Skip to content

E3 — Incident Response + Blameless Postmortem#

"An incident's value isn't measured by who you blame, but by how the system changes."

Block: E — Ownership · Duration: ~14h · Prerequisite: E2

🎯 When you finish this module#

  • You manage an incident with roles, communication, and a timeline.
  • You write a blameless postmortem, tying the root cause back to the system.
  • You make the action items from the postmortem trackable.

🧠 Why this, why now#

In E2, you knew what to do the moment an alert fired; E3 turns that moment into a process and then institutionalizes the learning. The writing discipline in F4 (ADR/RFC/postmortem) builds on the postmortem practice from this module.

📖 Read first#

Source For what Duration
11-SRE/Incident-Response.md incident management ~30 min
00-Culture/Blameless-Postmortem-Template.md template ~20 min

💥 Broken lab#

👉 labs/broken/K07-incident-sim/ — Symptom: a multi-fault incident simulation; multiple signals, real time pressure. Diagnosis + communication are measured together.

✅ Acceptance criteria#

Don't move to the next module until all of these are verified: - [ ] The K07 multi-fault incident simulation was managed with a UTC minute-precision timeline — written timeline - [ ] A blameless postmortem was written: quantified impact + root cause + "why wasn't this caught earlier" - [ ] At least one trackable action item (owner + due date) was extracted from the postmortem - [ ] bash labs/broken/K07-incident-sim/verify.sh passes with zero errors after the fix

🧪 Test yourself#

  1. What role should be assigned first in an incident, and why does it come before the technical fix?
  2. Why isn't "human error" a root cause?
  3. Which section of a postmortem is more valuable: the root cause, or "why wasn't this caught earlier"?
Answers 1. Incident Commander (coordination + communication). Because people working in parallel trample each other without a single decision point and clear communication; the fix turns into chaos. Role separation is covered in [`11-SRE/Incident-Response.md`](../../11-SRE/Incident-Response.md). 2. Because if you designed the system in a way that let a person get it wrong, the error was inevitable — the root cause is that design (missing guard-rail, fragile process). "Human error" stops the investigation instead of changing the system. 3. The second one. The root cause explains this particular incident; "why wasn't this caught earlier" (missing alert/test/review) is the layer that prevents recurrence — that's where the real learning is. The template is in [`00-Culture/Blameless-Postmortem-Template.md`](../../00-Culture/Blameless-Postmortem-Template.md).

🆘 If you're stuck#

Symptom Likely cause What to do
Everyone is trying to fix it at the same time No coordination role Assign an Incident Commander; clarify roles and the single decision point
The timeline can't be reconstructed afterward No real-time record was kept Keep timestamped notes as the incident unfolds (a chat channel is enough)
The postmortem blames a person No blameless framework Shift the language to the system: not "person X made a mistake" but "the system allowed X"
Action items aren't tracked No owner/date Add an owner + due date + where it will be tracked to every item

💼 Portfolio output#

A written blameless postmortem — also used as a writing sample in F4.

⏭️ Up next#

E4 — Database Production (Restore)


"'Human error' isn't a root cause — it's proof that a system let a person make a mistake."