Measure the signal.
Respect the limits.

Walk-forward structure-proxy results, complete sample sizes, subsequent returns and uncertainty—published without hiding weak buckets.