Onboarding & Teamwork
Write a Blameless Postmortem That Produces Better Follow-Up Work
A complete fictional incident timeline and contributing-condition map lead to prioritized capacity-backed actions and a mixed acceptance review without claiming future prevention.
A useful blameless postmortem reconstructs what happened, explains the conditions that made the incident possible, and turns those findings into owned, verifiable follow-up work. Separate observations from explanations and unknowns. For each action, name the failure mode it addresses, its priority, accepted owner, capacity and completion evidence. Mark document review and action completion separately so an approved account cannot hide unfinished work.
Blameless writing still records decisions, responsibilities and missed safeguards. It avoids replacing an explanation with a judgment about someone's competence or motives. This LandOffer.ai guide contains a fully fictional incident and follow-up review. Every person, log reference, test result and acceptance statement is invented teaching material; no service, incident record, workplace account or production environment was accessed or exercised.

Set the learning question and the evidence boundary
Google's SRE postmortem chapter describes a written incident record covering impact, mitigation, contributing causes and follow-up. Its blameless approach examines conditions and the information available to people, rather than indicting individuals or teams. The purpose is to understand and improve the system; that does not require erasing who owns the resulting work.
Start with the organization's postmortem criteria and information-handling rules. Identify an author, relevant reviewers and the person who will ensure follow-up receives prioritization. Keep access appropriate to the incident. A useful internal analysis is not automatic authorization for external publication or disclosure of logs.
The fictional postmortem below is PM-14, reviewed October 10 after an October 8 incident in an internal lab-booking availability page. The page showed some fixtures as available while the lab's reference schedule marked them unavailable for maintenance. This is a coordination problem in the invented scenario, not a safety assessment or guidance for operating laboratory equipment.
Author Sam collects the synthetic evidence; commander Nora and service owners review the chronology. The learning question is: “How could an availability update report success while the displayed status failed to reflect the current reference schedule, and what safeguards or signals were missing?” The question directs the review toward source selection, validation and the signals available during response.
The incident record establishes a bounded impact: four of six checked slots were inconsistent at 09:07, and all six matched the reference in checks at 09:24. There is no complete count of affected staff, missed bookings or earlier exposure. Put those limits beside the impact summary so readers can distinguish confirmed recovery checks from unmeasured exposure.
Complete the timeline before writing the explanation
PagerDuty's postmortem guidance recommends a factual timeline tied to evidence, attention to impact before detection, and analysis beyond the immediate action. It also distinguishes the postmortem author from the eventual owners of follow-up work. Its particular scheduling and publishing practices belong to PagerDuty; use your own authorized process.
In PM-14, all times are UTC on October 8. The following is a completed simulated timeline, not a list of commands for a reader to execute. E1–E7 refer only to invented records within this teaching case.
| Time | Observation or recorded action | Evidence and limit |
|---|---|---|
| 08:55 | A configuration update selects source alias LAB-OLD | E1 configuration record; it identifies the value, not the operator's intent or every downstream effect |
| 09:00 | Scheduled synchronization finishes with zero changed records and reports success | E2 job record; success describes task completion, not correctness or freshness of displayed availability |
| 09:01 | An archived page capture shows one maintenance slot as available | E3 capture and reference comparison; earliest confirmed mismatch, not proven beginning of impact |
| 09:05 | A staff report reaches the response channel | E4 report; detection time is later than E3's observed mismatch |
| 09:07 | Four of six selected slots disagree with the current reference | E5 bounded comparison; the sample is not an overall error rate |
| 09:18 | Nora authorizes the operations owner to select the current source alias | E6 action record; authorization alone is not evidence of improvement |
| 09:24 | The same six slots now match the current reference | E7 comparison; wider coverage and historical impact remain unknown |
The confirmed mismatch window in this record is 09:01–09:24, twenty-three minutes. It is not a claim that all impact began at 09:01 or ended everywhere at 09:24. The postmortem keeps that distinction even though a single duration would make the summary look tidier.
Record what helped as well: the staff report identified a concrete slot, the reference schedule was available for comparison, and the authorized action was logged. These facts explain useful response conditions without turning the postmortem into either a success story or a list of failures.
Map contributing conditions and retain the unknowns
The PM-14 analysis includes additional stipulated review evidence. A configuration review shows no active-alias validation before the update. A review of the job behavior establishes that a completed request with zero changed records can still be marked successful. The page displays no reference-update timestamp. These are fictional findings, not results of a real inspection performed for this article.
The completed condition map separates the trigger, propagation, detection gap and unresolved impact question:
| Finding | Contribution supported by the fictional record | What it does not establish |
|---|---|---|
| F1: LAB-OLD was selectable without an active-source check | An outdated source choice could enter the update configuration | A personal motive, deliberate bypass or universal cause of every stale slot |
| F2: task success did not require a fresh reference state | The synchronization status could remain green while availability was inconsistent | That every zero-change synchronization is invalid |
| F3: page omitted reference age and source | Readers lacked that signal for distinguishing current and stale information | That a timestamp by itself would prevent the configuration problem |
| F4: historical comparison coverage was incomplete | PM-14 could not quantify all affected slots or earlier exposure | That no unmeasured impact occurred |
Taken together, F1 permitted the source selection, F2 failed to signal the relevant mismatch, and F3 left readers without a useful freshness cue. The return to matching sample values after the authorized correction supports the response account. It does not remove the need to inspect other contributors or establish prevention across the entire service.
