What criteria do you use to filter out low-probability Inversion Fair Value Gaps during market retracements?

When an imbalance is invalidated, what lower-timeframe displacement structure must form on the retest to confirm that the gap has flipped into valid support/resistance?