Diagnostics

System Audit & Guardrail Synthesizer

Evaluates AI system prompts, architectures, and automated decision logic against 80+ governance failure modes and synthesizes executable guardrail code and legal SLA clauses.

Overview

When to use the System Audit & Guardrail Synthesizer.

AI engineers, compliance leads, and product architects designing high-stakes decision systems.

  • Locate system prompt or decision policy documentation.
  • Identify escalation and dispute handling workflows.

Estimated time: 5 minutes

Scholarly metadata

Authorship

Contact: diagnostics@ethotechnics.org

Publication details

  • Published: Dec 3, 2025
  • Last updated: Jan 9, 2026
  • Version: v1.1.0
  • DOI: Pending Zenodo deposit

License: CC BY 4.0

Credit Ethotechnics Institute Diagnostics Lab, include tool name + version, and link to the canonical permalink.

Archive snapshot: Wayback capture

Changelog

  • v1.1.0 · 2026-01-09 — Published method cards, transparency notes, and replicability guidance for each diagnostic.
  • v1.0.0 · 2025-12-03 — Initial diagnostics suite release.

Copy citation (APA/BibTeX)

Cite this page Formats: APA, MLA, Chicago, BibTeX, RIS

Version

v1.1.0

Last updated

Jan 9, 2026

DOI

Pending Zenodo deposit

APA

Ethotechnics Institute Diagnostics Lab. (2026). System Audit & Guardrail Synthesizer. Ethotechnics Institute. https://ethotechnics.org/diagnostics/system-auditor

MLA

Ethotechnics Institute Diagnostics Lab. "System Audit & Guardrail Synthesizer." Ethotechnics Institute, 2026, https://ethotechnics.org/diagnostics/system-auditor.

Chicago

Ethotechnics Institute Diagnostics Lab. "System Audit & Guardrail Synthesizer." Ethotechnics Institute. Jan 9, 2026. https://ethotechnics.org/diagnostics/system-auditor.

BibTeX

@misc{diagnostic_system-auditor,
  title={System Audit & Guardrail Synthesizer},
  author={Ethotechnics Institute Diagnostics Lab},
  year={2026},
  howpublished={Ethotechnics Institute},
  url={https://ethotechnics.org/diagnostics/system-auditor},
  version={v1.1.0}
}

RIS

TY  - WEB
TI  - System Audit & Guardrail Synthesizer
AU  - Ethotechnics Institute Diagnostics Lab
PY  - 2026
UR  - https://ethotechnics.org/diagnostics/system-auditor
ER  -

Methodology

Method, transparency, and replicability.

Automated evaluation against 80+ Ethotechnics failure modes, quantitative SLA bounds, and executable guardrail generation.

Inputs

  • System prompt, policy rules, or workflow specification.
  • Autonomy tier (Advisory, Semi-Autonomous, Autonomous).
  • Domain hazard tier (Low, Medium, High, Critical).

Procedure

  1. Parse text against Ethotechnics structural failure classifiers.
  2. Compute calibrated quantitative SLA bounds.
  3. Synthesize TypeScript, Python, and JSON Schema middleware.

Outputs

  • Governance Health Score (0-100) and risk level.
  • Detected failure mode cards with specific remedies.
  • Copyable production guardrail code and contract clauses.

Measures

  • Vulnerability to Unearned Closure, Administrative Shame, and Dead-User Zones.
  • Quantitative Time-to-Halt (TTH) and Reversal SLA ceilings.
  • Compliance with STD-01 Temporal Rights & Recourse standards.

Does not measure

  • Raw model inference token throughput or latency.
  • Underlying training dataset copyright clearances.
  • Adversarial jailbreaks unrelated to governance architecture.

Assumptions

  • System prompts or architecture specs reflect actual deployed logic.
  • Domain hazard tier accurately captures end-user stakes.
  • Decisions produce observable downstream state transitions.

Instrument prompts

  • Paste system prompt or load an industry preset.
  • Select autonomy tier and domain hazard stakes.
  • Review identified breach vectors and SLA limits.

Rubric

  • Critical Risk: missing contestability or irreversible denial.
  • Elevated Risk: unilateral closure or burden shifting.
  • Low Risk: verified claimant confirmation and active rollback lanes.

Scoring logic

  • Base 100 with weighted risk penalties per failure vector.

Validation notes

Tested against customer support, clinical benefits triage, credit underwriting, and content moderation pipelines.

Deterministic rule classification across 80+ Ethotechnics failure taxonomy classes.

