← Build site

Scoreboard

Every game the system has predicted, and how close it got. Updated Tue Nov 17, 2026 10:44 AM MT.

How to read this page — every prediction has two labels: what it learned from, and whether it was made before kickoff or after the game (a backtest). Tap for the full key.
made before kickoff
A real forecast: the number existed before the game started.
BACKTEST: made after the game
A test, not a forecast. The model was trained only on older seasons and then asked about games that had already been played. It shows how the method would have done — it was never available before those games.
trained on 2023 and earlier
The newest season the model was allowed to learn from.
S, SB, SQ, SBQ
Versions of the same model that see different evidence. S = plays only; B adds team decisions (depth chart, injury status); Q adds news and talk (beat writers, press conferences).
World model (simulator)
Plays each game out snap by snap thousands of times and reports what happened across those runs: the share it won (“% to win”), the average and median final margin, and the average total. For 2026 the forecast is the one sealed on a Tuesday before the game; for 2024–2025 it is a backtest from the walk-forward run. Every row names its kernel (the frozen version of the play model), the seasons it learned from, and a PW-143 warning: those kernels learned from team ratings that averaged in later games of the same season.
v0.1
The 2026 game forecast the system publishes. It has no fixed training cutoff: it updates on every final score as the season goes.
Market line
What the betting market expected, as the middle value across sportsbooks at the last time we saw the line before kickoff. It is the bar to beat. For 2024–2025 we only have a snapshot every day or two, so that line is usually about a day before kickoff, not the final closing line.
Average miss
How many points off a prediction was, on average, ignoring whether it was too high or too low. Lower is better.
95% range
If the season had gone a bit differently, the number would very likely land inside this range. A wide range means there are too few games to be sure. (Worked out by re-drawing whole weeks at random 2,000 times. With fewer than 5 weeks of games there is no range.)
Picked the winner
Share of games where the team predicted to win did win. Ties and exact pick’ems are left out.

How accurate

Live path vs World model: same games, same players

Two different methods, graded on exactly the same rows: only games and player-games that both forecast before kickoff and that have been played. Live path = the published v0.1 game forecast and the live player model. World model = the simulator's Tuesday issuance, its game scores and the player stat lines it credits inside the same simulated games. "World ahead by" is the row-by-row difference (positive = world model closer); the 95% range re-draws whole weeks 2,000 times and needs 5 weeks. Updated Tue Nov 17, 10:43 AM MT.

week 4: 16 games, 410 player-gamesweek 5: 15 games, 368 player-gamesweek 6: 14 games, 342 player-gamesweek 7: 14 games, 352 player-gamesweek 8: 14 games, 364 player-gamesweek 9: 15 games, 377 player-gamesweek 10: 14 games, 356 player-games

Shared played games: 102 (live only 15, world only 0). Shared player-games: 2570 (live only 1958, world only 71).

