Graduated Response Matrix¶
Decision matrix — instantiates Self-Targeting Defense Guardrail
A lookup table that maps classifier confidence and self-status ambiguity against response harm, so uncertain judgments are routed to weaker, more reversible actions.
A defense that answers every threat with its heaviest hammer is one false positive away from an autoimmune disaster. Graduated Response Matrix is the explicit lookup table that stops this: on one axis, how confident and unambiguous the threat judgment is; on the other, how much harm and irreversibility each available response carries; and in every cell, the most the defense is permitted to do at that level of certainty. Its defining idea is that response intensity must be a monotone function of evidence quality — high confidence unlocks strong action, low confidence caps the system at monitoring or a reversible challenge — so uncertainty shrinks the permitted response rather than being resolved by escalation. It is a graded permission table across the whole space of confidence and harm; it does not itself perform any single containment step, and it does not require two approvers — it decides which rung is even eligible.
Example¶
An ISP's DDoS-mitigation system watches traffic to a customer's hosted service and must choose between doing nothing, rate-shaping the suspect flows, challenging them, or null-routing the destination entirely — the last of which takes the customer's whole service offline, a heavy and blunt response. The Graduated Response Matrix governs the choice. A high-confidence, unambiguous volumetric attack from spoofed sources unlocks aggressive filtering. But a moderate-confidence anomaly that might be a legitimate traffic surge — a product launch — lands in a cell whose ceiling is rate-shaping plus an interactive challenge, never a null-route. The matrix simply forbids the nuclear option at that confidence level.
Setup to outcome: when a flash crowd from a viral post produced a mid-confidence anomaly, the matrix capped the response at shaping, and the legitimate surge got through degraded rather than blackholed. The distinction that matters is that the ceiling was set by the evidence quality, not by the on-call engineer's nerve under pressure.
How it works¶
- Two axes, explicit cells. Confidence-and-self-status-ambiguity crosses harm-and-reversibility; each cell names the strongest permissible response, pre-decided rather than improvised in the moment.
- Rank responses by harm and reversibility. Every available action is placed on the harm axis by how much burden it imposes and how hard it is to undo — monitoring at one end, irreversible engagement at the other.
- Carry the uncertainty state, don't collapse it. The classifier hands the matrix confidence, evidence independence, and ambiguity — not a bare hostile/benign bit — so ambiguous self-status is visible and pushes the lookup toward the low-harm rows.
- Monotonicity as an invariant. The table is constructed so that lower confidence can never map to a higher-harm ceiling than higher confidence — the guardrail against escalating on doubt.
Tuning parameters¶
- Grid resolution — how many confidence bands and harm tiers the matrix distinguishes. Finer grids match response to evidence precisely but are harder to validate and can imply false precision; coarser grids are robust but blunt.
- Cell ceilings — the strongest action allowed per cell. Raising ceilings buys speed and coverage against threats; lowering them protects self but lets more real threats operate under a capped response.
- Ambiguity penalty — how sharply self-status uncertainty pulls the lookup toward reversible rows. A steep penalty is protective but can neuter the defense whenever attackers manufacture ambiguity.
- Escalation gate — what new evidence is required to move up a row. Tight gates prevent creep toward heavy responses; loose gates let the system drift back to hammer-for-everything.
When it helps, and when it misleads¶
Its strength is that it encodes proportionality[n1] as a table instead of a hope: the response is bounded by evidence quality by construction, and the dangerous cell — low confidence, high harm — is simply left empty. It turns "use judgment" into an auditable artifact everyone can inspect before an incident, and it makes the escalation logic explicit rather than emergent from operator adrenaline.
Its failure mode is that a matrix is only as honest as its confidence axis. If the classifier's reported confidence is miscalibrated — overstated, as they often are — then well-designed cells still authorize heavy harm on thin evidence, because the lookup trusts a number that lies. Adversaries also learn the grid and engineer their behavior to land in a permissive cell. The classic misuse is a matrix whose ceilings creep upward over time under pressure to "be decisive," until low-confidence cells quietly permit irreversible action and the graduation is gone. The guarding discipline is to validate the confidence axis against real outcomes and to freeze the low-confidence/high-harm cells as permanently empty, no matter the operational push.
How it implements the components¶
harm_intensity_and_reversibility_map— the harm axis is the ranked map of responses by burden and reversibility, from monitoring to irreversible engagement.threat_classifier_with_uncertainty_state— the confidence axis consumes and preserves the classifier's uncertainty, base-rate context, and self-status ambiguity rather than a binary label.response_authorization_gate— each cell's ceiling is the authorization: it defines the strongest response permitted at that evidence level, gating intensity on certainty.
It does not carry out the actual reversible holding action it may select — taking the concrete reversible containment step before anything destructive is the reversible_containment_ladder, which belongs to Quarantine-Before-Destroy Rule; this matrix chooses the rung, that rule executes the reversible one.
Related¶
- Instantiates: Self-Targeting Defense Guardrail — supplies the confidence-to-harm mapping that keeps response intensity bounded by evidence.
- Consumes: Self-Status Cross-Check — the self-status verdict feeds the ambiguity axis of the lookup.
- Sibling mechanisms: Appeal and Rapid Restoration Workflow · Engagement Kill Switch · False-Positive Harm Budget Dashboard · Post-Incident Autoimmune Review · Protected-Self Allowlist with Expiry · Quarantine-Before-Destroy Rule · Self-Status Cross-Check · Shadow Mode and Canary Enforcement · Two-Key High-Harm Engagement
Editorial Notes¶
Form Classification¶
Form family: Rule, Policy & Commitment
Rationale: Graduated Response Matrix operates as a standing rule, threshold, contractual commitment, or policy constraint governing future conduct because it a lookup table that maps classifier confidence and self-status ambiguity against response harm, so uncertain judgments are routed to weaker, more reversible actions.
Independent corroboration: The frozen evidence defines Graduated Response Matrix as 'A lookup table that maps classifier confidence and self-status ambiguity against response harm, so uncertain judgments are routed to weaker, more reversible actions', so its operative form is Rule, Policy & Commitment.
Nearest alternative: Decision, Gate & Allocation — The matrix routes each uncertain classification to a bounded response, but its strongest-permissible cells are standing constraints.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Law & Governance
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Legal proportionality ties response severity to evidence, certainty, and expected harm.
Related originating lineages:
- Security Studies & Intelligence Analysis — Threat triage materially operationalizes confidence and ambiguity against response reversibility.
- Ethics of Technology & AI Governance — Classifier uncertainty and self-targeting risks shape its contemporary automated-system form.
Review resolution: Both reviewers agree that law_governance is primary: Legal proportionality ties response severity to evidence, certainty, and expected harm. I retain security_intelligence, tech_ethics_ai_governance only as formative lineage, not as a list of later applications. I resolve origin_mode as cross_disciplinary_synthesis because the artifact joins distinct disciplinary contributions. I resolve domain_reach as multi_domain because it transfers across several fields but is not a domain-free primitive. Encyclopedia synthesis is true because the exact generalized packaging is an encyclopedia-authored combination or refinement.
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¶
[n1] Proportionality — the principle that the severity of a response must be commensurate with the severity and certainty of the threat it answers. Long-standing in law and just-war doctrine, it is exactly the invariant this matrix mechanizes: no cell may authorize harm out of proportion to the evidence in hand. ↩