  • Unearned Closure
  • Dead-User Zones
  • Administrative Shame
  • Heroism-Dependent Systems

Replicability

  • Open the System Auditor workbench.
  • Select or enter your system prompt.
  • Export the verified Decision Object JSON receipt.

Example outputs

  • Audit Report JSON receipt.
  • TypeScript Express/Fastify guardrail middleware.
  • Contract-ready legal SLA clauses.

Sample output

Preview the audit readout.

Governance Health Score, detected breach vectors, calibrated SLA limits, and copyable code.

View methodology

Run the tool

Audit an AI system prompt or architecture spec.

Select an industry preset or paste your own system prompt to generate guardrails and contract terms.

1. System Architecture & Prompt Spec
Load Industry Preset
47

Elevated Risk

4 potential governance breach vectors detected

Recommended Quantitative SLA Thresholds
Time-to-Halt (TTH)30sMax automated execution before brake
Reversal SLA48hHuman contestability turnaround
User Burden Cap3 stepsMax proof steps before inversion
Human Saturation Ceiling10%Max allowable review fallback rate
Detected Failure Modes & Required Remediations
Unearned Closurehigh risk

System unilaterally marks disputes or cases as resolved without verified recipient affirmation or relief invariant checks.

Remedy: Implement verified claimant confirmation before case closure; start a 7-day Graceful Rollback Window.
Administrative Shame & Burden Shiftinghigh risk

Offloads evidentiary burden onto claimants while maintaining low institutional effort (cost assignment / fragility subsidy).

Remedy: Apply the Burden Inversion Rule: when system confidence is below 95%, the system must gather missing proofs or default in claimant favor.
Heroism-Dependent Escalationhigh risk

Escalation pathways require extreme measures (legal threats, panic) rather than predictable operational capacity triggers.

Remedy: Establish automated escalation horizons when model confidence falls below threshold or user frustration markers register.
Affect-Invariance Violationmedium risk

Conditioning relief or engagement on customer composure, emotional patience, or repetitive policy citations.

Remedy: Enforce affect-invariance: baseline safety and appeal rights must remain identical regardless of user emotional state.
Synthesized Guardrail Code & Contract Clauses
import { z } from "zod";
import type { Request, Response, NextFunction } from "express";

/**
 * Ethotechnics Governance Guardrail Middleware
 * System: Autonomous Customer Support & Refunds Agent
 * Target SLA: Time-to-Halt <= 30s, Reversal SLA <= 48h
 * Mitigated Failure Vectors:
 * - Unearned Closure (high risk)
 * - Administrative Shame & Burden Shifting (high risk)
 * - Heroism-Dependent Escalation (high risk)
 * - Affect-Invariance Violation (medium risk)
 */

// 1. Decision Object Schema enforcing Contestability
export const DecisionObjectSchema = z.object({
  decisionId: z.string().uuid(),
  systemId: z.literal("autonomouscustomersupportrefundsagent"),
  timestamp: z.string().datetime(),
  claimantId: z.string().min(1),
  actionClass: z.enum(["ADVERSE", "PERMISSIVE", "INTERIM_HOLD"]),
  modelConfidence: z.number().min(0).max(1),
  reasons: z.array(z.string()).min(1),
  bindingClock: z.object({
    startedAt: z.string().datetime(),
    reversalDeadline: z.string().datetime(),
    timeToHaltTargetSeconds: z.literal(30),
  }),
  appealPath: z.object({
    endpoint: z.string().url(),
    maxSteps: z.number().max(3),
    burdenInversionActive: z.boolean(),
  }),
});

export type DecisionObject = z.infer<typeof DecisionObjectSchema>;

// 2. Ethical Circuit Breaker & Safe-Pause Middleware
export function createEthotechnicGuard(config = { safePauseThreshold: 0.85 }) {
  let rollingReversals = 0;
  let totalEvaluations = 0;

  return async (req: Request, res: Response, next: NextFunction) => {
    const startTime = Date.now();

    // Check circuit breaker status
    if (totalEvaluations > 50 && (rollingReversals / totalEvaluations) > 0.3) {
      return res.status(503).json({
        error: "ETHOTECHNIC_CIRCUIT_BREAKER_TRIGGERED",
        message: "Excessive decision reversal rate detected. System safe-paused.",
        safePauseActive: true,
      });
    }

    res.on("finish", () => {
      const durationSec = (Date.now() - startTime) / 1000;
      if (durationSec > 30) {
        console.warn(`[SLA BREACH] Time-to-Halt exceeded: ${durationSec}s > 30s`);
      }
    });

    next();
  };
}