Aglet

Verify a staged rollout after cohort divergence

Verification must preserve the comparison that revealed divergence. Recheck the affected cohort, its control, and any newly exposed slice with measures that capture both failures and successful completion. A healthy aggregate should not erase a cohort-specific regression.

Check whether the outcome improved

  1. Recheck the affected slice

    Use the same cohort predicate, endpoint or job, error class, and comparison interval from the investigation. Record exposure volume and any membership changes so an apparent improvement is not just a different sample.

  2. Inspect expansion evidence

    If another slice was exposed, compare its outcomes separately before combining totals. Check completion, retries, latency, and downstream records. Keep results per cohort when a rollout boundary still separates behavior.

  3. Confirm the checkpoint rule

    Apply the predeclared progression or hold condition and record who made the decision. If evidence is mixed or too sparse, mark verification inconclusive and schedule another checkpoint instead of implying the rollout is safe.

What to carry forward

A verification result should say whether the rollout can proceed under the recorded rule, must remain scoped, or needs further evidence. It is complete only when the original divergent cohort and the next exposure slice are both represented.

Keep the decision with the work.

Use a Work Item in Aglet to record the problem, the evidence you have, and the next decision. Add an owner and priority, then keep updates in the discussion so the next person can follow the reasoning.

Create an account See the product workflow