Every decision. Immutable. Verifiable.
The ADJUDON Audit Trail captures every decision trace with full context, chains it cryptographically, and makes it exportable on demand. When a regulator asks "show me how this decision was made," you hand them a verifiable record — not a reconstructed narrative.
How It Works
The ADJUDON Audit Trail captures every decision trace with full context, chains it cryptographically, and makes it exportable on demand. When a regulator asks "show me how this decision was made," you hand them a verifiable record — not a reconstructed narrative.
How it works:
At write time, the audit logger computes chainHash = SHA-256(previousChainHash + entryPayload). The previousChainHash is the chainHash of the immediately preceding entry. The first entry in the chain uses a known seed value. This is the same integrity model used in blockchain systems — without the distributed consensus overhead.
One-click verification:
Call GET /api/audit/verify to verify the integrity of your entire audit chain. The endpoint walks every entry, recomputes each hash, and confirms the chain is unbroken. Returns a verification result with the total entry count, any broken links, and the timestamp of the last verified entry. Available via API and via the dashboard with a single click.
What this means for regulators:
When an auditor or regulator requests proof that your decision logs have not been altered, you provide the verification result. The cryptographic chain is the proof — not a policy document, not a promise.
Full decision context on every trace
Every trace submitted to ADJUDON generates an immutable audit record. The following fields are captured and stored for every decision — no configuration required.
Timestamp
ISO 8601, millisecond precision. Server-assigned at ingest.
Agent ID
The identifier of the AI agent that produced the decision.
Input Context
The full input payload sent to the agent — prompt, session context, user metadata, or any structured data your application provides.
Output Decision
The agent's proposed output — the decision, recommendation, or action it intended to take.
CPI Score
The Confidence Performance Index calculated by the Confidence Engine. Includes the composite score and the three pillar breakdowns (base probability, entropy, vector similarity).
Policy Matches
Every policy evaluated against the trace, with match/no-match status and the action triggered (approve, flag, block, notify).
Human Overrides
If the trace was routed to the review queue: the reviewer's identity, their decision (approve, reject, escalate), their rationale, and the timestamp of resolution.
IP Address
The originating IP address of the API call.
Request Metadata
API key scope, SDK version, workspace ID, and organization ID.
SHA-256 hash chain — tamper-evident by design
Every audit log entry is hashed using SHA-256. Each entry's hash incorporates the hash of the previous entry (previousChainHash → chainHash), forming a cryptographic chain. If any entry in the chain is modified, deleted, or reordered, the chain breaks — and the tampering is immediately detectable.
How it works:
At write time, the audit logger computes chainHash = SHA-256(previousChainHash + entryPayload). The previousChainHash is the chainHash of the immediately preceding entry. The first entry in the chain uses a known seed value. This is the same integrity model used in blockchain systems — without the distributed consensus overhead.
One-click verification:
Call GET /api/audit/verify to verify the integrity of your entire audit chain. The endpoint walks every entry, recomputes each hash, and confirms the chain is unbroken. Returns a verification result with the total entry count, any broken links, and the timestamp of the last verified entry. Available via API and via the dashboard with a single click.
What this means for regulators:
When an auditor or regulator requests proof that your decision logs have not been altered, you provide the verification result. The cryptographic chain is the proof — not a policy document, not a promise.
Export-ready for every audit scenario
The Audit Trail is designed to produce regulator-ready output on demand. No data engineering. No manual assembly.
CSV Export
`GET /api/audit/export` — Download the full audit log as a structured CSV file. Filterable by date range, agent ID, status, and policy match. Compatible with Excel, Google Sheets, and any compliance reporting tool.
PDF Export
`GET /api/audit/export/pdf` — Generate a formatted PDF report of the audit log. Includes summary statistics, individual trace records, and the hash chain verification status. Designed for submission to regulators (BaFin, data protection authorities) and for inclusion in compliance audit packs.
BaFin / GDPR DPA Ready
Export formats are structured to satisfy BaFin reporting requirements for algorithmic decision systems and GDPR Data Protection Impact Assessment (DPIA) documentation. The audit trail provides the technical evidence layer for Art. 13 transparency and Art. 35 impact assessments.
Configurable Retention
Data retention is configurable from 7 to 365 days per workspace. Default: 90 days. Deletion is automatic and irreversible after the retention window expires. Right-to-erasure requests are supported via API and dashboard.
Granular event tracking across the full decision lifecycle
The Audit Trail does not log decisions as single events. It tracks the full lifecycle — from ingest to resolution — with discrete, queryable event types.
Trace Events
trace.created—A new decision trace was ingested.
trace.approved—The trace passed all policies and was approved (201).
trace.flagged—The trace matched a flag policy and was routed to review (202).
trace.blocked—The trace matched a block policy and was rejected (403).
Policy Events
policy.matched—A policy condition set matched the trace.
policy.action.executed—The policy action (block, flag, approve, notify) was executed.
policy.alert.dispatched—An alert was sent via Slack, email, or webhook.
Review Events
review.created—A review item was auto-created from a flagged trace.
review.assigned—The review item was assigned to a reviewer.
review.approved—A human reviewer approved the decision.
review.rejected—A human reviewer rejected the decision.
review.escalated—The review item was escalated (SLA breach or manual escalation).
System Events
audit.exported—An audit export (CSV or PDF) was generated.
audit.verified—The hash chain integrity verification was executed.
retention.purged—Entries were deleted per the configured retention policy.
Prove every decision your AI agents make.
The Audit Trail is not a feature. It is the compliance foundation. Start logging immutably in under 10 minutes.