version | Disclosure schema version | v0.8, or v0.8+tier2-v0.2 when the Tier-2 block below is present |
plaintext_to_provider | Whether the serving Mac sees your prompt in plaintext | true |
model_identity | Where the model identity comes from | provider_reported |
hardware_attestation | How much of the pool is hardware-attested for this purpose | none, partial, all or unsupported |
model_hash_verified | How much of the pool reports a model hash that matches the signed catalog | all, partial or none |
provider_leg_encryption | Encryption of the coordinator-to-provider leg beyond TLS | Reported from live metadata |
untrusted_provider_safety | Behavioral safety checks on providers | Reported from live metadata |
sticky_affinity | Whether requests with the same conversation tag are preferentially routed back to one provider, and for how long | enabled: true, ttl_seconds: 1800 |
prefix_cache | Automatic KV-cache reuse on the serving Mac, with a plain-language description. It does not pin routing | enabled: true |
relay_blind_request_encryption | Request encryption hidden from the relay. Present only when enabled | Omitted (default-off pilot) |
model_verification_limit | Plain-language limit of model verification | Text |
verified_model_settlement | Which paid entrypoints settle on receipts, the outcome meanings, and the claim limits | Object |
compute_integrity | Status of sampled compute-integrity checks | unavailable, with no settlement effect |
tier2 | Per-pillar detail: model_hash (with verified and uncatalogued provider counts), encrypted_leg, attestation, behavioral_safety | Present when any Tier-2 check is active |