WEATHERPRED / AUTORESEARCH
Test trades.
Keep the evidence.
Two registered batches · January–September 2025 data · Updated 2026-09-06 14:17 UTC
1,836Policy × cost comparisons
0Validated profitable strategies
−$4.45Monthly selector, hypothetical $100 account
Trading returns drive the search
Price momentum, reversals, favorites, longshots and observed-temperature constraints. A strategy does not need to win an overall probability-accuracy contest.
Historical candle trades are conditional estimates. Actual depth and fills remain unverified. The final holdout stays sealed.
What the batches found
Price behavior: 30 of 576 policies showed positive costed validation P&L. None survived the statistical search adjustment. Selecting using earlier months lost money.
Observed highs: no positive costed result. Preliminary reports can disagree with final settlement; cheap contracts were not guaranteed payouts.
Explore every result
| Policy | Net $ | Trades | Adjusted p |
|---|
P&L covers the 92-day development validation period, one conditional contract per selected event. Variations reuse the same events and are not independent trials. Positive rows are exploratory leads.
Next experiments
Faster reaction to newly received observations; maker spread capture after queue and adverse-selection checks; transient inconsistencies across related brackets. Each new batch is registered before scoring.
Live paper orders retain their frozen model and execution rules. No real-money orders.