Prospective prediction or held-out test

Lock the rule first, then test it on future or reserved cases it has never seen.

Your stock-screening spreadsheet "would have" returned 22% a year, and you're ready to fund it. The catch: you built the rule while staring at the historical answers, so of course it fits them. A rule tuned on the past explaining the past is a story, not a test.

The move is to freeze the rule first, then test it on future or reserved cases it has never seen. The cue is any system that was designed and evaluated on the same data. Steps: write the rule down exactly and timestamp it, apply it on paper to next quarter's picks (or to a slice of history you deliberately never looked at while building), pre-commit what would count as success, then score the frozen predictions against what actually happened. When it works, hindsight and skill come apart: a rule that only ever worked on the data that shaped it dies cheaply on paper instead of expensively in your account. Don't use it when waiting for fresh cases costs more than the decision is worth, or when the rule is too fuzzy to pin down, because a rule you can quietly reinterpret afterward can never fail.