Non-Target Impact Pre-Mortem¶
Foresight ritual — instantiates Bycatch-Aware Selective Intervention Design
Before deployment, imagines the intervention has already caused off-target harm and works backward to name who gets caught and how — turning bycatch into a design input rather than a post-mortem finding.
A Non-Target Impact Pre-Mortem is a structured foresight exercise run before a selective intervention ships. The team assumes the intervention has already gone live and inflicted collateral harm, then works backward to answer one question: who got caught that we never meant to catch, and by what path? Its defining move is that it fixes the target and non-target rosters at design time — while they are still cheap to change — rather than discovering the off-target classes from complaints after the fact. Unlike the archetype's measurement tools, it produces hypotheses about whom to watch, not readings of who was hit; it is the step that makes later monitoring targeted instead of a blind sweep.
Example¶
A video platform is about to switch on an automatic takedown rule that removes uploads matching a set of known extremist-propaganda fingerprints. The target class looks obvious — the propaganda itself. Before launch, the trust-and-safety team runs a pre-mortem: "It's six months from now and we're on the front page for censoring the wrong people. What happened?" Working backward, they name the non-targets the fingerprint rule will sweep up: journalists documenting the same clips, human-rights monitors archiving them as evidence, counter-speech creators quoting propaganda in order to rebut it. Each becomes a line in a non-target register with a plausible capture path — the flagged clip is present, but the surrounding context that flips its meaning is never read by a hash match.
The output isn't a rate; nobody has been wrongly removed yet. It is a sharpened target definition — propaganda that promotes, not merely depicts — plus a watchlist of classes the later monitoring must sample. That watchlist is what tells the team to build a verified-researcher appeals fast-lane before the rule goes live, not after the first wave of wrongful strikes.
How it works¶
- Assume the harm as already certain. Start from "the intervention caused off-target damage" as a given. Treating failure as a fact, not a risk, licenses the team to name victims without the optimism a forward-looking risk list carries.
- Sharpen the target boundary. Force an explicit, testable definition of what is genuinely on-target — because "off-target" is undefined until "on-target" is pinned down.
- Enumerate non-targets by capture path. For each plausible non-target class, name how the selector would confuse it for the target: a shared feature, adjacency, coincident timing.
- Rank and route. Prioritize the register by plausibility × harm and hand each entry to the mechanism that will actually watch, release, or score it — nothing is "noted" without an owner.
Tuning parameters¶
- Imagination horizon — how far-fetched a capture path to entertain. Wider surfaces rare but severe bycatch; too wide drowns the register in noise.
- Roster granularity — a few broad non-target classes versus many fine ones. Finer catches edge cases but multiplies what downstream monitoring has to cover.
- Facilitation independence — whether an outsider runs the session. An external facilitator blunts the team's optimism and its sunk commitment to the current design.
- Target-definition strictness — how narrowly the on-target class is drawn. A tighter definition reclassifies more of the world as acknowledged non-target — safer, but it raises the bar the selector must clear.
- Cadence — one-shot pre-launch, or re-run at each major design change as the intervention and its environment move.
When it helps, and when it misleads¶
Its strength is timing: it moves bycatch discovery to the cheapest possible moment — before code ships or a rule takes effect — and defeats the specific blind spot that a team optimizing for target capture never volunteers its own collateral. Saying the non-target classes out loud is also what lets downstream monitoring be aimed rather than exhaustive.
Its foresight is only as good as the team's imagination: a non-target class nobody pictures gets no line in the register, so a pre-mortem complements but never replaces live sampling. Its classic misuse is running it as theater — holding the workshop to satisfy a governance checkbox, then shipping the original design unchanged — which yields a register nobody wires into anything. The discipline that guards against this is to treat every entry as a commitment: each named non-target class is assigned to a real downstream watcher (a sentinel, an audit, an appeals path) or explicitly, on the record, accepted as tolerated harm. The technique's real name is prospective hindsight, and its power comes precisely from imagining the failure as already inevitable.[n1]
How it implements the components¶
target_class_definition— its first product: a sharpened, testable boundary for what the intervention is actually meant to catch, so "off-target" becomes a meaningful category.non_target_class_register— its central output: the prioritized roster of classes that could be caught but shouldn't be, each tagged with the path by which the selector would mistake it for the target.
It stops at naming and defining. Measuring how often those classes are actually hit belongs to Non-Target Sentinel Sampling and the Bycatch Rate Dashboard; building the releases and exclusions it recommends belongs to Escape Hatch or Release Protocol and Negative Filter or Exclusion Device.
Related¶
- Instantiates: Bycatch-Aware Selective Intervention Design — the pre-mortem supplies the target and non-target rosters the rest of the design watches and scores against.
- Sibling mechanisms: Non-Target Sentinel Sampling · Selectivity Window Test · Selector Retuning Cycle · Success Metric Reweighting · Bycatch Rate Dashboard · Bycatch Tolerance Stop Rule · Compensation and Restoration Trigger · Escape Hatch or Release Protocol · False-Capture Audit · Negative Filter or Exclusion Device
Editorial Notes¶
Form Classification¶
Form family: Experiment, Test & Rehearsal
Rationale: Non-Target Impact Pre-Mortem operates as an active test, trial, simulation, drill, or rehearsal that generates evidence through a deliberate attempt or perturbation because it before deployment, imagines the intervention has already caused off-target harm and works backward to name who gets caught and how — turning bycatch into a design input rather than a post-mortem finding.
Independent corroboration: The frozen evidence defines Non-Target Impact Pre-Mortem as 'Before deployment, imagines the intervention has already caused off-target harm and works backward to name who gets caught and how — turning bycatch into a design input rather than a post-mortem finding', so its operative form is Experiment, Test & Rehearsal.
Nearest alternative: Assessment, Review & Assurance — Non-Target Impact Pre-Mortem includes features of a bounded evaluation of existing evidence or work that produces a finding or disposition, but its defining operation is an active test, trial, simulation, drill, or rehearsal that generates evidence through a deliberate attempt or perturbation.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Organizational & Management Science
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Managerial decision practice developed the pre-mortem: assume a plan has failed and work backward to concrete causes before commitment.
Related originating lineages:
- Engineering & Design — Safety and risk engineering contributed systematic off-target hazard identification and design modification before deployment.
- Environmental Science & Climate Studies — Environmental impact assessment contributes explicit target and non-target harm classes.
- Psychology — Judgment research on prospective hindsight supplies the cognitive basis for generating neglected failure paths.
- Ethics of Technology & AI Governance — Algorithmic impact assessment materially shapes use against automated moderation and other selective technologies.
Review resolution: Both independent reviews agree on primary origin organizational_management; reconciliation resolves alternate_origin_disagreement. Formative alternate lineages retained: engineering_design, environmental_climate, psychology, tech_ethics_ai_governance. The broader reach of later applications is kept separate as domain_reach=multi_domain; origin_mode=cross_disciplinary_synthesis describes the historical relationship among lineages. Confidence is conservatively reconciled to high, and encyclopedia_synthesis=true preserves the reviewers' boundary judgment.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
The register's value is entirely in being routed. An unrouted pre-mortem finding is worse than none: it leaves a paper trail proving the harm was foreseen and then ignored, which converts a design oversight into a foreseeability problem. The ritual's real deliverable is not the workshop but the set of downstream assignments it forces.
[n1] Prospective hindsight — imagining that a plan has already failed and then explaining why — was shown to improve people's ability to generate concrete, plausible failure causes; Gary Klein adapted it into the managerial pre-mortem. Assuming the harm as already certain is what strips out the optimism a forward-looking risk list carries. ↩