The questions a skeptical reviewer, engineer, or auditor would actually ask — answered directly, with links to artifacts they can verify themselves.
What this site actually is, and what it deliberately is not.
No. This site is the public interface, documentation hub, and artifact surface for Project-AI. The runtime architecture is documented separately; this portal is not evidence that every described enforcement path is deployed. Surfaces should distinguish documented claims, generated artifacts, simulations, and demos.
Use the provenance badge on the specific surface: SIMULATION, DEMO, DOCUMENTED, or VERIFIED. For a strict receipt-v1 verification bundle, /verify recomputes the canonical hash and checks the declared Ed25519 signature locally. That proves only the supplied bundle against the portal's declared key; it does not prove ledger membership, signer custody, freshness, or the truth of the receipt's decision.
No. Project-AI documents an execution-governance architecture. A chat surface may be one client of that design; the contract targets deny-by-default capability adjudication before governed effects, while complete runtime-path enforcement remains unverified.
Pre-execution control, not post-hoc dashboards.
The documented governance contract requires capability requests to pass the Triumvirate and STATE_REGISTER before execution, producing ALLOW, DENY, or SAFE_HALT. End-to-end conformance across every execution entry point is an assurance target, not a fact established by this portal.
The documented deny-by-default rule treats absence of an explicit ALLOW with a valid capability token as DENY. Non-bypassability across all runtime paths remains subject to executable evidence. See the Glossary entry for AC0–AC5 authority classes.
The contract defines SAFE_HALT as a terminal outcome alongside ALLOW and DENY when safe adjudication or continuity fails. Receipt production is a required invariant; the local verifier can check a supplied strict bundle but does not prove that every runtime SAFE_HALT produced a valid signature.
RLHF and Constitutional AI primarily shape model behavior; policy engines and guardrail libraries can also participate in runtime enforcement depending on their integration. Project-AI documents a separate pre-execution governance design, but this comparison does not establish exclusivity or benchmark superiority.
What an external reviewer can inspect now, and which proof capabilities remain withheld.
Paste a strict receipt-v1 verification bundle at /verify. The browser validates the exact schema and timestamp syntax, recomputes the canonical SHA-256 receipt hash, resolves the declared portal key, and checks the Ed25519 signature locally. It does not resolve opaque ledger references, prove ledger inclusion, verify external timestamps or freshness, establish private-key custody, or prove that the recorded decision was semantically correct.
/keys renders the Ed25519 audit public key declared by the current portal build. Deployment-private-key pairing, warrant and PGP keys, rotation history, release checksums, and RFC 3161 timestamp-authority material are not established there.
The Claims Registry at /claims publishes bounded PAI claims with separate assurance status and implementation maturity, plus pinned evidence, limitations, and open promotion criteria. The current promotion freeze prevents unsupported upward movement.
Use /disclosure for the coordinated vulnerability policy (scope, SLA, safe-harbor). Machine-readable contact metadata is published at /.well-known/security.txt per RFC 9116.
Live system posture and what is currently shipped vs planned.
Subsystem posture is published at /status. The portal heartbeat is real; Triumvirate, Redis, NIRL, and kernel signals are marked as live, simulated, or planned per subsystem.
Legion’s ambassador charter and read-only conversation are public at /legion. Legion is an assigned role serving AGI, humanity, and planetary stewardship, not a Genesis event. The separate signed-receipt demonstration requires sign-in. Request execution capabilities through /contact.
A measure of unresolved evidence required to keep an action admissible under the governance frame. The /admissibility-debt page is live as a planned calculator surface and currently documents the metric before public runtime execution is exposed.