ARI 24 @ NYG 36: live +3.4, world -1.1 (home margin)ATL 45 @ NO 24: live +3.1, world +1.3 (home margin)DAL 34 @ HOU 30: live +6.0, world -0.7 (home margin)DEN 14 @ SF 24: live +5.0, world +2.9 (home margin)DET 26 @ CAR 32: live -2.7, world -2.9 (home margin)GB 17 @ TB 14: live +1.6, world -0.6 (home margin)IND 30 @ WAS 13: live -0.4, world +0.9 (home margin)JAX 22 @ CIN 17: live -5.2, world -3.2 (home margin)KC 30 @ LV 27: live -4.4, world -4.0 (home margin)LAC 23 @ SEA 30: live +12.5, world +7.1 (home margin)LA 24 @ PHI 20: live -2.8, world -2.6 (home margin)MIA 10 @ MIN 15: live +13.6, world +2.9 (home margin)NE 29 @ BUF 26: live +4.3, world +3.2 (home margin)NYJ 12 @ CHI 23: live +14.0, world +3.8 (home margin)PIT 24 @ CLE 27: live -2.1, world +0.6 (home margin)TEN 18 @ BAL 24: live +13.9, world +8.7 (home margin)BAL 30 @ ATL 23: live +1.8, world +2.1 (home margin)BUF 37 @ LA 34: live +3.4, world +2.7 (home margin)CHI 7 @ GB 31: live -2.1, world +1.5 (home margin)CIN 31 @ MIA 24: live -6.3, world -4.6 (home margin)CLE 12 @ NYJ 17: live +1.4, world -1.2 (home margin)DEN 23 @ LAC 17: live -3.1, world -0.6 (home margin)DET 45 @ ARI 21: live -5.1, world -2.7 (home margin)HOU 26 @ TEN 3: live -6.7, world -4.5 (home margin)IND 22 @ PIT 34: live +2.1, world +1.4 (home margin)LV 17 @ NE 45: live +5.5, world +4.8 (home margin)MIN 23 @ NO 13: live -3.6, world -3.1 (home margin)NYG 27 @ WAS 27: live +2.3, world +1.1 (home margin)PHI 17 @ JAX 19: live +8.2, world +7.8 (home margin)SF 32 @ SEA 29: live +3.1, world +2.6 (home margin)TB 24 @ DAL 16: live +3.6, world +4.1 (home margin)ARI 18 @ LA 16: live +13.0, world +8.0 (home margin)BAL 10 @ CLE 13: live -3.7, world -4.8 (home margin)BUF 33 @ LV 30: live -5.8, world -4.4 (home margin)CAR 25 @ PHI 14: live -0.7, world -0.2 (home margin)CHI 31 @ ATL 23: live -1.1, world +2.4 (home margin)DAL 13 @ GB 23: live +2.4, world +1.2 (home margin)HOU 34 @ JAX 31: live +4.1, world +4.8 (home margin)LAC 6 @ KC 22: live +7.4, world +5.7 (home margin)NO 34 @ NYG 27: live +0.8, world +2.1 (home margin)NYJ 19 @ NE 26: live +9.3, world +9.3 (home margin)PIT 26 @ TB 19: live -1.8, world +1.9 (home margin)SEA 14 @ DEN 17: live -2.3, world -2.7 (home margin)TEN 26 @ IND 20: live +8.1, world +3.5 (home margin)WAS 17 @ SF 23: live +9.4, world +6.8 (home margin)CIN 35 @ BAL 13: live +0.4, world +2.5 (home margin)CLE 27 @ TEN 24: live -2.0, world -1.2 (home margin)DAL 38 @ PHI 21: live -1.8, world -1.7 (home margin)DEN 30 @ ARI 16: live -5.0, world -0.7 (home margin)GB 17 @ DET 20: live +3.8, world +2.2 (home margin)IND 10 @ MIN 16: live +7.0, world +4.7 (home