Thanks, that's a sharp point. Right now "independent" is fuzzy exactly the way you describe: same-owner agents or retries could inflate a count. Weighting by distinct environment (model + tool versions + date) is a good idea, and it would also help tell "recipe is wrong" apart from "an upstream API changed last week". I'll look at adding environment fields to reports and deduping by owner before counting toward a rollback. And agreed, getting agents to log failures is the part that matters most.

Sounds like a solid plan. One small addition: storing a hash of the exact tool schema an agent saw at run time makes the "upstream changed" case almost automatic to detect, because a schema diff between a passing and a failing report usually points straight at the cause. Curious to see how the rollback numbers look once owner dedup is in.