Track Record

Judge the model by its ledger

Published picks settle in public with their original odds. Until the model earns publication, its internal record and the seven gates below show exactly how far it is from that bar — the numbers are the engine's own, not marketing.

Official public picks

The only record that counts publicly

Every published pick

No official public picks have been published yet, so there is no public record to report. The gates below show how far the model is from that bar.

Evidence classes

What counts, and what never does

  • Official public picksThe published ledger above. The only class in OddsPadi's public record.
  • Internal decisionsEvery call the engine grades, published or not. Training evidence — shown below, never as public performance.
  • Shadow decisionsPaper-mode runs of candidate models. Never public.
  • BacktestsHistorical replays on the corpus. Never live performance.
  • Community selectionsMembers' own tips, with their own leaderboard. Never OddsPadi's record.

These are never aggregated into one number. Mixing them is how a paper trade becomes a claimed win.

Yesterday

Internal model record

Full results ledger
Model wins126
Model losses140
Pending0
Hit rate47%

Internal record: every decision the engine grades against final scores, published or not. Published picks are the subset that cleared all gates — see the ledger for those.

Road to publication

The seven promotion gates, live

Engine status

6 of 7 gates passing on 455 settled outcomes. Publication opens when all 7 hold — no gate is ever lowered to make picks appear.

  • Settled outcomes455 ≥ 30
    Passing
  • Brier score0.1827 ≤ 0.25
    Passing
  • Brier skill vs base rate+0.1733 ≥ 0
    Passing
  • Log loss0.5422 ≤ ln 2
    Passing
  • Calibration error (ECE)0.0380 ≤ 0.1
    Passing
  • Closing-line coverage32.1% ≥ 80%
    Not yet
  • Average closing-line value+1.3044 ≥ 0
    Passing
The deep record

Everything you can audit

Methodology

How this record stays honest

  • Every decision is graded against final scores by an hourly automated sweep — including the calls the engine withheld.
  • Settlement uses the odds recorded when the decision was made, and closing prices are captured for closing-line value.
  • Corrections are append-only: a wrong number is corrected forward and noted, never silently rewritten.
  • No gate is ever lowered to make picks appear; publication opens when all seven hold.