margin)KC 3 @ SEA 20: live +3.7, world -0.1 (home margin)LA 35 @ LV 21: live -4.1, world -3.5 (home margin)MIA 28 @ NYJ 25: live +3.9, world +2.7 (home margin)NE 24 @ CHI 30: live +2.0, world +0.2 (home margin)NYG 27 @ HOU 21: live +9.9, world +6.1 (home margin)PIT 16 @ NO 10: live -2.9, world -0.8 (home margin)SF 20 @ ATL 29: live -3.4, world -2.7 (home margin)TB 43 @ CAR 13: live +3.4, world +1.2 (home margin)ARI 33 @ DAL 30: live +7.0, world +6.3 (home margin)ATL 0 @ TB 16: live +2.9, world +2.0 (home margin)BAL 10 @ BUF 36: live +4.0, world +0.4 (home margin)CAR 19 @ GB 27: live +4.6, world +4.7 (home margin)CHI 34 @ SEA 17: live +4.0, world +2.4 (home margin)CLE 6 @ PIT 27: live +4.7, world +2.2 (home margin)IND 7 @ JAX 42: live +10.1, world +8.1 (home margin)KC 24 @ DEN 28: live +0.4, world -1.6 (home margin)LAC 9 @ LA 24: live +7.4, world +6.8 (home margin)LV 38 @ NYJ 33: live +1.3, world +1.5 (home margin)MIN 15 @ DET 19: live +2.7, world +1.8 (home margin)NE 24 @ MIA 19: live -7.3, world -4.8 (home margin)PHI 10 @ WAS 27: live +3.9, world +2.8 (home margin)TEN 36 @ CIN 43: live +12.8, world +7.9 (home margin)ARI 3 @ SEA 17: live +9.8, world +6.8 (home margin)BUF 23 @ MIN 6: live -2.2, world -2.9 (home margin)CIN 34 @ ATL 17: live -2.8, world -1.4 (home margin)CLE 22 @ NO 23: live +3.1, world +0.4 (home margin)DAL 27 @ IND 23: live -2.9, world -4.2 (home margin)DEN 26 @ CAR 21: live -3.1, world -1.7 (home margin)DET 19 @ MIA 9: live -6.6, world -4.4 (home margin)GB 45 @ NE 27: live +3.6, world +2.9 (home margin)HOU 44 @ LAC 22: live -5.2, world -4.4 (home margin)JAX 33 @ BAL 10: live -9.1, world -6.3 (home margin)LA 20 @ WAS 13: live -5.0, world -5.5 (home margin)LV 17 @ SF 38: live +10.4, world +6.8 (home margin)NYG 19 @ PHI 11: live -1.8, world -2.2 (home margin)NYJ 12 @ KC 35: live +9.9, world +5.2 (home margin)TB 24 @ CHI 27: live +5.8, world +4.5 (home margin)BUF 31 @ NYJ 12: live -9.6, world -8.8 (home margin)CAR 20 @ NO 17: live +2.0, world -0.1 (home margin)HOU 24 @ CLE 27: live -4.9, world -7.2 (home margin)JAX 40 @ TEN 23: live -11.9, world -10.0 (home margin)KC 35 @ ATL 13: live -3.2, world -2.6 (home margin)LAC 23 @ BAL 20: live +4.7, world +3.5 (home margin)LA 34 @ ARI 6: live -9.0, world -7.2 (home margin)MIA 22 @ IND 24: live +4.3, world +0.7 (home margin)MIN 30 @ GB 10: live +2.9, world +5.3 (home margin)NE 10 @ DET 35: live +2.7, world +2.5 (home margin)PIT 18 @ CIN 38: live +3.6, world +2.9 (home margin)SEA 21 @ LV 13: live -7.3, world -5.0 (home margin)SF 46 @ DAL 40: live -4.2, world -3.2 (home margin)WAS 24 @ NYG 21: live +3.1, world +3.0 (home margin)

