An organisation adopts blameless postmortems but engineers still avoid admitting mistakes and incidents are under-reported. What is missing beyond the stated policy?

What is missing

The incentive structure contradicts the stated policy, and people respond to incentives rather than to policies.

The specific contradictions to look for:

1. Performance review consequences. If incidents appear in performance conversations, or if "caused an outage" is remembered informally at promotion time, blamelessness is a document rather than a practice. This is the most common contradiction and the least often acknowledged.

2. Leadership behaviour during incidents. A senior person asking "who deployed this?" in the incident channel undoes a year of policy in one sentence. What leaders do during the incident is the actual policy; what the document says is aspirational.

3. Action items that are disguised blame. "Add a training session for the team that made this mistake" is blame with a process wrapper. Genuine action items change the system so the mistake becomes unlikely — a confirmation that shows context, a staged rollout, a command that cannot target production by accident.

4. Asymmetric visibility. Incidents are highly visible; the work preventing them is not. If the only time reliability work is noticed is when it fails, the rational individual strategy is to avoid being near it.

What actually builds the culture

Leaders modelling it. The most senior person describing a mistake they made, publicly and without self-flagellation, does more than any policy statement.

Postmortems that consistently find systemic causes. After enough postmortems conclude "the tooling made this easy to get wrong" rather than "the engineer was careless", people believe it.

Rewarding disclosure. Someone who reports a near-miss nobody would have noticed should be visibly thanked. Near-miss reporting is the strongest available indicator of psychological safety, and it is the cheapest source of learning available — an incident that did not happen.

Separating learning from accountability, explicitly. Blameless does not mean consequence-free for genuine negligence; it means the learning process is separated from any accountability process, and people are told which one is happening.

Time to act on findings. If postmortem action items never get prioritised, the process is theatre and participation declines accordingly. Nothing corrodes the practice faster than doing the analysis and then doing nothing.

The measurable signals

• Near-miss reports. Rising is good. Zero means people are not reporting, not that nothing happens.
• Time from incident to postmortem, and the proportion of action items completed.
• Who writes them. If it is always the same senior people, others do not feel safe.
• Whether junior engineers declare incidents. If only seniors declare, the ladder is not safe to climb.

The underlying point

Culture is what the organisation rewards, not what it publishes. A stated policy contradicted by observed consequences produces cynicism rather than compliance — and the cost is not the cynicism, it is the incidents you never hear about.