QuantProp. Analyze a trade file ↗
methodology / article

The missing trials
change the story.

The best curve is less informative when you do not know how many curves were tried.

QUANTPROP EDITORIAL  /  UPDATED 2026-10-27  /  EDUCATIONAL

A family is not a collection of independent discoveries

Nearby lookbacks, thresholds and entry delays often describe closely related hypotheses. Counting them as independent discoveries is misleading; deleting them from the search history is worse. Preserve the complete search lineage before choosing a multiplicity correction.

Exploration consumes information

Once a validation period has influenced a decision, it is no longer untouched evidence for that decision. Renaming a file or moving it to a new repository does not reset its informational history. QuantProp labels reused partitions as exploratory.

The practical response

Pre-register a bounded candidate set. Keep rejected variants, compare against simple benchmarks and reserve future observations for confirmation. The portable screen does not produce a significance badge or claim an effective number of independent trials.