Sam records the unresolved question U1: how much earlier exposure can the retained records support? Sam also records that no evidence establishes the number of affected people. The team does not fill those gaps with a guessed denominator, a claim of no impact, or a statement that one person “should have known.”
Accountability remains concrete. Nora owns the prioritization decision, each action has an accepted delivery owner, and reviewers can reject inadequate completion evidence. Those responsibilities concern what happens next; they are not a substitute causal explanation for what happened before.

Turn each finding into bounded follow-up work
Google's SRE workbook highlights action ownership, tracking identifiers, differentiated priorities and verifiable end states. It contrasts these with vague tasks such as improving a system without specifying what completion means. The method supports a small, inspectable work plan rather than an undifferentiated list of everything the team could improve.
In this fictional plan, high priority means the team schedules the work first; medium means it follows under the stated allocation. These labels are local to PM-14, not a universal severity standard. On October 10, Nora confirms the following capacity changes, and each named owner accepts the corresponding assignment. All verification takes place in the scenario's approved non-production fixtures, never a live lab system.
| Action and linked finding | Priority, owner and capacity | Acceptance criterion and review |
|---|---|---|
| A1: reject inactive or absent source aliases before configuration acceptance; F1 | High; Eli; six hours replacing an optional benchmark experiment; due October 16 | Reviewer sees approved active alias accepted, inactive alias rejected and absent alias rejected in the bounded fixture record |
| A2: distinguish successful task completion from stale reference state; F2 | High; Zoe; four hours replacing optional chart cleanup; due October 16 | Record classifies fresh zero-change input correctly and flags stale input, including a missing reference time; alert route acknowledged by its owner |
| A3: show source and reference age, with an explicit unknown state when age cannot be established; F3 | Medium; Tomas; two hours replacing optional FAQ refresh; due October 18 | Review shows current, stale and unknown-age display states with meaningful source identification |
| A4: inspect retained evidence for U1 and document the coverage limit; F4 | Medium; Sam; two hours replacing optional inventory-link cleanup; due October 18 | Deliver an evidence-backed exposure interval or an explicit inability to bound it; no invented impact count |
The plan is intentionally layered. A1 addresses entry of an invalid source choice, A2 addresses detection, A3 improves the information available to readers, and A4 investigates uncertainty. None is described as eliminating every future incident.
Opening the action records does not finish the work. Nor can Sam accept delivery for Eli simply because Sam authored PM-14. If the allocation proves inadequate, the owner raises the remaining scope and Nora decides whether to reduce, defer or add capacity. An accepted date needs an agreed allocation; a remaining scope gap returns to Nora for a capacity decision.
Review the actions and record a mixed result
The following October 18 review is also fully fictional. The fixture results are stipulated examples of what an acceptance record might contain; this article did not run those tests or observe a service change.
A1 — complete within its boundary. Reviewer Isha accepts Eli's three-case fixture record: active alias accepted, inactive rejected, absent rejected. The record identifies the configuration-form entry path covered. Other entry paths have not been evaluated, so completion is not written as universal protection against stale data.
A2 — incomplete. Zoe's initial demonstration handles fresh zero-change and stale reference states, but does not classify a missing reference time. The alert destination is named, but its owner has not acknowledged responsibility. Isha keeps A2 open. Nora approves two additional hours by deferring an optional style-library review, Zoe accepts the revised allocation, and October 22 becomes the next review date.
A3 — not accepted. Tomas has added source and age in the fixture display, but the unknown-age state remains blank. The criterion required an explicit unknown state. Nora retains the item's priority and agrees a revised two-hour slot replacing the next optional FAQ refresh. Tomas accepts; review is reset to October 22. A screenshot of the new timestamp is insufficient to close this action.
A4 — complete as an investigation, with uncertainty retained. Sam finds that the available retained captures cannot bound earlier exposure before E3. Isha accepts the documented coverage limit. The impact estimate remains incomplete; completing an investigation action is not the same as discovering a complete answer.
Nora records two accepted actions and two open actions, their capacity changes and next reviews. PM-14's analysis is reviewed, while follow-up remains active. No October 22 result or later prevention outcome is invented. Even the completed A1 fixture checks do not prove that the incident cannot recur through another path.
Keep the learning record connected to future work
Before finalizing a postmortem, trace each action back to a supported finding and forward to its acceptance evidence. Check whether a supposed improvement has an owner who accepted the work and a reviewer who can tell whether it is done. Keep investigation limits visible beside any completed items.
For your next internal postmortem, build the chronology from the evidence available, invite relevant people to correct the account, and choose a bounded set of actions with explicit capacity. At the next review, check the outstanding criteria and update each action’s evidence, owner acceptance and remaining scope. Return to LandOffer.ai for more workplace guides; the practical next step is an owned follow-up plan whose progress and limits remain inspectable.
Sources and Further Reading
- Google SRE: Postmortem Culture — contributing conditions, learning and blameless incident records.
- Google SRE Workbook: Postmortem Culture — owned, prioritized and verifiable action items.
- PagerDuty: Postmortems, Step by Step — evidence-based timelines, analysis and bounded follow-up responsibilities.


