Bet ledger — MLB Edge Engine

Is it profitable

No. Across 24 settled slates and 437 tracked picks the ledger is down $5,056 at flat $100 stakes, a return of −11.8%. The model has since been measured directly against the prices it was betting into and found to be the less accurate of the two, which is the reason rather than the excuse. The paper works through how that was established.

Cumulative profit and loss at flat $100 stakes, by slate. The dashed vertical marks 2026-07-13, when a target leak in the model's dominant feature was found and fixed; everything left of it was produced by a model reading the label off its own feature sheet.

By era

The record splits cleanly at the leak fix. The leaky model's picks hit nine points below even the market's implied rate — the signature of fabricated confidence. The honest model's picks land on the market's number, which is the fix working and the problem restated: matching the market is not beating it.

By market and by claimed edge

Edge is model probability minus the no-vig market probability. If the model is the less accurate of the two, the largest claimed edges are mechanically the model's largest mistakes — and that is what the right-hand column shows.

Every pick

All 437 settled picks, newest first. Staked marks the ones that cleared every risk gate and received a real Kelly stake; the rest were tracked at paper stakes only.

era market result stake

Raw data

The ledger is generated by the pipeline itself and settled every morning against final scores. Both exports are the same 437 rows used to draw this page: mlb-ledger.csv · mlb-ledger.json. Source is at github.com/samarpreetxd/mlbmodel.

Ledger data failed to load — the tables above need /static/mlb-ledger.json. The CSV link still works.