Skip to content

Structural Filter Postmortem

Retrospective review — instantiates Structural Filter Intersection Audit

After an output that should have surfaced didn't, reconstructs the trace of every filter it hit and how they combined — blamelessly, because each filter was locally reasonable and each producer sincere.

A Structural Filter Postmortem is an event-triggered retrospective, run after a specific consequential failure — a warning that never surfaced, a story spiked that turned out to matter, a gap that caused harm — that reconstructs the causal trace of that one output through the whole filter stack: which filters it hit, in what order, with what verdict and stated rationale, and how locally-reasonable filters combined to produce a bad joint outcome. Its defining move is that it is blameless by necessity: the archetype's premise is that everyone was sincere, so a postmortem hunting for a culprit finds none and misses the structure. Its deliverable is a durable traceability record for the case — not an assignment of guilt, and not the general forward model of what survives (Intersection Matrix), but the backward trace of one real failure.

Example

At a drug- and device-safety agency, a signal — a pattern of adverse reports about a device — should have triggered a public warning months before it did. After the harm surfaces, a structural filter postmortem reconstructs the trace: the initial reports passed a triage filter that downgraded "anecdotal" signals; a legal-review filter softened the language to limit liability exposure; a resourcing filter deprioritized the item behind higher-volume issues; a final approval filter required sign-off from an official who was overloaded. No one suppressed anything — each filter did its job. But the trace shows the signal died at the intersection, softened here, delayed there, deprioritized elsewhere, and that the same combination would kill the next signal too. The record names the combination, not a villain, and points at which junction to change. Blameless-postmortem practice is what keeps that reconstruction honest.[1]

How it works

  • Trigger on a specific, consequential surfacing-failure — not on routine rejection.
  • Reconstruct the output's full path: every filter encountered, its verdict, its stated rationale, and its point of application, in sequence.
  • Ask the counterfactual at each node — would it have surfaced if this filter had passed it? — to separate where the outcome was overdetermined from where a single node was decisive.
  • Keep it blameless: attribute to filter structure and interaction, not to individual intent.
  • Emit a durable traceability record others can inspect, and that feeds the periodic drift review.

Tuning parameters

  • Trigger threshold — how bad an outcome must be to convene one; set too high, lessons are lost, too low, it becomes noise.
  • Trace depth — how many upstream filters to reconstruct, including pre-emptive self-editing, which leaves the faintest trace of all.
  • Blamelessness enforcement — how firmly individual culpability is ruled out of scope; weak enforcement collapses the review into a witch-hunt and dries up candour.
  • Counterfactual rigour — how carefully each "would it have surfaced?" is tested rather than merely asserted.
  • Record durability — a one-off writeup versus an entry in a standing, searchable traceability record.

When it helps, and when it misleads

Its strength is converting a painful specific failure into a durable structural lesson, and it uniquely exposes the interaction — how several sincere filters compound on a single case — which forward models and aggregate analyses can miss. The blameless framing is what keeps the producers who witnessed the failure willing to reconstruct it truthfully.

Hindsight makes the missed signal look obvious after the fact, tempting the panel to invent a clear villain, or an obvious rule that "should" have caught it; the counterfactual — "it would have surfaced but for X" — is easy to assert and hard to prove. It is misused when convened to assign blame, which produces scapegoats and silence, or run backwards to exonerate the institution: "nobody could have known." The discipline is to enforce blamelessness, test each counterfactual instead of asserting it, focus on the filter combination rather than a single node, and write a record that feeds the periodic drift review rather than a one-off catharsis.

How it implements the components

  • filter_traceability_record — its deliverable: the reconstructed, durable trace of how one output moved through the filter stack, showing where in the intersection it was softened, delayed, or killed.

It reconstructs one failure after the fact; it does not model the general surviving_intersection_model forward (that's Intersection Matrix), maintain the viewpoint_coverage_map and run its periodic_filter_drift_review on a schedule (that's Viewpoint Presence Dashboard), or register each filter's standing filter_legitimacy_standard (that's Filter Rationale Register).

Notes

A postmortem is reactive and case-specific; its findings only compound if each one feeds a standing traceability record and the periodic drift review. Left as isolated writeups, the same structural lesson is re-learned from scratch after the next harm — which is the failure the archetype exists to prevent.

References

[1] A blameless postmortem — a practice from site-reliability and incident-response culture — analyzes a failure for systemic and process causes with individual blame explicitly excluded, on the premise that people tell the truth about what happened only when they are not being hunted for it. It fits this archetype because the producers were sincere: hunting a culprit finds none and hides the structure. ↩