A choir-attendance run looks brilliant on one random seed and collapses on the next; a packed file for the lobby kiosk was never re-scored; a new model is about to replace one that may already be better. How should testers map those three risks?

Brilliant on one random seed and collapsed on the next is a stochastic misread, so score it with ML functional performance testing. A packed lobby-kiosk file that was never re-scored is a conversion or deploy-pack defect (smoke, ML functional performance, or A/B). A new model about to replace one that may already be better goes to shadow testing. One live canary, a card review, and a Chapter 5 type constraint do not cover those three.