NAMED_MECHANISMS
The Science of Engineering Incidents and Identity: Why It's Rarely the Villain's Fault
3 named mechanisms — "Who broke prod?" feels like the right question after an outage. Systems research gives a less satisfying but far more useful answer: complex systems
Blameless postmortem culture is the organizational practice of investigating failures by focusing on the conditions and system factors that enabled them, rather than on which individual caused them. Codified by Google's SRE Book and pioneered at scale by Allspaw and Robbins at Etsy, the approach holds that punishment deters the honest information-sharing required to understand and prevent future failures.
How it sounds in your headThe inner script after an incident: 'I pushed the deploy and the site went down — I'm going to get fired.' Blameless postmortem culture reframes the question: what conditions made this deploy reach production, what monitoring gaps let the failure propagate, and what system changes prevent it next time — not who pressed what button.
Invisible infrastructure work describes the engineering labor — on-call rotations, system maintenance, dependency updates, documentation, code review, community management — that keeps software systems running but generates no visible output in conventional contribution-tracking systems. Champion et al.'s 2024 empirical study finds this work is systematically under-credited relative to feature development, creating a recognition gap that shapes career trajectories and self-perception for the engineers who do it.
How it sounds in your headThe inner script: 'I spent three months migrating us off the deprecated database driver and nobody noticed — maybe I should just focus on features people can see.' The research names this as a structural recognition problem, not an individual failure to communicate value: the contribution-tracking systems don't count migration work, so it disappears.
The Swiss cheese failure model, introduced by James Reason in 1990, represents complex system defenses as slices of Swiss cheese — each with holes representing individual weaknesses. A failure occurs not when a single layer fails, but when holes across multiple layers happen to align, allowing an accident trajectory to pass through all of them. The model fundamentally shifts incident causation from 'who did the wrong thing' to 'what conditions converged.'
How it sounds in your headThe inner script after a deploy incident: 'I should have caught that — I'm clearly not good enough for this role.' The Swiss cheese model re-reads the situation: the deployment passed code review, CI, staging, and feature flags before reaching production. The incident needed all those layers to be porous simultaneously — the individual who clicked deploy inherited a system with many holes already open.