Games

MetricLive pathWorld modelWorld ahead byVerdict
Picked the winner101 games69%67%-2 pts95%: -7 to 3No clear winner yet: Live path ahead by 2 points of winner %, but the 95% range includes no difference.
Margin, average miss102 games9.2 pts9.5 pts-0.3 pts95%: -0.7 to 0.3No clear winner yet: Live path ahead by 0.3 pts of average miss, but the 95% range includes no difference.
Total, average miss102 games10.4 pts10.6 pts-0.2 pts95%: -0.5 to 0.2No clear winner yet: Live path ahead by 0.2 pts of average miss, but the 95% range includes no difference.
Log loss on the winnerlower is better0.6000.632-0.03295%: -0.038 to -0.027Live path ahead by 0.032 in log loss; the 95% range excludes no difference.
$100 on the spread102 bets each, early line at -110$309$-645$-9.3695%: $-36.65 to $11.45No clear winner yet: Live path ahead by $9.36 per $100 bet, but the 95% range includes no difference.
$100 on the total102 bets each, early line at -110$973$-1127$-20.5995%: $-38.92 to $1.87No clear winner yet: Live path ahead by $20.59 per $100 bet, but the 95% range includes no difference.

Players

MetricLive pathWorld modelWorld ahead byVerdict
Fantasy points, average miss2569 player-games4.014.02-0.0195%: -0.18 to 0.15No clear winner yet: Live path ahead by 0.01 pts of average miss, but the 95% range includes no difference.
Fantasy points, average miss · QB219 player-games6.046.44-0.3995%: -1.15 to 0.30No clear winner yet: Live path ahead by 0.39 pts of average miss, but the 95% range includes no difference.
Fantasy points, average miss · RB688 player-games4.234.220.0195%: -0.23 to 0.26No clear winner yet: World model ahead by 0.01 pts of average miss, but the 95% range includes no difference.
Fantasy points, average miss · WR1047 player-games3.933.920.0295%: -0.06 to 0.10No clear winner yet: World model ahead by 0.02 pts of average miss, but the 95% range includes no difference.
Fantasy points, average miss · TE615 player-games3.183.130.0595%: -0.19 to 0.31No clear winner yet: World model ahead by 0.05 pts of average miss, but the 95% range includes no difference.
Actual inside the 10–90 range2569 player-games85%86%target 80%Live path closer to the 80% a 10-90 range should hold (live 85%, world 86% on 2569 shared rows).
Targets, average miss1.441.400.0395%: -0.03 to 0.11No clear winner yet: World model ahead by 0.03 of average miss, but the 95% range includes no difference.
Catches, average miss1.081.060.0295%: -0.03 to 0.06No clear winner yet: World model ahead by 0.02 of average miss, but the 95% range includes no difference.
Receiving yards, average miss13.8513.88-0.0395%: -0.53 to 0.49No clear winner yet: Live path ahead by 0.03 of average miss, but the 95% range includes no difference.
Carries, average miss1.051.08-0.0395%: -0.11 to 0.03No clear winner yet: Live path ahead by 0.03 of average miss, but the 95% range includes no difference.
Rushing yards, average miss5.906.05-0.1595%: -0.39 to 0.08No clear winner yet: Live path ahead by 0.15 of average miss, but the 95% range includes no difference.
Passing yards, average miss5.375.170.1995%: -0.54 to 0.77No clear winner yet: World model ahead by 0.19 of average miss, but the 95% range includes no difference.
Passing TDs, average miss0.070.07-0.0095%: -0.01 to 0.00No clear winner yet: Live path ahead by 0.00 of average miss, but the 95% range includes no difference.
Rushing TDs, average miss0.100.10-0.0095%: -0.00 to 0.00No clear winner yet: Live path ahead by 0.00 of average miss, but the 95% range includes no difference.
Receiving TDs, average miss0.180.180.0095%: -0.01 to 0.01No clear winner yet: World model ahead by 0.00 of average miss, but the 95% range includes no difference.
Interceptions, average miss0.060.050.0095%: -0.00 to 0.00No clear winner yet: World model ahead by 0.00 of average miss, but the 95% range includes no difference.
Fumbles lost, average miss0.060.06-0.0095%: -0.00 to 0.00No clear winner yet: Live path ahead by 0.00 of average miss, but the 95% range includes no difference.

What the late information adds: weekly run vs last forecast before kickoff

Each path graded against itself on the same played games: the world model's frozen Tuesday issuance vs its last re-issue before kickoff (daily 7:30 PM MT and at kickoff −80 / −30 min), and the live path as of the Wednesday-noon weekly lock (its Tuesday full run, v0.1's Wednesday lock) vs its last refresh before kickoff. Only rows with a newer forecast count; positive = the later forecast was closer.

