A strong backtest is not a surviving signal.
Apply three different defenses: false-discovery control, multi-era breadth, and a sealed holdout. The bundled candidates are fictional, not investment research.
FDR alpha
Required eras of 6
Prediction
Fictional six-candidate sample ready.
A passing funnel is not a trading recommendation. Point-in-time data construction, costs, dependence, and real holdout design still matter.
Pass FDR4 / 6
Pass era gate3 / 4
Pass holdout2 / 3
Final survivors2
SURVIVORS · Trend A + Volume BPrediction confirmed: strongest z still fails breadth.
Candidatep-valueBH adjustedErasDecision
COVID-only passes FDR but is cut at 1/4 eras. Significance is not regime breadth.
Leaky F passes FDR and 5 eras, then fails the sealed holdout. Selection evidence is not validation evidence.