RDL
Decision Evidence Substrate. Open protocol for signed, replayable, hash-anchored decision artifacts.
The Replayable Decision Ledger
RDL is the open protocol that defines the decision receipt: the publishable, citable, verifiable artifact emitted by any AI system that wants to claim conformance with the eight properties of a governed AI decision.
The protocol is open. The reference implementation will be published under MIT-equivalent license. Anyone can implement RDL; anyone can verify an RDL artifact; anyone can submit an artifact to the public receipts registry.
The receipt artifact (public-level fields)
| Field | Purpose |
|---|---|
decision_id | Stable, ULID-keyed identifier for this decision |
parent_decision_id | Hash chain — every decision references its predecessor in the same context |
timestamp_utc | ISO-8601 with monotonic clock anchoring |
system_attestation | Reference to the attested execution environment that produced this decision |
policy_bundle_hash | SHA-256 of the signed policy bundle that authorized this class of decision |
inputs | Provenance-traced inputs: prompts, tool calls, retrieved evidence, prior decision references |
output | The decision itself (what the AI did) |
suppression_history | What was filtered out and why |
review_path | The contestation procedure: who can challenge, on what grounds, with what consequences |
verifier_state | Which controls were running, specified, or skipped under override at decision time |
signature | Ed25519 signature over the canonical-form serialization of the above |
Protocol vs. product
RDL has two distinct surfaces:
| RDL-as-Protocol | RDL-as-Product |
|---|---|
| Open spec, MIT-equivalent | OptimaX commercial offering |
| Anyone can implement | Hosted ledger + verifier service |
| Federation-friendly | Tiered enterprise compliance / regulatory-grade SLA |
Lives at github.com/optimax-solutions/rdl-spec on launch | Lives at optimax.solutions/services/rdl |
This separation is enforced by the brand wall: governedai.ai is not a sales surface. RDL-as-Product information lives on the OptimaX corporate site and is not promoted from this publication.
Conformance to RDL is the G2 / G3 threshold
An AI system whose decisions emit RDL receipts with valid signatures reaches G2 (Signed). A system whose RDL receipts are also independently replayable through the published replay bundle reaches G3 (Replayable). G3 + GDTK-100 pass + external verifier reaches G4 (Conformant). See the conformance ladder.
Status
RDL Protocol v0.1 specification is on the publication queue. The reference verifier and example artifact set will publish concurrently. Federation Layer 9 — multi-operator recognition — is horizon doctrine for 2027–2028.