Assignment without correct exposure logging mixes people who never saw the change into the treatment arm. Results flatten, and teams conclude the canary “proved nothing.”
Check that exposure waits for the variant UI, that bots are filtered, and that primary stability metrics can physically move from the change under test.
Secondary guardrails still matter — support tickets, payment failures, latency — but they should not silently become the freeze metric mid-rollout.
A short telemetry audit before a high-visibility window costs less than a week of arguing about a contaminated canary.