Track record

Every pick flagged by our tracking rule is graded here: wins, losses, pushes and voids. Picks are never removed, re-priced or filtered after the fact.

What the tracking rule is (and isn't): the rule's earlier backtest result turned out to be a data leak. Leak-free, it was about break-even. We keep grading it in public anyway, because a live, pre-lock record is the only honest way to measure a model.

No graded picks yet (n = 0). Live logging started in Week 4 of 2026: 621 picks and 22045 forecasts were logged before lock and are waiting on results (2026 wk4). Until picks are graded, we make no win-rate or ROI claims.

0graded picks (n)

0-0-0W-L-P · 0 void

—win rate · break-even 52.4% at -110

—ROI per 1 unit risked · 0.00 units

—avg. line CLV (points moved our way by close), n=0

0all graded forecasts (incl. non-picks)

By week

Week	n	W-L-P	Win rate	Units	ROI
No graded weeks yet. Week 4 is graded after the games, usually Tuesday.

Every graded pick

Player	Prop	Line	Side	Actual	Result	Units	Close	Logged (ET)	Kickoff (ET)
No graded picks yet.

* Graded at an assumed -110 price because the line source publishes no prices. "hidden" = graded against a line from a source we aren't licensed to republish (ESPN/DraftKings feed); the result still counts. Download every graded forecast (CSV) · every logged forecast (CSV)

Calibration (all graded forecasts)

Do the props we price at 60% hit about 60% of the time? This check covers every graded forecast, not just picks, so it fills in faster than ROI does.

Model P(over)	n	Predicted	Actual over rate
Fills in once Week 4 is graded.

How to audit this log

The rule: Tracking rule (NOT a validated edge): p = 0.15 x model + 0.85 x market reference, flag if the blended probability beats the price by more than 3 points. Its earlier backtest result came from outcome leakage; leak-free, it was about break-even (-0.5% on 2022). Lines without published prices are graded at an ASSUMED -110.
Before lock: the logger refuses lines for games that have already started. Every row carries its UTC log timestamp and kickoff time.
First forecast counts: if we re-price a prop, only the first logged forecast is graded.
Hash chain: each record stores the SHA-256 of the record before it, so any edit or deletion breaks the chain.

Week	Log file	Records	Chain	File SHA-256
2026 wk04	lines.jsonl	7882	verified	e1c232b85e967bd6b0d78b4d…
2026 wk04	forecasts.jsonl	22045	verified	4967fd0228de957d038c4b82…
2026 wk04	grades.jsonl	0	verified	e3b0c44298fc1c149afbf4c8…

Historical backtests are not included in these numbers. Only live, pre-lock, graded picks count. Past results, when we have them, don't guarantee future results. For entertainment and informational purposes only. 21+. 1-800-GAMBLER.