Near-Miss Response Tier¶
Policy rule — instantiates Counterfactual Proximity Signal Calibration
A standing policy that maps a case's proximity band to a bounded, graduated response — monitor, review, redesign, escalate — without ever booking it as a completed loss.
A Near-Miss Response Tier is a pre-committed policy: it partitions near-misses into a small number of tiers by how close they came, and it fixes, in advance, the bounded response each tier is allowed to trigger. Its defining move is that the response is graduated and capped — a low tier logs and monitors, a middle tier convenes a review, a high tier forces redesign or escalation — and no tier may convert the near-miss into a completed harm, a full reward, or a blame record. Where other mechanisms measure or price an almost, this one decides, by rule and ahead of time, what may be done about it. Deciding the response in advance is precisely what stops each dramatic close call from being re-argued from scratch under the influence of the moment.
Example¶
A security operations team runs a tiered policy for blocked intrusion attempts. A scan that hits a closed port and stops is Tier 1: logged, counted, no action. An attempt that gets valid credentials but is stopped by multi-factor authentication is Tier 2: a review of the exposed account and a credential rotation. An attempt that reaches an internal service and is halted only one action short of privilege escalation — the last control was the only one that held — is Tier 3: mandatory architecture review and a redesign of the affected trust boundary within a fixed window. The policy is written down before any incident, so when the Tier 3 case arrives at 2 a.m. the response is not negotiated; it is triggered. Crucially, none of these is recorded as a breach, because no breach occurred — the tier governs the response, while the security record still reads "attempt blocked." The proximity earned a proportional reaction without inflating the incident statistics.
How it works¶
The policy has two moving parts. A band partition cuts the proximity range into a few labeled tiers with stated cut-points, so any case lands in exactly one tier. A learning-update boundary attached to each tier enumerates the permitted responses and, just as importantly, the forbidden ones: a tier may mandate monitoring, review, retraining, or redesign, but may never create a completed-outcome statistic, full credit, or blame from an event that did not actually occur. Because the mapping is fixed ahead of time, applying it is mechanical — read the proximity band, execute the pre-agreed bounded response — which is what removes in-the-moment discretion and its biases.
Tuning parameters¶
- Number and cut-points of tiers — how finely proximity is banded and where the lines fall. More tiers give finer proportionality but more boundary disputes; the cut-points set who gets the heavy response.
- Response ceiling per tier — the strongest action each tier may compel. Raising the top tier's ceiling makes the policy bite harder on the closest calls but risks overreaction.
- Escalation trigger — whether repeated low-tier calls roll up into a higher-tier response. Rolling up catches drift; not rolling up keeps tiers independent and simple.
- Mandatory-vs-advisory — whether a tier's response is compelled or recommended. Mandatory removes discretion (and its biases) but also its judgment.
When it helps, and when it misleads¶
Its strength is proportionality on autopilot: the closest calls automatically get the strongest bounded response, and nobody has to win an argument in the heat of the moment. It is a direct counter to near-miss normalization, the slow drift by which repeated close calls come to feel routine. Its failure mode is threshold gaming — once a tier boundary carries a costly response, people have an incentive to record cases just under the line, and the tier that was meant to measure risk instead measures reporting behavior, a textbook instance of Goodhart's law.[n1] The guarding discipline is to decouple the reporter from the tier assignment where possible, audit the distribution of cases for suspicious bunching just below cut-points, and keep the boundary on responses firm so a gamed tier still never becomes a falsified outcome.
How it implements the components¶
threshold_band_partition— it defines the proximity tiers and their cut-points, assigning each case to exactly one band.learning_update_boundary— each tier carries an explicit, bounded set of permitted responses and a prohibition on manufacturing a completed harm, reward, or blame.
It does NOT implement signal_transformation_rule or value_delta_frame — turning a case's value gap into a bounded, example-level learning signal is Almost-Reward Annotation's job; this policy maps a band to an organizational response, not a case to a signal magnitude.
Related¶
- Instantiates: Counterfactual Proximity Signal Calibration — supplies the pre-committed, bounded response the archetype's proximity signal is meant to drive.
- Consumes: Near-Miss Distance Scorecard supplies the proximity score that decides which tier a case falls in.
- Sibling mechanisms: Almost-Reward Annotation · Threshold Band Map · Close-Call Review Protocol · Counterfactual Plausibility Filter · Counterfactual Value-Delta Table · Near-Miss Distance Scorecard · Proximity Signal Backtest · Regret-Weighted Decision Log · Salience Overweighting Check
Editorial Notes¶
Form Classification¶
Form family: Rule, Policy & Commitment
Rationale: Near-Miss Response Tier operates as a standing rule, threshold, contractual commitment, or policy constraint governing future conduct because it a standing policy that maps a case's proximity band to a bounded, graduated response — monitor, review, redesign, escalate — without ever booking it as a completed loss.
Independent corroboration: The frozen evidence defines Near-Miss Response Tier as 'A standing policy that maps a case's proximity band to a bounded, graduated response — monitor, review, redesign, escalate — without ever booking it as a completed loss', so its operative form is Rule, Policy & Commitment.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Engineering & Design
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Safety engineering supplies the practice of grading near misses by proximity and linking severity to bounded corrective action.
Related originating lineages:
- Disaster Management & Risk Reduction — Emergency-management escalation levels contribute the monitor-to-escalate response ladder.
- Public Administration & Policy — Policy design contributed precommitted response bands, escalation authority, and safeguards against discretionary overreaction.
- Statistics & Experimental Design — Measurement practice shaped calibrated thresholds and the need to monitor gaming around cut points.
Review resolution: Both independent reviews agree on primary origin engineering_design; reconciliation resolves reported_ambiguity, alternate_origin_disagreement. Formative alternate lineages retained: public_administration_policy, statistics_experimental_design, disaster_management. 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 medium, and encyclopedia_synthesis=true preserves the reviewers' boundary judgment.
Attribution caveat: The exact tier policy is a synthetic governance layer over established near-miss assessment. The tier rule is an encyclopedia synthesis over common risk-escalation practices rather than a uniquely named historical method.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; medium confidence.
Notes¶
[n1] Goodhart's law — "when a measure becomes a target, it ceases to be a good measure." Once a proximity tier gates a costly response, the recorded proximity becomes a target and invites gaming (bunching cases just under a cut-point), which is why tier assignment should be insulated from the party the response falls on. ↩