The unchanged strategy did not pass the broad untouched holdout and must not advance.
The unchanged rules were tested from 04-03-18 through 08-17-26 with realistic executable-side fills. This is historical research, not approval to trade.
Training, validation and untouched holdout
These are consecutive historical stages with different jobs. Completed positions passed every eligibility check and reached an exit; skipped opportunities failed a frozen data, liquidity, contract-selection, or risk rule and were not simulated.
| Period | ROI | Win rate | Profit factor | Sharpe | Max drawdown |
|---|---|---|---|---|---|
| −0.04% | +39.22% | 0.98 | -0.03 | −0.88% | |
| −0.00% | +25.00% | 1.00 | -0.00 | −0.41% | |
| +0.14% | +33.33% | 1.71 | 0.33 | −0.20% |
Does it survive higher costs?
The same frozen strategy is rerun with different fill-cost and maximum-risk assumptions. A robust result should not disappear after small, realistic changes.
| Scenario | Holdout return |
|---|---|
| $0.01 adverse slippage per leg · 0.50% risk | +0.14% |
| $0 adverse slippage per leg · 0.50% risk | +0.16% |
| $0.03 adverse slippage per leg · 0.50% risk | −0.23% |
| $0.05 adverse slippage per leg · 0.50% risk | −0.26% |
| $0.01 adverse slippage per leg · 0.25% risk | +0.00% |
Not approved for paper trading
A favorable historical test must still complete forward paper validation, operational reconciliation and kill-switch drills.
This result has a permanent audit record
The exact strategy, code, data snapshots and assumptions are fingerprinted together. Changing any of them creates a different research record; licensed raw quote data remains private.
39b0abcb4175…