Vivid

Four model families, one ledger.

Backtests exclude 2026. When Core-4 opposes Punt Rate's side, there's no play (14-10, 58.3% in the backtest, n=24).

Restated Oct 8, 2026: UOTW (old System 2, 1 play, 0-1, −2.2u) and Core-6 standalone (old System 3, 11 plays, 5-5-1, −0.5u) were retired and removed. Punt Rate moved from System 4 to System 2, so any System 2 row before Oct 8 means UOTW. The previous ledger of 29 plays, +0.5u, reconciles: +3.2u live − 2.7u removed.

System 1 was rebuilt on archived forecast data (2021–2025, 5 seasons); its 113-58 isn't comparable to the old figure. 2026 System 1 plays are qualified on the archived forecast.

Overlap rule: when two systems land on the same game in the same market, it's one bet at the larger system's size and the better number. Different markets (a side and a total) are two separate bets, since they're independent (correlation −0.012 across 584 games).

Rebuilt Oct 3, 2026; the old systems are retired. Blind validation (parameters picked on 2013–19, applied cold to 2020–25): System 1 62.2%, System 2 60.1%, System 3 ~55%. Qualification is read at the closing line.

Ledger = placed bets, all logged at -110 (switched Oct 10). The 1u System 1 bets in Weeks 1-4 and the 2u System 2 bets in Weeks 3-4 are legacy sizes from before the Oct 3 rebuild.

System 3 rule as of Week 6: conference game, home favored by 10+, and in Weeks 1-6 a total of 52 or less (61.2%); in Weeks 7+ a total of 48 or less with a ranked home favorite (62.4%).

Regular season only, 2020-21 excluded, graded at the closing spread, pushes excluded. No picks in conference tournaments or the NCAA tournament; none of the systems were tested there.

Big House, Thin Air and Grinders all bet the home team, so when they overlap it's one bet at the larger size. Combined card: 0 losing seasons in 18. Grinders is probationary, with a review in January.