Research
Research Lab
Illustrative data
12 validations running
Generated
1,248
Agent-proposed candidates this cycle
→Failed screening
1,061
Look-ahead, survivorship, thin liquidity
→Failed out-of-sample
142
Did not persist beyond the fitting window
→Failed multiplicity
38
Not significant after correction
→Survived the gate
7
Eligible for paper deployment
Candidates under validation
Cycle 2026-W36 · figures illustrative| Candidate | Stage | OOS periods | Verdict |
|---|---|---|---|
Cross-asset carry, regime-gated CAND-2291 |
Walk-forward | 6 / 8 held | Survived |
Vol-surface dislocation, index options CAND-2287 |
Multiplicity | 7 / 8 held | Testing |
Momentum + liquidity filter, small cap CAND-2279 |
Out-of-sample | 2 / 8 held | Killed |
Term-structure reversal, energy CAND-2274 |
Screening | — | Killed · look-ahead |
Dispersion, sector-neutral CAND-2268 |
Walk-forward | 5 / 8 held | Testing |
Validation gate
CAND-2291✓
Look-ahead check
No feature uses information unavailable at decision time.
✓
Survivorship check
Universe includes delisted and failed instruments.
✓
Out-of-sample persistence
Behaviour holds on data never used in fitting.
✓
Walk-forward stability
Held across 6 of 8 rolling windows, including two adverse regimes.
✓
Multiple-comparison correction
Significant after adjusting for every candidate tested this cycle.
✓
Cost & slippage modelling
Survives realistic transaction costs and capacity limits.
•
Paper deployment
Required observation period before any capital is allocated.
✕
Human authorisation
Not requested. No strategy reaches capital without a named approval.