Part 5 of 12 · 2 min read
The post-mortem: structure that forces honesty
A checklist that catches a missing mitigation, a single root cause, or blame creeping into the writing.
Think of a pre-flight checklist for pilots. It exists not because any one item is hard to remember. It exists because skipping one under pressure is exactly how something gets missed.
In plain words: a runbook or post-mortem checked against a required structure catches the specific failures that happen under pressure — a mitigation that's really just "investigate," a post-mortem with only one root cause, or language that quietly blames a person instead of the system.
A worked example. A runbook written for "agent cost per minute above budget" is checked for eight required fields (owner, severity, first check, mitigation, rollback, escalation path, last drill date) and for whether its mitigation names an actual control — the check rejects a mitigation that says only "investigate" because it names nothing concrete to flip, pause, or roll back. A post-mortem is separately checked for seven required sections, for phrases like "should have known" or "his fault" that signal blame has crept into the writing, and for whether it lists more than one root cause and at least one action item that improves either detection or mitigation, not just prevention.
The key lines.
# from code/mini-incident/incident.py
BLAME = ("careless", "should have known", "negligen", "stupid", "his fault", "her fault", "their fault", "lazy")
def check_postmortem(pm):
problems += [f"blame wording: {w!r}" for w in BLAME if re.search(r"\b" + re.escape(w), text)]
if len(pm.get("root_causes") or []) < 2:
problems.append("one root cause only: look for contributing factors")
Read it as: the checker does not judge whether the post-mortem is well written. It only catches a small, specific list of structural and wording failures. These are exactly the ones a tired team under pressure tends to make. A game day is a rehearsal where a failure is deliberately injected to test alerts, switches and people. It is how a team finds out whether its runbooks actually work before a real incident forces the question.
Runbook, blameless post-mortem, game day
A runbook is the written procedure tied to one alert; a blameless post-mortem is a written review focused on system causes and actions rather than individual fault; a game day is a rehearsal where a failure is injected to test alerts, switches and people.
Example: the "mitigation names no concrete control" check above, catching a runbook that would otherwise read as complete while telling a 3 a.m. responder nothing useful to actually do.