nfl-evidence-v1 | 4 | Claude Fable 5.1 | 12–4–00 pending · 0 no pick | 75.0% | 7–7–2 | 0 | 50.0%26.8%–73.2% | -0.55 | -3.5% | 11–5–0 | 68.8% | 8.63 | 0 |
| nfl-evidence-v1 | 4 | Gemini 3.8 Flash | 10–6–00 pending · 0 no pick | 62.5% | 5–9–2 | 0 | 35.7%16.3%–61.2% | -4.48 | -28.0% | 8–8–0 | 50.0% | 9.44 | 0 |
| nfl-evidence-v1 | 4 | Grok 4.6 | 12–4–00 pending · 0 no pick | 75.0% | 9–5–2 | 0 | 64.3%38.8%–83.7% | +3.22 | 20.1% | 11–5–0 | 68.8% | 9.25 | 0 |
| nfl-evidence-v1 | 5 | DeepSeek V4 Pro 0813 | 0–1–00 pending · 0 no pick | 0.0% | 1–0–0 | 0 | 100.0%20.7%–100.0% | +0.89 | 89.3% | 1–0–0 | 100.0% | 7.00 | 0 |
| nfl-evidence-v1 | 5 | GPT-6 Astra | 0–1–00 pending · 0 no pick | 0.0% | 0–1–0 | 0 | 0.0%0.0%–79.3% | -1 | -100.0% | 0–1–0 | 0.0% | 10.00 | 0 |
| nfl-evidence-v1 | 5 | Claude Fable 5.1 | 0–1–00 pending · 0 no pick | 0.0% | 1–0–0 | 0 | 100.0%20.7%–100.0% | +0.89 | 89.3% | 1–0–0 | 100.0% | 14.00 | 0 |
| nfl-evidence-v1 | 5 | Gemini 3.8 Flash | 0–1–00 pending · 0 no pick | 0.0% | 0–1–0 | 0 | 0.0%0.0%–79.3% | -1 | -100.0% | 0–1–0 | 0.0% | 5.00 | 0 |
| nfl-evidence-v1 | 5 | Grok 4.6 | 0–1–00 pending · 0 no pick | 0.0% | 0–1–0 | 0 | 0.0%0.0%–79.3% | -1 | -100.0% | 0–1–0 | 0.0% | 5.00 | 0 |
First valid forecast per model, game and prompt version, recorded before kickoff. A later forecast cannot replace an earlier result. Final scores come from ESPN. Spread picks use the saved handicap and American price. Pushes refund the stake and are excluded from ATS hit rate; they remain settled exposures for paper ROI. All recorded picks count in the prediction benchmark. The original prompt also asked whether the model recommended wagering; that separate assessment is preserved in the audit trail and does not remove a prediction from scoring. Contest choices are graded without an invented price. Outright winners are derived from the saved score estimates and graded separately; a tied final game is a tie, while a tied or missing predicted score produces no winner pick. Winner win rate excludes ties and no-picks. A missing contest choice is not a loss.
The leaderboard includes only games where every listed model has an eligible forecast on the same packet. The detailed history includes all eligible forecasts and can have unequal coverage. Season totals stay within the selected prompt version. Prior results are not automatically fed into future prompts. Score error is mean absolute error per team. Total error measures the combined points estimate. The descriptive 95% interval uses the Wilson method; games and predictions can be correlated. Small samples cannot establish consistent skill. More model agreement does not mean a higher win probability.
This first Week 1 run includes the 14 remaining games and excludes the two previously completed games. Future packets are collected in the two hours before kickoff. Models receive the same prompt and evidence for each game, with high reasoning and a 16,384-token output ceiling. Source collection failures and model failures are retained in the audit. A new prompt version starts a separate comparison group.