Unit 11.02: Blameless review that changes something

Blameless does not mean actionless, and a review that changes nothing has cost several people an hour.

Five questions, then owned actions

What was known, what made the wrong thing look right, what would have made it visible, what would have made it impossible, and what is changing.

The code lists the questions and the resulting actions.

QUESTIONS = [
    ("what did the person know at the time?", "not: what should they have known"),
    ("what made the wrong action look right?", "not: why did they do that"),
    ("what would have made it visible sooner?", "the monitoring gap"),
    ("what would have made it impossible?",     "the structural fix"),
    ("what are we changing, and who owns it?",  "the output"),
]
print(f"{'ask':42} rather than")
for ask, rather in QUESTIONS:
    print(f"{ask:42} {rather}")

ACTIONS = [
    ("add q42 to the eval set", "owner: A", "done"),
    ("gate on newly-broken-cases == 0", "owner: B", "in progress"),
    ("alert on grounding drop > 0.05", "owner: B", "not started"),
]
print("\nactions:")
for action, owner, status in ACTIONS:
    print(f"   {action:38} {owner:10} {status}")

# Blameless does not mean actionless. A review that produces no owned, dated
# changes has cost an hour of several people's time and altered nothing.

"What made the wrong action look right?" is the question that distinguishes a blameless review from a polite one. It assumes the person acted reasonably given what they could see, and asks what they could see - which is a system question with a system answer.

The actions have owners and statuses. That is the output; everything before it is the method, and a review that produces discussion without owned changes has performed the ritual and skipped the point.

The mistake this prevents

Watch for repeats across reviews. Three incidents of the same shape are not three people making the same mistake - they are one missing control, and treating them individually produces three fixes and a fourth incident. When a category recurs, the action should be a control that makes the class impossible rather than another instance-level fix.

The mistake is treating the absence of blame as the goal. The goal is change, and blamelessness is the technique that makes people share enough information for the change to be the right one.

Takeaway

Ask what made the wrong action look right, and end with owned, dated actions. A review without changes has cost time and altered nothing.

Mark complete