## Slide 1

State clearly that this is a synthetic incident in a fictional service. The purpose is to rehearse a blameless review. Describe the observed delay and recovery without attributing personal fault or declaring an unproven root cause.

## Slide 2

Draw the boundary around the evidence. API acceptance, queue persistence and partner completion are distinct states. A retained request identifier does not prove immediate delivery, and delayed delivery does not prove a lost write.

## Slide 3

Read the definitions, not only the large numbers. The 47-minute interval follows a declared threshold. The zero lost-write count is limited to the reviewed reconciliation sample and must not become an absolute claim.

## Slide 4

All timestamps use UTC and are strictly ordered. Explain which record supports each event. The recovery threshold and the backlog-drain completion are different milestones; only the former appears in this timeline.

## Slide 5

Review decisions using the information available at the time. Restoring a known configuration was a recovery choice, not proof of a single cause. The bounded communication deliberately avoided unsupported conclusions.

## Slide 6

Keep confirmed factors and hypotheses visibly separate. Replay supports retry amplification within its fixture. Partner timeout clustering remains a hypothesis until a controlled test isolates its contribution.

## Slide 7

Give every unknown an investigation and an owner. This creates a path to evidence without pretending the review is complete. If the full reconciliation changes the impact count, update the record and communications.

## Slide 8

A control can work and still have a gap. The latency alert detected real impact, but earlier queue signals may reduce detection delay. The load test gap concerns scenario coverage, not an individual’s diligence.

## Slide 9

Read the verification condition for each action before discussing implementation. Proposed and In progress are not complete. A code change alone cannot satisfy a behavior-based success condition.

## Slide 10

Explain the artifact required to close each action. Replay results need input fixtures and configuration identity. A runbook needs a successful use by someone other than its author, including ambiguity notes.

## Slide 11

Communication should preserve the distinction between known facts, operating guidance and unresolved hypotheses. These messages are examples for review and are not automatically sent to any recipient.

## Slide 12

Close with the evidence review. Keep unresolved questions visible and verify the follow-up actions before closure. The learning loop should improve system behavior while preserving the limitations of the original record.