Accountability

MODEL RECORD

Every published projection is graded against the final score and stays on this page, hit or miss. Two separate things are tracked: B1G+A's pregame game projections and its in-game win-probability model.

Projections: Week 6, posted Oct 10, 2026 · graded from public feeds finals · win-probability metrics from held-out 2025 testing.

B1G+A RECORD

green = .500 or better · red = below .500

	ALL-TIMEsince 10/10/26 · 7 graded	LATEST WEEKWeek 6 · 7 graded · 0 pending
Straight upprojected winner won	4-3.571	4-3.571
Against the spreadB1G+A lean vs. DraftKings	5-2.714	5-2.714
Over/underB1G+A lean vs. DraftKings	3-4.429	3-4.429
Avg margin errorprojected vs. actual margin	8.9pts	8.9pts

Iowa at Washington (Fri., Oct 9, 2026, 8 PM CT) kicked off before B1G+A launched (Oct 10, 2026), so it is not in the pick record.

The B1G+ take: the sample is tiny. Calibration check: B1G+A picks carried an average pregame FPI win chance of 67%, so you'd expect about 9.4 of 14 right; it got 8. With 14 games that is far too small a sample to say the model is well calibrated, and B1G+A will keep tracking it in the open. Straight-up pick percentages also lean on FPI (comparison), which supplies the win chance. The projected score is B1G+A's own blend of the market-implied score and team scoring stats.

Game by game · Week 6

7 games · B1G+A first, FPI (comparison) for comparison · Original post ↗

Game	B1G+A proj.	Final	B1G+A pick	Winner	Spread	Total	Margin err	FPI (comparison) win %	B1G+A live WP
IU @ NEB	IU 28–24	20–17	IU	Hit	NEB +7.5 Hit	OVER 47.5 Miss	1	70%	NEB 24% → low <1% / high 83%RECONSTRUCTED after the game, not logged live: public play-by-play feeds replayed through the same B1G+A model · 172 plays · last play 12%
BALL @ NU	BALL 13–40	14–45	NU	Hit	BALL +35.5 Hit	UNDER 53.5 Miss	4	99%	NU 99% → low 99% / high >99%RECONSTRUCTED after the game, not logged live: public play-by-play feeds replayed through the same B1G+A model · 194 plays · last play >99%
UCLA @ ORE	UCLA 29–34	17–15	ORE	Miss	UCLA +10.5 Hit	OVER 58.5 Miss	7	85%	No live datatracker did not log this game
ILL @ MSU	ILL 26–22	38–37	ILL	Hit	ILL -2.5 Miss	OVER 47.5 Hit	3	54%	No live datatracker did not log this game
MD @ OSU	MD 16–41	16–45	OSU	Hit	MD +33.5 Hit	OVER 56.5 Hit	4	98%	No live datatracker did not log this game
USC @ PSU	USC 27–28	31–23	PSU	Miss	USC +1.5 Hit	OVER 54.5 Miss	9	62%	No live datatracker did not log this game
MINN @ PUR	MINN 30–24	13–41	MINN	Miss	MINN -2.5 Miss	OVER 49.5 Hit	34	53%	No live datatracker did not log this game

FPI (comparison) win % is the public feed's pregame win chance for the B1G+A pick, shown for comparison (not B1G+A). B1G+A live WP key: the home team's win probability from our in-game model. First number = first logged run, then the lowest and highest values logged. "Last logged" is the latest run before the tracker stopped; the tracker can stop before the final whistle and never records finals, so it is not a final result. "No live data" = the tracker didn't log that game. "Reconstructed" = the tracker missed the game, so values were rebuilt afterward by replaying public play-by-play feeds through the same model (first/low/high over every play).

B1G+A projected score averages the Vegas-implied score (from the DraftKings spread and total) with a stat blend: each offense's points per game averaged with the opponent's points allowed per game, using stats through Oct. 3. Win chance from FPI (comparison). Lines from DraftKings via public feeds as of Saturday morning. Projections are estimates, not guarantees. Not projected: Iowa @ Washington (Fri Oct 9; the Week 6 card covered Saturday games only).

In-game win-probability model

tested on all of 2025, held out

Brier (lower is better)0.1242125,614 plays, 764 games

vs cfbfastR built-in WP0.1507same plays: B1G+A 0.1239

vs public feeds live WP0.1254 / 0.1261B1G+A / public feeds · 300 games, 49,907 plays

Brier gap vs public feeds, 95% CI-0.0049 to +0.0035includes zero: a statistical tie

Calibration, 2025

Projected home win probability vs. how often the home team actually won, by decile. On the diagonal = well calibrated.

Brier score by quarter: B1G+A vs public feeds

Same 300 games. B1G+A is slightly better in Q1–Q2, even in Q3–Q4, and worse in overtime (only 95 plays).

The B1G+ take: The win-probability model clearly beats a spread-only baseline (0.177) and cfbfastR's built-in model on the same plays. Against public feeds it's a statistical tie: the 95% interval for the Brier gap spans zero. Overtime is weak because there's so little OT data. The pregame projection record is only 14 games, which proves nothing yet. Every graded game gets added here.