Proof before picks

Show the maths. Show the receipts.

Historical model evidence and real published results stay separate here. A large corpus does not excuse a weak backtest, and a promising backtest does not count as live performance.

Public outcome evidence unavailable. Some data is delayed. The figures shown are the last we could confirm. Accuracy, ROI, calibration and CLV are withheld; zero is not substituted for missing evidence.
01 / Engine Health

Is the daily system operating?

unavailable
Latest provider attemptNo recorded attempt
Fixtures analysed0
Decisions generated0
Value picks on today's slate0
Official picks published—
Stale decisions0
Settlement backlog0

Data sources contributing: none in the latest run. Recorded data gaps: 1.

02 / Historical foundation

What is the engine learning from?

failed

Training corpus census failed while reading Supabase row counts.

Finished fixtures0
Odds snapshots0
Feature rows0
Backtest runs0
Player availability0
Lineup snapshots0
Player match facts0
Learning pipeline

From raw history to a live-approved adjustment

Each gate is separate. Later stages cannot borrow authority from a large corpus or a promising backtest.

  1. 010

    Finished fixtures

    Chronological results and provider facts.

  2. 020

    Benchmark backtests

    Walk-forward holdouts, kept separate from runtime authority.

  3. 030/3

    Runtime parity

    Exact model entrypoint and feature-contract receipts.

  4. 040

    Calibration runs

    Settled live decisions grouped by model version.

  5. 050/0

    Review-ready

    Candidates passing sample, skill, error and CLV gates.

  6. 060

    Approved promotions

    Operator-approved, model-bound live adjustments.

football0% complete features
fixtures
0
odds
0
features
0
backtests
0
basketball0% complete features
fixtures
0
odds
0
features
0
backtests
0
tennis0% complete features
fixtures
0
odds
0
features
0
backtests
0

Latest walk-forward backtests

SportWeight sourceOdds evidenceTest samplePicksBrierLog lossYieldCLV
03 / Public Pick Performance

What happened after publication?

WithheldNo public outcome metric is inferred from an unreadable ledger
Outcome ledger not read

No false zeroes

Settled picks, wins, losses, ROI, odds, edge, Brier score and simulated profit will return only after the canonical public ledger is readable.

Inspect ledger availability
04 / Calibration

Confidence versus reality

A small gap is better
Withheld

Calibration needs real settled outcomes

The probability ladder is not drawn because the outcome ledger could not be read. An empty repository and an unavailable repository are different states.

05-09 / Outcome breakdowns

No breakdown without a ledger

unavailable

Sport, league, market, confidence, data-quality and closing-line tables are omitted because their source could not be read. This prevents a grid of zeroes from looking like measured performance.

Sport and league
Withheld
Market performance
Withheld
Confidence and data quality
Withheld
Closing-line value
Withheld
10 / Warnings

What could make these numbers misleading?

2
action

Public outcome ledger is unavailable

Accuracy, ROI, calibration and closing-line metrics are withheld because the public-pick repository could not be read. Zero is not substituted for unavailable evidence.

watch

Provider coverage has gaps

unavailable provider state with 1 recorded gap. The engine does not substitute demo fixtures.

How these numbers are counted

Wins divided by wins plus losses. Pending, push and void results are excluded.

One unit per settled public pick; pushes return the unit and voids are excluded from stake.

Binary published-selection Brier score. Lower is better; pending, push and void results are excluded.

Always excluded
  • mock or demo predictions
  • internal model runs
  • watchlist-only candidates
  • non-positive-edge analyses
  • unsettled picks
Play it smart, padi: good analysis helps you decide better, but no analysis removes risk. OddsPadi presents probabilities rather than certainty, doesn't take bets, and doesn't hold your money. If you bet, keep it fun — only stake what you can afford to lose. Limits and support near you.