These mappings are draft legal-interpretation artifacts, pending
compliance-counsel review — not legal advice, and they cover gateway-routed
traffic only. Wardin produces Art-12-grade runtime records; it does not make
you “EU AI Act compliant,” and no software can. Every pack is shown with a
DRAFT
status, its disclaimer verbatim, and its content hash.Packs
A pack maps Wardin’s receipt checks to a framework’s controls. Six ship today, each versioned and effective-dated:
Each pack carries a version, an effective date, a verbatim disclaimer, and a
pack hash (a content hash of the pack’s raw bytes). The gateway stamps the pack
version onto each receipt as it’s written, so a receipt is always read against the exact
mapping that was in force when it was produced — even after a pack is later revised.
The control mappings above are drafts, pending compliance-counsel review. Which
articles, functions, and controls each pack cites — and whether a check “proves” a
given control — is Wardin’s current engineering interpretation, not a
counsel-reviewed crosswalk and not legal advice. Mappings will change as review
completes; the pack
version + packHash let you pin exactly which interpretation
produced any claim. Treat them as a starting point for your own compliance team, not a
substitute for one.Coverage levels
Every control the pack cites gets one of three coverage postures:ENFORCED— a signed in-path check proves the control on every request (e.g. EU AI Act Art 12, satisfied by the signed, chained event log itself).PARTIAL— gateway checks cover part of the control; the rest is your process. The control row spells out the gap under “Your responsibility.”EXTERNAL— no gateway check maps to the control. Its evidence lives outside the request path; Wardin can reference it, not prove it.
n enforced · n partial · n external), so you can see a framework’s request-path coverage at a glance.
Proven by — the checks behind a control
AnENFORCED or PARTIAL control lists the in-path checks whose signed pass/fail results
are its evidence:
AUTH— virtual-key authenticationBUDGET— per-key spend-limit enforcementALLOWLIST— model allowlistGUARDRAIL— prompt-injection / jailbreak detectionPII— PII redactionUPSTREAM— the provider call’s outcome
checks on the receipt, so a control’s claim is
traceable straight down to individual signed records.
Querying evidence by control
Coverage tells you what a control means; the per-control drill-down tells you which requests it applied to. For any in-path control you can pull a window (24h / 7d / 30d) of the receipts that applied it, each with its outcome (enforced / failed /
indeterminate) and pack version, and open any one to verify it.
Where to see it
The EVIDENCE rail stage in the app renders the full coverage map: pick a framework, read each control’s posture and evidence, and drill into the signed receipts behind it. It sits next to the chain statistics and a sample verified receipt, because framework coverage is a lens on the same receipt chain — read-only, never a place you configure anything.The one-click auditor Evidence Bundle export and the offline verifier (spec
published; reference CLI, public release in preparation) are available today. In
development: long-term WORM retention beyond the hot analytics window, so a
bundle can span years rather than the recent window. Until it ships, treat framework
coverage as a lens over the recent, verifiable receipt window — see Signed
Receipts for exactly what’s verifiable today.