Done — merged to main (core b2bd86b, /health exposure a60c9f1).
harness::admission::AdmissionController replaces the per-model unbounded, untimed lock-wait FIFO with a bounded batch-1…
Done — merged to main (cdf8728). The A0 seatbelt.
Before dispatch, cortex reserves prompt_estimate + max_output (max_output from max_completion_tokens/max_tokens, else the…
Done — merged to main (4f16b8c).
cortex-gateway::metering: real per-request usage is captured at completion and fed to the spend ledger + per-principal Prometheus counters (`cortex_spend_to…
Done — merged to main (486d7e9).
cortex-gateway::auth::require_principal axum middleware: parses Authorization: Bearer <key> (Bearer only — no custom headers/body fields), resolves it…
Done — merged to main (bc74e0e).
cortex_core::entitlements: EntitlementProvider trait (resolve / reserve / settle / release / snapshot) with Principal, CapWindow (Balance