Your RCA is not measuring what you think.

The individual root-cause analysis produces a document. The blameless group post-mortem produces understanding. Most teams still ship the wrong one.

The hook

An SEV-2 hit at 03:14 last Tuesday. The on-call found the immediate cause by 03:47. Graphs were green by 04:20. Sixteen hours later, a document landed in the incident channel: five headings, one author, two thumbs-up emoji. Three weeks later, a nearly identical incident happened. I have seen this movie before. I have been the person writing that document, alone, at 7pm, convinced I understood what went wrong.

राजस सुकुमार मदनाचा पुतळा · Royal, Delicate, an Idol Sculpted of Love

राजस सुकुमार मदनाचा पुतळा । रविशशिकळा लोपलिया ॥ १ ॥ ॥ ध्रु. ॥
कस्तुरीमळवट चंदनाची उटी । रुळे माळ कंठीं वैजयंती ॥
मुगुट कुंडलें श्रीमुख शोभलें । सुखाचें ओतलें सकळ ही ॥ २ ॥
कासे सोनसळा पांघरे पाटोळा । घननीळ सांवळा वाइयानो ॥ ३ ॥
सकळ ही तुम्ही व्हा गे एकीसवा । तुका म्हणे जीवा धीर नाहीं ॥ ४ ॥
Some perceptions of reality exceed what one person is built to hold. The correct response is to call the group.

What I keep seeing

The RCA document has five headings: Summary, Timeline, Root Cause, Impact, Action Items. It was written by one person, usually the on-call engineer who resolved the incident. It is honest. It is also one head's honesty, compressed under time pressure and edited for defensibility.

Two people react with the thumbs-up emoji. The channel moves on. The same failure mode sits quietly in the system, waiting.

The mechanics

An RCA written by one person is a specific epistemic artifact. It contains what that one person understood by the time they finished writing. It is a compression of their mental model, and mental models under pressure develop blind spots the way windshields develop cracks: exactly where you are looking hardest.

A group post-mortem is a different object entirely. It runs 45 to 90 minutes with the on-call engineer, at least one engineer from the service that failed, at least one from an adjacent service, and someone from customer support. The facilitator asks a specific sequence: what did you see? What did you think it meant? When did you realise it meant something else? What were you doing that would have caught this earlier?

The output of that meeting is not the document. The document is a byproduct. The output is that four to six people now share a corrected model of a system that any one of them, alone, was wrong about. That shared model is the thing that prevents the next incident. Not the action items. The model.

Where Tuka comes in

The closing line of the abhanga: "सकळ ही तुम्ही व्हा गे एकीसवा, तुका म्हणे जीवा धीर नाहीं." All of you, come and stand beside me; my soul cannot hold steady alone. This is not a leadership virtue statement. It is a claim about a specific class of perception. Some pictures of reality are too large for one nervous system.

An incident is exactly this class of perception. Nobody on your team, alone, can see what actually happened. The actual happening includes: what the graphs showed, what the runbook did not say, what support was hearing from customers, what the morning deploy did not warn about, and what the on-call engineer misread on a Grafana panel at 03:31. Each piece lives in a different head. The person who admits this and calls the group is not weak. They are the only one accurately estimating the size of what they are looking at.

What I would actually do

Ban single-author RCAs for anything above SEV-3. Require a group post-mortem within 72 hours. The facilitator is not the on-call engineer; it is a rotating role from a different team. If the meeting cannot happen in 72 hours because everyone is too busy, the incident was more expensive than you have admitted. Put the meeting on the calendar as SEV-1 work. Your operating cadence will feel slower for the first quarter. Your incident-repeat rate will drop by more than half.

Chetan Dhandal