Path · metricWeekly runLast before kickoffLater ahead byVerdict
World model: Tuesday issuance vs last before kickoff86 of 87 played games and 2038 of 2071 player-games had a later forecast
Margin, average miss9.49.40.095%: -0.1 to 0.2The later forecast was closer by 0.0 pts of average miss, but the 95% range includes no difference.
Total, average miss10.610.50.195%: 0.0 to 0.2The later forecast was closer by 0.1 pts of average miss; the 95% range excludes no difference.
Log loss on the winner——0.00395%: -0.004 to 0.015The later forecast was closer by 0.003 in log loss, but the 95% range includes no difference.
Fantasy points, average miss4.634.080.5695%: 0.50 to 0.62The later forecast was closer by 0.56 pts of average miss; the 95% range excludes no difference.
Live path: forecast as of the Wednesday-noon weekly lock vs last before kickoff102 of 102 played games and 4283 of 4458 player-games had a later forecast
Margin, average miss9.09.2-0.295%: -0.7 to 0.1The weekly run was closer by 0.2 pts of average miss, but the 95% range includes no difference.
Total, average miss10.510.40.195%: -0.1 to 0.2The later forecast was closer by 0.1 pts of average miss, but the 95% range includes no difference.
Log loss on the winner——-0.00395%: -0.018 to 0.007The weekly run was closer by 0.003 in log loss, but the 95% range includes no difference.
Fantasy points, average miss3.683.350.3395%: 0.25 to 0.42The later forecast was closer by 0.33 pts of average miss; the 95% range excludes no difference.

World model player lines carry known simulator issues (PW-144), labelled on every row and not fixed here: a target is credited only on a completion (world targets = catches); simulated pass volume runs about 5 attempts per team-game high; the QB carry share is checked on every issuance; overtime credits are unverified. Live path win chances are the normal curve implied by v0.1's own margin and spread of outcomes. Contract: docs/contracts/prediction-contract.md.

The prediction matrix: three models × the same research

Each model (player, game, world) runs a stats-only version and versions with the behavioural (B) and language (Q) items added, on the same games at the same time. Champion = stats plus every item that has passed the promotion gate; Frozen S never changes, so Champion vs Frozen S is research progress. A gap counts only on rows two or more cells forecast before kickoff; "closer" means a smaller average miss than the model's own Champion, and the 95% range re-draws whole weeks (needs 5). Updated Tue Nov 17, 10:43 AM MT.

Champion today: stats only (S) in all three models · 0 of 17 items promotedgraded so far: 102 games, 3430 player-games

Frozen S
Champion
Champion + B
Champion + Q
Champion + all
Player model
Frozen Splayers graded3374 · miss 3.27 ptssame rows as Champion7261 forecasts on file · arm S
Championplayers graded3374 · miss 3.27 ptssame rows as Frozen S7261 forecasts on file · arm S
Champion + Bplayers graded3374 · miss 3.27 ptsvs Champion0.01 pts closer [-0.01, +0.04] · within noise · 3374 rows7261 forecasts on file · arm SB
Champion + Qplayers graded3374 · miss 3.27 ptsvs Champion0.01 pts closer [-0.02, +0.04] · within noise · 3374 rows7261 forecasts on file · arm SQ
Champion + allplayers graded3374 · miss 3.26 ptsvs Champion0.01 pts closer [-0.02, +0.06] · within noise · 3374 rows7261 forecasts on file · arm SBQ
Game model
Frozen Sgames graded102 · margin miss 9.2same rows as Champion239 forecasts on file · arm S
Championgames graded102 · margin miss 9.2same rows as Frozen S239 forecasts on file · arm S
Champion + Bgames graded72 · margin miss 9.7vs Champion0.01 pts closer [-0.00, +0.02] · within noise · 72 rows72 forecasts on file · arm SB
Champion + Qnot builtno B/Q version of the game model exists yet
Champion + allnot builtno B/Q version of the game model exists yet
World model
Frozen Splayers graded2083 · miss 4.60 ptsgames graded87 · margin miss 9.4same rows as Champion2430 forecasts on file · arm S
Championplayers graded2083 · miss 4.60 ptsgames graded87 · margin miss 9.4same rows as Frozen S2430 forecasts on file · arm S
Champion + Bgames graded72 · margin miss 9.4vs Champion0.13 pts further off [-0.54, +0.27] · within noise · 72 rows72 forecasts on file · arm SB
Champion + Qnot builtno B/Q input reaches the simulator yet
Champion + allnot builtno B/Q input reaches the simulator yet

Player arms on held-out seasons (backtest)

Fitted on 2023 and earlier, scored on every 2024–26 player-week it never saw. RMSE removed vs stats only; positive = better. The live cells above replace this as weeks are graded.

