AI Governance · Decision Logging

AI Decision Logging Platform

Cryptographic dataset certification answers the question "what data was this AI trained on?" Decision logging answers "how did this AI reach this output?" Together, they close the provenance chain that EU AI Act Article 12 requires for high-risk AI systems.

The Complete AI Provenance Chain

Dataset generationSynthetic data created via CTGAN

CertificationSHA-256 fingerprint + Ed25519 signature issued

Transparency logCertificate appended to append-only public log

Model trainingAI system trained on certified dataset

DeploymentModel deployed to production

Decision loggingOperational decisions logged with certificate reference

AuditFull chain verifiable: data → model → decision

Decision Record Structure

Each decision log entry references the certification ID of the dataset used to train the model that produced the output. This creates a verifiable link from the operational decision back to the certified training data.

{
  "decision_id": "uuid",
  "timestamp": "2026-03-19T00:00:00Z",
  "actor": "model-v2.1",
  "decision_label": "credit_risk_assessment",
  "confidence": 0.94,
  "certificate_ref": "cert-uuid",         // links to training data cert
  "input_hash": "sha256:...",              // hash of the input data
  "output_hash": "sha256:...",             // hash of the model output
  "signature": "ed25519:base64url...",    // tamper-evident
  "public_hash": "sha256:...",            // public log entry hash
  "sequence": 18203                       // position in public decision log
}

The certificate_ref field is a foreign key into the CertifiedData transparency log — allowing any auditor to traverse from a decision back to the exact dataset bytes that informed the model's training.

EU AI Act Compliance Coverage

Art. 12(1)

Automatic logging of AI system events

Decision log records each inference event automatically via API call

Art. 12(2)

Logging of reference database access

certificate_ref field records which certified dataset was the reference data

Art. 13(1)

Transparency toward deployers

Public decision log is verifiable by deployers without provider involvement

Art. 19(1)

10-year retention of technical documentation

Decision records retained indefinitely; certificate references remain valid

Art. 26(6)

Deployer logging of operating periods

Decision log timestamps + sequence numbers satisfy operating period traceability

Integration Pattern

Decision logging integrates at the inference layer — after the model produces an output, before the response is returned to the end user. The logging call is a single API request that does not add measurable latency to the inference path.

POST /api/decisions
Authorization: Bearer <api_key>

{
  "actor": "credit-model-v2",
  "decision_label": "credit_risk_assessment",
  "confidence": 0.94,
  "certificate_ref": "cert-uuid-of-training-dataset",
  "input_hash": sha256(JSON.stringify(inputRecord)),
  "output_hash": sha256(JSON.stringify(modelOutput))
}

The platform signs the decision record with the CertifiedData Ed25519 key, assigns a sequence number in the public decision log, and returns the signed artifact. The artifact can be retained as the operational compliance record for that decision event.

Generate a certified dataset →Decision lineage overview View public decision log

Article 12 logging obligations →Compliance guide →AI compliance logging →Technical reference →