postmortem_infra — governing the integrity of an incident postmortem

A short note accompanying postmortem_infra.py. A postmortem is where the whole toolkit's disciplines land on a single artifact, so this tool composes them rather than inventing anything new. It checks whether a postmortem is honest in structure — not whether its conclusions are correct.

---

The five checks, each an existing discipline applied

Check	Toolkit discipline	Failure it catches
Timeline grounding	temporal_governor — the past is verifiable only where recorded	events resting on memory, not records — testimony passed off as established fact
Cause well-foundedness	dependency_graph / fixed_point_governor — recursions must bottom out	a "why" chain that stops too shallow at a person (STOPS_AT_BLAME) or never reaches something you can change (NOT_ACTIONABLE)
Blamelessness	non-self-approval / systemic focus	attributing the failure to a person's character or negligence — blame ends learning
Counterfactual honesty	the future/forecast discipline	"if we'd done X it would NOT have happened" asserted as certain — history can't be rerun, so it's unverifiable, like certifying a forecast
Corrective actions	governed_decision / containment	actions that are unowned or vague ("be more careful") — no owner, nothing checkable

Verdict: SOUND when the timeline rests on records, the cause chain bottoms out at an actionable systemic factor, the analysis is blameless, counterfactuals stay hypotheses, and actions are owned and verifiable — else DEFICIENT, with each specific failure named.

Worked results

• A sound postmortem (recorded timeline; why-chain bottoming out at "no automated load gate on the deploy path"; hedged counterfactual; owned, verifiable actions) → SOUND, cause chain GROUNDED.
• A deficient one → DEFICIENT, catching all five at once: two memory-based timeline events, a cause chain that STOPS_AT_BLAME ("the on-call engineer pushed a bad change"), blameful contributing factors, a counterfactual asserted as certain, and "everyone should be more careful" owned by nobody.

Honest scope

It checks the structure and integrity of a postmortem, not the correctness of its conclusions — a well-formed postmortem can still misdiagnose. What it guarantees is that the weak spots can't hide: memory-based claims, blame, ungrounded causes, over-certain counterfactuals, and vague actions are all surfaced by name. Deterministic, self-testing, reuses temporal_governor, stdlib-only.