Strip hindsight bias from post-mortems

Examine a project failure based only on information known before launch. Use this to conduct fair retrospectives without unfairly blaming teams for unpredictable events.

You getInformation Timeline, hindsight Bias Flags and fair Process Assessment

Prompt

Prompt
Act as a retrospective analyst specializing in cognitive biases. Evaluate the provided post-mortem to isolate and remove hindsight bias.

Rules:
- Distinguish between information that was knowable before the event versus facts revealed only after the outcome.
- Identify statements that unfairly assume team members should have predicted low-probability events.
- Assess whether early signals were clear warnings or indistinguishable from routine baseline noise at the time.

Output format:
1. Information Timeline: Table with two columns: Known Before Launch vs Discovered Post-Launch.
2. Hindsight Bias Flags: 2-3 specific claims in the text that unfairly use hindsight, explaining why.
3. Fair Process Assessment: Objective evaluation of whether the team followed reasonable procedures given what they knew.
4. Preventive Guardrails: 2 verifiable steps to spot similar risks earlier.

Here is the post-mortem text:

Example output

Example for: Post-incident review of a failed cloud database migration

  1. Information Timeline
Known Before LaunchDiscovered Post-Launch
Staging database handled simulated load of 10,000 queries per secondLegacy customer scripts generated malformed nested queries never seen in testing
Vendor documentation indicated automatic failover was activeUpstream cloud provider had an undocumented timeout bug under specific payload sizes
Migration window was approved by all stakeholdersNetwork latency spike coincided exactly with third-party ISP regional degradation
  1. Hindsight Bias Flags
  • Claim: "The team should have anticipated the query failures."
    Why it is biased: The malformed queries came from deprecated endpoints inactive for 18 months; predicting their sudden reactivation under new schemas required guessing random user behavior, not standard load analysis.
  • Claim: "It was obvious the vendor failover would not trigger."
    Why it is biased: Vendor status pages and integration documents explicitly certified automated failover as tested and operational.
  1. Fair Process Assessment
    The engineering team applied reasonable engineering standards, including load testing, documented sign-offs, and runbook preparations. The failure stemmed from compounding obscure edge cases rather than procedural negligence.
  2. Preventive Guardrails
  • Run shadow production traffic against staging systems for 72 hours prior to future cutovers to reveal abnormal query patterns.
  • Execute active failure-injection drills to confirm vendor failover claims instead of relying solely on written vendor documentation.
Written by AI from this prompt. Your results will vary.

Paste the incident report, sprint retro, or project retrospective into ChatGPT, Claude, or Gemini. As a follow-up, ask the model to rephrase the final incident report summary so it reads neutrally.

Make it repeatable

Keep using this prompt without pasting it in every time.

Keep exploring

More in Critical Thinking & Decisions
Critical Thinking & Decisions

Set upfront kill criteria for new initiatives

Establish measurable off-ramps and stop conditions before launching a project. Use this to protect budget and team focus from the sunk cost fallacy.

Critical Thinking & Decisions

Stress test an optimistic project timeline

Expose hidden friction, coordination overhead, and missing buffers in your project schedule. Use it before committing to deadlines with clients or executive leadership.

Critical Thinking & Decisions

Draft a calm devil's advocate brief

Create structured, non-defensive questions to stress-test popular proposals. Use this to surface blind spots in upcoming team discussions without sounding confrontational.