Pipeline check: the served release vs its own refit (Champion + all): 0.02 pts closer [-0.05, +0.10] · within noise · 3373 rows. Near zero means the arm gaps above are arm effects, not a refit effect.

Research progress: Champion vs Frozen S

Player model: flat at zero by construction — no item has passed the gate, so the Champion is still the stats-only model.

Game model: flat at zero by construction — no item has passed the gate, so the Champion is still the stats-only model.

World model: flat at zero by construction — no item has passed the gate, so the Champion is still the stats-only model.

Does each item transfer?

An item is kept only if it helps every model its mechanism should reach. Helping just one model is flagged as possible overfit.

ItemShould reachPlayerGameWorldVerdict
depth_chart_rank B · candidateplayer, game, world+0.01 [-0.01, +0.04]no cellno cellTransfer untested: only 1 of the 3 models its mechanism should reach has a graded cell carrying it.
depth_delta_1 B · candidateplayer, world+0.01 [-0.01, +0.04]–no cellTransfer untested: only 1 of the 2 models its mechanism should reach has a graded cell carrying it.
depth_delta_3 B · candidateplayer, world+0.01 [-0.01, +0.04]–no cellTransfer untested: only 1 of the 2 models its mechanism should reach has a graded cell carrying it.
weeks_at_rank B · candidateplayer, world+0.01 [-0.01, +0.04]–no cellTransfer untested: only 1 of the 2 models its mechanism should reach has a graded cell carrying it.
depth_best_4 B · candidateplayer, world+0.01 [-0.01, +0.04]–no cellTransfer untested: only 1 of the 2 models its mechanism should reach has a graded cell carrying it.
rank_disagree B · candidateplayer, world+0.01 [-0.01, +0.04]–no cellTransfer untested: only 1 of the 2 models its mechanism should reach has a graded cell carrying it.
chart_buries_starter B · candidateplayer, world+0.01 [-0.01, +0.04]–no cellTransfer untested: only 1 of the 2 models its mechanism should reach has a graded cell carrying it.
coach_pos_tgt B · candidateplayer, world+0.01 [-0.01, +0.04]–no cellTransfer untested: only 1 of the 2 models its mechanism should reach has a graded cell carrying it.
coach_pos_car B · candidateplayer, world+0.01 [-0.01, +0.04]–no cellTransfer untested: only 1 of the 2 models its mechanism should reach has a graded cell carrying it.
inj_status B · candidateplayer, game, world+0.01 [-0.01, +0.04]no cellno cellTransfer untested: only 1 of the 3 models its mechanism should reach has a graded cell carrying it.
qb_start B · candidateplayer, game, world+0.00 [-0.01, +0.01]+0.01 [-0.01, +0.02]-0.16 [-0.45, +0.11]Did not transfer: gain in player of 3 reached (held-out, pre-registered) ⚑
qb_start_v2 B · candidateplayer, game, world+0.00 [-0.00, +0.01]+0.01 [-0.00, +0.02]-0.13 [-0.54, +0.27]Did not transfer: gain in player of 3 reached (held-out, pre-registered) ⚑
q_not_injury Q · candidateplayer, game, world+0.01 [-0.02, +0.04]no cellno cellTransfer untested: only 1 of the 3 models its mechanism should reach has a graded cell carrying it.
q_part_risk Q · candidateplayer, game, world+0.01 [-0.02, +0.04]no cellno cellTransfer untested: only 1 of the 3 models its mechanism should reach has a graded cell carrying it.
q_source_delta Q · candidateplayer, game, world+0.01 [-0.02, +0.04]no cellno cellTransfer untested: only 1 of the 3 models its mechanism should reach has a graded cell carrying it.
q_parts Q · candidateplayer, world+0.01 [-0.02, +0.04]–no cellTransfer untested: only 1 of the 2 models its mechanism should reach has a graded cell carrying it.
q_words Q · candidateplayer, world+0.01 [-0.02, +0.04]–no cellTransfer untested: only 1 of the 2 models its mechanism should reach has a graded cell carrying it.

