AlphaVerdict
daf9b08d5f27469b · synthetic-evidence-compositeEquity path
Teal: strategy · gray: configured benchmark. Hypothetical, after configured friction.
Research metrics
Priority validation work
- Reduce turnover or demand a larger gross edge; the current result does not survive plausible friction.
- Simplify the rule or broaden the sample; a stable edge should not depend on one contiguous window.
- Collect more independent periods and inspect the lower confidence path.
- Treat the edge as unproven and collect more independent observations.
- Describe the failing regime explicitly and test a pre-declared exposure gate rather than tuning on the full sample.
- Repeat the evaluation without those symbols and inspect point-in-time universe breadth.
- Repeat the run with licensed or user-owned real point-in-time data.
Run identity
Data
5a55c243a6156b4de57d2e52ec4babcf3adda65710e188be2184da1708ac63b6Strategy
0a36531a0f6d1555b0c1c194351745d78f5075860529db9334d89f07d0b32e26Findings
Approximate breakeven friction is -126.3 bps versus 15.0 bps configured.
Next: Reduce turnover or demonstrate a wider edge on untouched data.
0 of 2 folds had positive total return.
Next: Simplify the rule and pre-declare a fresh holdout before further tuning.
Probability of a positive resampled total return is 6.0%.
Next: Collect more independent periods and inspect the lower confidence path.
Two-sided randomization p-value is 0.178.
Next: Treat the edge as unproven and collect more independent observations.
Only one evaluated benchmark regime had positive compounded return.
Next: Treat regime dependence as a hypothesis and validate it on a fresh sample.
The top three symbols contribute 68.8% of absolute gross contribution.
Next: Repeat the evaluation without those symbols and inspect point-in-time universe breadth.
Synthetic fixtures validate plumbing only and cannot support a market claim.
Next: Repeat the run with licensed or user-owned real point-in-time data.
Research boundary
A pass means only that this run survived the configured tests. It is not evidence of future returns, a recommendation, or permission to deploy capital.
Results are research simulations, not investment advice or execution instructions.
Signals are formed after a decision close and applied at the following session open.