Response Matrix¶
Decision table — instantiates Proportionality Calibration
Maps severity, recurrence, intent, impact, and context to response ranges so decision-makers can see the proportional response band.
A Response Matrix is a multi-axis grid whose cells hold a range of proportionate responses — a band — for each combination of case features. You locate a case by its severity, its recurrence, its intent and impact, and its context, read down to the intersecting cell, and find not a single verdict but the zone a fitting response must fall within. Its defining move, and what separates it from its nearest twin, is that it deliberately outputs a band rather than a point: the matrix shows the decision-maker where proportional lives for this class of case and leaves the final choice inside that band to judgment. Context is a first-class axis, not a footnote — the same severity can land in a different band once intent or impact shifts. The matrix's job is to make the proportional range visible and to keep judgment inside it, not to remove the judgment.
Example¶
An AI platform runs a decision matrix for model-safety incidents. One axis is severity of the surfaced harm (from a low-quality answer up to output that could enable serious real-world damage); a second is user exposure (an internal tester, a small cohort, the full user base); a third is context (a jailbreak against guardrails versus a spontaneous failure, a first occurrence versus a recurring class). An incident — a guardrail bypass producing dangerous instructions, caught at small-cohort exposure, recurring for the third time — lands in a cell whose band reads: "human-review gating through throttling of the affected capability; rollback if recurrence continues." That band, not a single order, is the output. The on-call lead still chooses within it — gating plus a targeted throttle rather than a full rollback — because the matrix's role is to bound the response, showing that a full decommission would be above the band and a mere monitoring note below it. Move the same harm to full-user exposure and the cell's whole band shifts upward.
How it works¶
- Define the axes. Severity, recurrence, intent/impact, and context each become a dimension, so a case is placed by several features at once rather than by one label.
- Populate cells with bands, not points. Each intersection holds a low-to-high range of proportionate responses; the width of the band signals how much discretion the class warrants.
- Locate and read the band. A case is positioned on every axis; the intersecting cell yields the response range within which any fitting answer must sit.
- Shift the band on recurrence and context. Repeat occurrences and aggravating context move the case to a higher cell, so the escalation is a visible jump between bands rather than a hidden decision.
Tuning parameters¶
- Axis count — how many case features are dimensions. More axes capture context finely but multiply cells and can outrun the evidence available to place a case.
- Band width — how wide the range in each cell is. Wide bands preserve discretion but risk laundering an arbitrary call; narrow bands constrain tightly but approach a fixed verdict.
- Severity granularity — how many rows the severity axis carries. Fine granularity distinguishes cases well but invites false precision at the boundaries.
- Recurrence weighting — how strongly a repeat occurrence shifts the case upward. Heavy weighting deters patterns but can over-escalate a scattered history.
When it helps, and when it misleads¶
Its strength is that it makes context legible and bounds discretion without erasing it: two deciders reading the same case reach the same band, over-reaction and under-reaction both show up as choices outside the band, and the recurrence axis turns escalation into an auditable jump. It is genuinely better than a single-column penalty list precisely because it refuses to collapse a multi-feature case into one number.
Its failure mode is the well-documented flaw of any such grid: distinct cases get compressed into the same cell, so a matrix can assign identical treatment to situations that differ materially while lending the whole thing a false air of objectivity.[1] The classic misuse is treating the band as the decision — picking the midpoint mechanically and calling it proportional, which is "proportionality theater" with extra axes. The guarding discipline is to keep bands only as wide as real discretion warrants, to require a written reason for the point chosen inside a band, and to review whether cases are piling up at cell boundaries where the axes are too coarse.
How it implements the components¶
severity_assessment— the severity axis is an explicit rubric that places each case on a graded scale before any response band is read.context_adjustment_factor— intent, impact, and situational context are first-class axes, so the same severity can resolve to different bands as context changes.escalation_threshold— the recurrence axis defines the crossings at which a case jumps to a higher-response cell, making escalation a visible, reasoned step.
It does not enumerate the concrete penalty ladder or hold cases to precedent — fixing a specific sanction level per conduct category and anchoring it to response_scale and precedent_comparator is the job of its nearest twin, the Sanction Matrix, which prescribes a point where this matrix shows a band.
Related¶
- Instantiates: Proportionality Calibration — it renders the proportional band across a case's features so discretion stays bounded and visible.
- Sibling mechanisms: After-Action Proportionality Review · Disciplinary Guideline · Moderation Strike System · Proportional Remedy Menu · Risk-Based Enforcement Protocol · Sanction Matrix · Least-Restrictive Alternative Screen
Editorial Notes¶
Form Classification¶
Form family: Decision, Gate & Allocation
Rationale: Response Matrix operates by places each case across severity, recurrence, intent, and context and selects a proportional response band. That concrete deployed or enacted form is Decision, Gate & Allocation under the frozen taxonomy.
Nearest alternative: Representation, Specification & Plan — Although Representation, Specification & Plan can support this mechanism, the frozen evidence makes its operative form the act that places each case across severity, recurrence, intent, and context and selects a proportional response band; the alternative is therefore secondary rather than defining.
Review outcome: Adjudicated after independent review; high confidence.
Origin Attribution¶
Primary origin: Law & Governance
Origin pattern: Convergent development
Present-day reach: Multi-domain
Rationale: Mapping severity, intent, recurrence, and impact to proportional sanctions is rooted in legal and disciplinary governance.
Related originating lineages:
- Organizational & Management Science — Workplace discipline and incident management materially adapted proportional bands.
- Public Administration & Policy — Enforcement agencies independently developed standardized response matrices.
Review resolution: Both blind reviewers agree that law_governance is the primary historical origin. Explicit reconciliation of alternate origin disagreement, domain reach disagreement adopts reviewer_a's evidence: Mapping severity, intent, recurrence, and impact to proportional sanctions is rooted in legal and disciplinary governance. The selected record uses alternates=organizational_management, public_administration_policy, origin_mode=convergent, and domain_reach=multi_domain; the other review proposed alternates=criminology_forensic, organizational_management, public_administration_policy, origin_mode=convergent, and domain_reach=universal. The selected combination better preserves the mechanism-specific formative lineages and calibrated scope; broader present-day use is not treated as proof of additional historical origin.
Review outcome: Reconciled after independent review; high confidence.
References¶
[1] Cox, L. A., Jr. "What's Wrong with Risk Matrices?". Risk Analysis 28(2), 497–512 (2008). Shows that typical risk matrices can compress quantitatively different risks into the same rating and thereby misdirect treatment decisions. registry ↩