Aglet

Triage divergent behavior during a partial rollout

Partial rollouts can hide a real regression inside an apparently healthy aggregate. Start with the rollout boundary and compare exposed and unexposed cohorts using the same behavior measure. Keep rollout configuration, traffic mix, and data differences visible while scope is still uncertain.

Establish what is happening

  1. Map exposure

    Record which environments, regions, tenants, percentages, or feature cohorts received the new version and when. Confirm the exposure source instead of inferring it from a request header or dashboard label that may be stale.

  2. Choose matched cohorts

    Compare exposed traffic with the closest unexposed control on the same endpoint, job, customer type, and time interval. Note differences in volume, device mix, data shape, and dependency routing that could explain unequal outcomes.

  3. Describe the divergence

    State the exact behavior that differs, its first observation, and whether it is increasing, stable, or intermittent. Include one representative success and failure from each cohort where available, while marking missing samples explicitly.

What to carry forward

Triage is complete when the rollout slice, comparison slice, symptom, and evidence gaps are explicit. Escalate as a scoped rollout concern when exposed behavior differs materially; avoid broad claims until controls and traffic mix have been checked.

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