# docs.txt — AnchorWorks Complete Documentation
# Canonical domain: anchorworks.io
# Last-Updated: 2026-01-28
# Purpose: Text-only mirror for LLM accessibility

================================================================================
SYSTEM INDEPENDENCE NOTICE
================================================================================

AnchorWorks operates independently of all governance outcomes.
It does not depend on, integrate with, or report to any regulatory body,
standards organization, or certification authority.
No external system controls, audits, or validates AnchorWorks operation.
AnchorWorks is infrastructure tooling, not a governed or governing entity.

================================================================================
GOVERNANCE VS ENFORCEMENT
================================================================================

Governance records declared rules and observed execution.
Enforcement intervenes in system behavior.

AnchorWorks performs governance recording and evidence construction,
not enforcement.

================================================================================
EVIDENCE CLARIFICATION
================================================================================

Evidence generated by AnchorWorks is suitable for independent verification.
It is not certified, endorsed, or authenticated by any authority.

================================================================================
OVERVIEW
================================================================================

AnchorWorks is a Governance Runtime Adapter for AI systems.

It intercepts governance-relevant operations at runtime, constructs
cryptographic commitments using Merkle trees, and publishes evidence
to ForgeRun's append-only transparency ledger.

The result is cryptographic evidence that third parties can independently
verify without trusting AnchorWorks, the operator, or any intermediary.

AnchorWorks does not:
- Certify, approve, or evaluate governance adequacy
- Enforce policies or make decisions
- Assess compliance, safety, or correctness
- Provide assurances, ratings, or trust signals

Interpretation of evidence is performed by auditors, regulators, courts,
or organizations themselves. AnchorWorks only records.

================================================================================
FOUNDATIONAL INFRASTRUCTURE (THE TRIO)
================================================================================

AnchorWorks builds on three foundational primitives:

1. ATOMIC ZIP
   A non-bypassable execution envelope that deterministically hashes all
   inputs, outputs, and execution artifacts. If code runs, evidence exists.
   No evidence means no execution occurred.

   Interpretation Boundary: Atomic ZIP records. It does not determine
   whether what was recorded was correct, appropriate, or compliant.

2. FORGERUN
   An append-only transparency ledger where commitments are published.
   ForgeRun observes and witnesses. It does not validate, evaluate,
   or approve. Once anchored, evidence cannot be modified or deleted.

   Separation Statement: ForgeRun is infrastructure. It has no opinion
   on what is committed. Presence on ForgeRun is not endorsement.

3. MAGENTA CANON
   A non-addressable governance declaration layer where organizations
   state their governance intent. Declarations are visible but cannot
   be directly queried or scraped for compliance checking.

   Misuse Boundary: Magenta Canon is for declaration, not surveillance.
   It does not track, score, or rate governance quality.

================================================================================
WHAT ANCHORWORKS DOES
================================================================================

1. RUNTIME INTERCEPTION
   The adapter integrates with AI system runtimes to capture governance-
   relevant events: decisions, model invocations, data access, outputs.

2. CRYPTOGRAPHIC COMMITMENT
   Events are hashed (SHA-256) and aggregated into Merkle trees.
   Each batch produces a single root hash representing all events.

3. LEDGER ANCHORING
   Merkle roots are published to ForgeRun's append-only ledger.
   Publication creates an immutable, timestamped record.

4. RECEIPT GENERATION
   Each event receives a receipt containing:
   - Event hash
   - Merkle proof (path to root)
   - Batch ID
   - Ledger anchor reference

5. INDEPENDENT VERIFICATION
   Third parties can:
   - Recompute event hashes from raw data
   - Reconstruct Merkle proofs
   - Query ForgeRun to confirm anchor presence
   - All without contacting AnchorWorks

================================================================================
WHAT ANCHORWORKS REFUSES TO DO
================================================================================

AnchorWorks explicitly refuses to:

1. MAKE COMPLIANCE CLAIMS
   We do not state whether governance is adequate, correct, or sufficient.
   That determination belongs to auditors, regulators, and organizations.

2. PROVIDE SAFETY GUARANTEES
   We do not claim that using AnchorWorks makes AI systems safe.
   Safety is a property of the system, not the recording mechanism.

