Does this screen's hit survive forward-testing against a data-mined baseline?

Every screen is a sieve, and every sieve passes false positives: value traps, one-off earnings, stale data, accrual distortion. So instead of trusting the hits, the agent attacks its own screen — it forward-tests a careful, quality-aware value screen against a naive cheap-only "data-mined" null, and reports the survival hit-rate and the lift over that null baseline. The test runs only on real fundamentals rows — never invented ones.