S14 — Integrity Monitor

Automated detection of logical traps, coercion patterns, and rule drift. Advisory output only — warnings, not blocks (unless mapped to a security invariant).

Detections

Circular logic

Scans for: reputation used to justify the event that generated it; rules referencing each other’s output; token promotion depending on conditions it creates. Method: directed dependency graph, flag cycles.

Coercion traps

Identifies: all available actions lead to reputation loss; forking blocked or penalized; dispute filing effectively impossible. Method: enumerate participant actions, compute outcomes, flag all-negative or empty action spaces.

Axiom drift

Monitors: cumulative parameter changes approaching constitutional limits; sequences of small changes that would violate an axiom if applied together. Method: sliding window tracking over 6-month periods.

Advisory roles

Role Access Can execute
Translator Read input Suggest commands, cannot execute
Sentinel Read event queue + rule set Detect patterns, cannot veto
Guide Read state Explain state, cannot modify

Output

{check, result: PASS|WARN, severity, details, evidence, decision_hash, reasoning_trace}

decision_hash = SHA-256(check + input + result + model_identity). Full reasoning trace recorded for auditability.

When advisory becomes enforcement

WARN → hard block only when the detection maps to a rule engine invariant:

  • Circular logic in rule update → rule rejected
  • Coercion in admission gate → event rejected
  • Axiom drift beyond limits → governance proposal rejected

Implementation Status

Runtime: Verified against donor AMS Node.js (projects/unified-mcp/src/); not yet ported to Colibri Node.js

Verified against source: 2026-04-06

Claim Status Notes
Circular logic detection (dependency graph, flag cycles) Spec-only No directed dependency graph scanner or cycle detector found in src/security/ or elsewhere
Coercion trap detection (enumerate actions, compute outcomes) Spec-only No action-space enumeration or all-negative outcome detection in source
Axiom drift monitoring (sliding window, 6-month periods) Spec-only No parameter-change tracking or sliding window monitor found
Advisory roles: Translator, Sentinel, Guide Spec-only No role definitions matching these names exist in source; src/security/ has audit/validator/sanitizer only
Output format {check, result: PASS|WARN, severity, ...} Spec-only No integrity-check output envelope found; src/security/audit.js returns different structures
decision_hash = SHA-256(check + input + result + model_identity) Spec-only No decision_hash or reasoning_trace field found anywhere in source
WARN to hard block escalation (rule engine invariant mapping) Spec-only Security modules detect threats but do not implement WARN-to-block escalation logic
General security scanning (SQL injection, secrets, path traversal) Implemented src/security/audit.js and audit-comprehensive.js provide robust pattern-based detection
Input validation and sanitization Implemented src/security/validator.js (Zod-based) and sanitizer.js provide production-ready input security

Back to top

Colibri — documentation-first MCP runtime. Apache 2.0 + Commons Clause.

This site uses Just the Docs, a documentation theme for Jekyll.