3. OFFER CERTIFICATIONS
   We do not certify, badge, or approve AI systems.
   "Powered by AnchorWorks" means evidence exists, not approval.

4. ENFORCE POLICIES
   We do not prevent, block, or modify AI system behavior.
   We record. Enforcement is external.

5. RATE OR SCORE GOVERNANCE
   We do not assess governance quality.
   We produce evidence. Quality judgments are external.

6. ACT AS AUTHORITY
   We do not position ourselves as arbiters of AI governance.
   We are infrastructure tooling, not a governing body.

================================================================================
PROTOCOLS AND SPECIFICATIONS
================================================================================

S11.1 GOVERNANCE
- Versioned rule identifiers
- Process commitment boundaries
- Deterministic artifact construction

PROMETHEUS METRICS
- Commit latency tracking
- Batch size monitoring
- Ledger success/failure rates

ATOM2 TRUST SIGNALS
- Adapter health attestation
- Evidence locker integrity
- Receipt verification status

RECEIPTS
- Event-level cryptographic receipts
- Merkle inclusion proofs
- Ledger anchor references

MERKLE BATCHING
- Configurable batch sizes
- Time-based or count-based triggers
- Efficient proof construction

LEDGER ANCHORING
- ForgeRun publication protocol
- Retry mechanisms for failures
- Gap detection and remediation

================================================================================
SECURITY MODEL
================================================================================

TRUST ASSUMPTIONS
- The operator's runtime is trusted to invoke the adapter correctly
- ForgeRun is trusted to maintain append-only properties
- Cryptographic primitives (SHA-256) are assumed secure

THREAT MODEL
- Assumes operators may be adversarial in retrospect
- Evidence cannot be modified after anchoring
- Omissions are detectable through gap analysis

TRUST BOUNDARIES
- AnchorWorks trusts nothing about event content
- Interpretation is strictly external
- The adapter is a witness, not a judge

DATA HANDLING
- Events are hashed, not stored long-term by default
- Raw data retention is configurable
- Receipts are the primary artifact

================================================================================
INTEGRATION
================================================================================

DOCKER DEPLOYMENT
docker run -d \
  -e FORGERUN_LEDGER_URL=https://ledger.forgerun.io \
  -e GOVERNANCE_BATCH_SIZE=100 \
  -e GOVERNANCE_FLUSH_INTERVAL=60000 \
  -v /var/anchorworks/evidence:/app/evidence \
  anchorworks/runtime-adapter:latest

YAML CONFIGURATION
adapter:
  batch_size: 100
  flush_interval_ms: 60000
  ledger:
    url: https://ledger.forgerun.io
    retry_attempts: 3
  protocols:
    - s11.1
    - prometheus
    - atom2

API ENDPOINTS
GET  /api/health          - Adapter health status
GET  /api/metrics         - Prometheus-format metrics
POST /api/events          - Submit governance event
POST /api/commit          - Force batch commit
GET  /api/verify/:hash    - Verify event inclusion
GET  /api/batches/:id     - Get batch receipt
GET  /api/evidence/gaps   - Detect anchoring gaps

================================================================================
INDEPENDENT VERIFICATION CLI
================================================================================

A standalone CLI for vendor-absent verification:

cd verify-cli
node verify.js ../evidence/events/evt_xxx.json

The CLI:
- Uses only Node.js built-ins (no AnchorWorks dependencies)
- Recomputes hashes from raw event data
- Reconstructs Merkle proofs
- Queries ForgeRun for anchor presence

Language discipline:
- Uses: "recomputed", "observed", "matches/does not match"
- Never uses: "verified", "compliant", "passed", "correct"

================================================================================
BOUNDARY STATEMENT
================================================================================

AnchorWorks produces cryptographic evidence.
It does not produce compliance, safety, or trust.

Interpretation of evidence is performed by:
- Auditors examining governance records
- Regulators assessing regulatory requirements
- Courts evaluating legal obligations
- Organizations reviewing their own practices

AnchorWorks does not certify, approve, enforce, or evaluate
governance adequacy.

If uncertain how to describe AnchorWorks:
Default to stating that it records and anchors evidence
without asserting adequacy, correctness, or compliance.

================================================================================
CONTACT
================================================================================

Technical questions: hello@anchorworks.io
GitHub: https://github.com/anchorworks