Promotion gate

Code, not judgement: held-out gain over the floor with a 95% range above zero, a placebo that comes back null, a planted signal that is found, and the same-sign gain in every model the item should reach. Today: 0 promoted. Failing checks: heldout (15 items); placebo (11 items); plant (10 items); transfer (17 items).

Promotion / demotion log: nothing promoted or demoted yet.

Research changes

Player props: accuracy per model

Every player prop captured before kickoff (William Hill first, else the median regulated US sportsbook, else offshore US books; a line must be under 7 days old), priced by each model and taken on the side the model thinks more likely, $100 at the book's price. W-L-P leaves out voids (the player did not play: no action). "Model − market" is the average gap in the chance of the OVER on the same props: negative means the model leans under. Brier on the over: lower is better, model / market. Backtest rows were rebuilt after the games from lines and forecasts that existed before kickoff. Updated Tue Nov 17, 10:08 AM MT.

ModelPropsW-L-PHitP&LROIOver / under picksHit over / underModel − marketBrier
The Modellogged before kickoff66333892-2524-0 (113 void)104 pending60.7%−$21,234-4.7%1918 / 471545.2% / 67.0%-0.8 pts0.2481 / 0.2475
The Modelrebuilt after the games753358-382-0 (13 void)0 pending48.4%−$9,812-13.3%397 / 35649.2% / 47.4%-1.4 pts0.2571 / 0.2489
The Simulatorlogged before kickoff65273955-2365-0 (101 void)106 pending62.6%−$11,855-2.7%1684 / 484350.6% / 66.8%+0.9 pts0.2557 / 0.2478
The Simulatorrebuilt after the games6041-19-00 pending68.3%+$1,46824.9%29 / 3175.9% / 61.3%+1.7 pts0.2069 / 0.2504

Hit rate by market (live and backtest)

MarketThe ModelThe Simulator
Receiving yds48.5%706-751 · −$13,35750.1%622-619 · −$7,088
Receptions56.4%812-629 · −$5,51256.0%692-544 · −$4,458
Rushing yds52.4%351-319 · −$1,08955.1%328-267 · +$2,338
Rush attempts53.8%221-190 · −$98957.3%209-156 · +$1,963
Rush + rec yds48.9%140-146 · −$2,31952.6%154-139 · −$369
Passing yds49.8%112-113 · −$1,42352.2%105-96 · −$313
Passing TDs60.7%133-86 · −$45260.2%121-80 · −$194
Interceptions53.8%98-84 · −$95150.2%100-99 · −$2,168
Anytime TD74.0%1677-588 · −$4,95481.3%1665-384 · −$99

Rule v1 (raw negative binomial) vs v2 (role hurdle + rolling calibration + market shrink), same settled props

Calibration fitted only on earlier weeks: model:11: n=7156, weight on the model 0.45; sim:11: n=6380, weight on the model 0.50.

ModelPropsBrier v1Brier v2Brier marketW-L v1W-L v2
The Model7300.29530.25740.2490361-369353-377
The Simulator420.20290.20290.251730-1230-12

Calibration: how often the side won, by the model's own chance (settled, pushes and voids left out)

Model's chanceThe ModelThe Simulator
50–60%50.6%said 52.2% · n=517053.9%said 52.9% · n=3897
60–70%62.4%said 64.2% · n=31959.8%said 64.8% · n=498
70–80%72.8%said 75.5% · n=39469.4%said 75.2% · n=602
80–90%88.0%said 85.6% · n=75879.8%said 85.3% · n=667
90–100%93.6%said 95.3% · n=51590.4%said 94.9% · n=716

Games

Tap a game to see each player's predicted and actual fantasy points.