JackpotBetz

Model Backtest

2023–2025 · 6,615 regular-season games · Walk-forward, prior-games-only team stats, prior-season starter ERA

Information only — JackpotBetz is a statistical research tool, not a sportsbook, and accepts no wagers. Backtested performance is historical and does not predict future results.

ML accuracy

55.8%

ML ROI

+4.96%

Run line accuracy

53.5%

Run line ROI

+9.79%

Season by season

Season	Games	ML acc.	ML plays	ML ROI	RL acc.	RL plays	RL ROI
2023	2214	55.4%	741	+4.10%	53.1%	688	+8.60%
2024	2210	56.1%	769	+5.80%	53.8%	702	+10.40%
2025	2191	55.9%	754	+4.90%	53.5%	695	+10.40%
All	6615	55.8%	2264	+4.96%	53.5%	2085	+9.79%

Plays counted only when model probability beat the no-vig market proxy by at least 2%.

Calibration — predicted vs actual win rate

40–45%: 41.8% actual (612g)

45–50%: 47.2% actual (1188g)

50–55%: 53.1% actual (1235g)

55–60%: 58.1% actual (1104g)

60–65%: 63.4% actual (843g)

65–70%: 68.8% actual (512g)

70%+: 76.0% actual (296g)

Actual tracks slightly above predicted in the high buckets — the model is mildly conservative on strong favorites.

Closing line ingestion

Backtests default to a market proxy. Ingest real closing prices here and the priced results below settle against actual numbers, including closing line value. Provider pulls need an ODDS_API_KEY (The Odds API historical endpoint); CSV import always works.

Date

Book

CSV or JSON import — date,away,home,awayML,homeML,awayRL,homeRL

Store

0 lines

0 dates

0 with model probs

0 final scores

Date	Games	ML	RL	Source
No closing lines ingested yet.

Min edge (pp)

Priced against real closing lines — requires model probabilities on the ingested rows plus final scores.

No priced bets yet. Ingest closing lines with model probabilities (JSON rows accept modelHomeProb and modelHomeRlProb), then load final scores.

Configuration comparison

Leakage-free (prior-season starter ERA)

ML acc 55.8% · ML ROI +4.96% · RL ROI +9.79%

Honest, deployable configuration — what the live board uses.

Same-season starter ERA (hindsight)

ML acc 57.7% · ML ROI +17.10% · RL ROI +22.00%

Upper bound only. Uses full-season ERA unavailable at game time.

Team-only baseline (no pitchers)

ML acc 55.8% · ML ROI +4.20% · RL ROI +6.10%

Starting pitcher input is the main source of edge over this baseline.