Transparency HubPublic Audit & Compliance

Execution Control Infrastructure
for Autonomous Systems

Every decision logged. Every rule inspectable. Every verdict reproducible. TEOS Sentinel is a deterministic, auditable governance engine — not a black box.

25 Deterministic Rules BLOCK / WARN / ALLOW Publicly Auditable JSON Schema No ML Black Box
01 / Engine Architecture

Deterministic Rule Engine v2.4

TEOS Sentinel uses a purely deterministic regex-based rule engine. No ML inference. No probabilistic verdicts. Every rule is a named, versioned, publicly inspectable pattern with explicit severity and scoring. The same input will always produce the same output — guaranteeing reproducibility for audit and compliance.

Engine FlowINPUT: command string ↓ SCAN: 25 named rules (regex patterns) ↓ SCORE: highest-severity match wins (0-100) ↓ VERDICT: BLOCK (>=80) | WARN (40-79) | ALLOW (0-39) ↓ OUTPUT: { verdict, score, rule, severity, reasons, timestamp } ↓ PERSIST: appended to events.json (rotating, max 500)

All 25 Rules

02 / Public API

Machine-Readable Endpoints

All endpoints return JSON. CORS enabled for cross-origin access. No authentication required for read endpoints.

POST/scan

Submit a command or code snippet. Returns deterministic verdict with rule match, score, and reasoning. Body: { "command": "..." }

GET/stats

Aggregated scan counters: total, blocked, warned, allowed, block rate, top triggered rules.

GET/events

Paginated event log. Query params: ?page=1&limit=100&verdict=block. Returns recent scan events with full verdict details.

GET/audit

Compliance export. Last 200 events reversed. Includes engine version, rule count, and generation timestamp.

GET/health

Engine health check. Returns status, engine version, rule count, uptime, and current time.

GET/rules.json

Static machine-readable rule definitions. 25 rules with regex patterns, severity levels, scores, and categories.

GET/test-cases.json

37 attack simulation test cases with expected verdicts. Use to validate engine correctness.

GET/audit-example.json

Sample audit log entries showing the schema, field descriptions, and example records.

03 / Verdict Schema

Scan Response Format

Every POST /scan response follows this deterministic schema. Fields are guaranteed present for valid input.

FieldTypeDescription
verdictstring"BLOCK" | "WARN" | "ALLOW" | "ERROR"
scorenumberRisk score 0-100. Higher = more dangerous.
rulestringMatched rule identifier. Format: "R{id}.{NAME}" e.g. "R01.DESTRUCTIVE_SHELL"
ruleIdstringShort rule ID: "R01" through "R25", or "R00" for clean.
severitystring"critical" | "high" | "medium" | "none"
reasonsstring[]Human-readable explanations for the verdict.
commandstringThe original input command (echoed back).
timestampstringISO 8601 timestamp of the scan. UTC.
Example Response{ "verdict": "BLOCK", "score": 100, "rule": "R01.DESTRUCTIVE_SHELL", "ruleId": "R01", "severity": "critical", "reasons": [ "rm -rf permanently destroys all filesystem data", "Wiper malware signature detected" ], "command": "rm -rf /", "timestamp": "2026-05-07T03:14:22.000Z" }
04 / Compliance & Governance

Auditability Guarantees

TEOS Sentinel is designed for environments that require provable, deterministic security decisions. Key compliance properties:

Deterministic

Same input always produces the same output. No randomness. No model drift. Fully reproducible for auditors.

Publicly Inspectable

All 25 rules, 37 test cases, and audit schemas are available as static JSON. Anyone can verify engine behavior.

Event Persisted

Every scan decision is logged with timestamp, rule, score, and verdict. Rotating store capped at 500 entries.

Export Ready

Audit logs exportable as CSV or JSON. Compatible with SIEM ingestion and compliance review workflows.

No ML Dependency

Zero ML inference. No training data, no model weights, no probabilistic outputs. Pure pattern matching.

Versioned Engine

Engine version reported in every endpoint. Rule updates are additive — existing patterns are never silently changed.

05 / Bug Bounty

Circumvent a Rule, Claim a Reward

Found a bypass? We reward responsible disclosure. If you can craft a command that matches a threat pattern but gets an ALLOW verdict, report it.

$500
Critical Bypass
$100-250
Rule Evasion
$50
False Positive
📧 Report a Bypass →