{"schema":"bay-run.performance-contract.v1","service":"bay-run","proof_schema":"bay-run.performance-proof.v1","harness":{"version":"1","entrypoint":"performance_harness.py","execution":"offline_local_process_only","remote_targets_supported":false,"model_artifacts":"existing local Hugging Face cache required; network disabled"},"live_probe":{"method":"POST","url":"https://run.huggingbay.xyz/v1/speed-test","auth":{"token_endpoint":"https://run.huggingbay.xyz/oauth/token","grant_type":"client_credentials"},"scope":"one bounded three-stage embedding+rerank+classification run; use the harness for distributions"},"workload":{"request":{"model":"Qwen/Qwen2.5-0.5B-Instruct","messages":[{"role":"user","content":"Reply with exactly: Bay Run ready."}],"max_tokens":8,"temperature":0.0},"workload_sha256":"sha256:ca3c56ab4cb82dc3f865090e5b68cabdf14b32805e76b9dd02d856c26c177158","determinism":"explicit model, fixed prompt, temperature zero, bounded output"},"workloads":[{"id":"free.chat","access":"free_bearer","method":"POST","path":"/v1/chat/completions","billing_mode":"none","model":"Qwen/Qwen2.5-0.5B-Instruct","workload_sha256":"sha256:ca3c56ab4cb82dc3f865090e5b68cabdf14b32805e76b9dd02d856c26c177158"},{"id":"paid.x402.chat","access":"paid_x402","method":"POST","path":"/x402/chat/completions","billing_mode":"local_fake_no_external_io","model":"Qwen/Qwen2.5-0.5B-Instruct","workload_sha256":"sha256:ca3c56ab4cb82dc3f865090e5b68cabdf14b32805e76b9dd02d856c26c177158"},{"id":"paid.stripe_credit.chat","access":"paid_stripe_credit","method":"POST","path":"/v1/credits/chat/completions","billing_mode":"local_fake_no_external_io","model":"Qwen/Qwen2.5-0.5B-Instruct","workload_sha256":"sha256:ca3c56ab4cb82dc3f865090e5b68cabdf14b32805e76b9dd02d856c26c177158"}],"measurement":{"clock":"time.perf_counter_ns","latency_unit":"milliseconds","latency_scope":"in_process_asgi; includes model execution and local wrapper overhead","excludes":["bearer auth dependency","rate limiting","model artifact acquisition"],"throughput_unit":"successful_requests_per_second","percentiles":{"reported":["p50","p95","p99"],"method":"nearest-rank"},"cold":{"definition":"model absent from process residency before every measured request","artifact_cache_preserved":true,"concurrency":1},"warm":{"definition":"same process after at least one successful unmeasured prime"}},"runtime_latency_classes":{"warm":{"definition":"the selected model was resident before origin computation began","telemetry":["x_bay_run_meta.latency_class","x_bay_run_meta.residency_class","x_bay_run_meta.cold_load"],"guarantee":"status only; prefer_warm is a preference, not a residency promise"},"cold":{"definition":"the selected model was not resident before origin computation; construction may be owned by this request","telemetry":["x_bay_run_meta.latency_class","x_bay_run_meta.residency_class","x_bay_run_meta.cold_load"],"guarantee":"no public numeric SLA"},"generative":{"definition":"legacy workload-lane label for generative origin computation","telemetry":["x_bay_run_meta.latency_class","x_bay_run_meta.workload_kind","x_bay_run_meta.residency_class","x_bay_run_meta.cold_load"],"residency_class":["warm","cold"],"guarantee":"no interactive SLA for arbitrary cold models; guarded interactive calls return a typed recovery response"},"native":{"model_load":"none","examples":["calculate","validate_json","parse_pdf","parse_document"],"sla":"no public numeric SLA"},"pinned_specialist":{"model_load":"resident only when readiness/cache status confirms the configured warm model","examples":["embeddings","rerank","classification"],"sla":"no public numeric SLA"},"on_demand":{"model_load":"artifact/model load may dominate the first request","examples":["tail specialists","generative chat","schema extraction"],"interactive_sla":false,"telemetry":["x_bay_run_meta.cold_load","x_bay_run_meta.residency_class","x_bay_run_meta.latency_ms"]},"contract":{"schema":"bay-run.latency.v1","metadata_field":"x_bay_run_meta","latency_class_field":"latency_class","workload_kind_field":"workload_kind","residency_class_field":"residency_class","residency_header":"x-bay-run-residency-class","cold_load_field":"cold_load","cold_deadline_error_code":"cold_model_deadline_unavailable"},"guidance":"measure both cold and warm phases; do not infer one from the other or treat prefer_warm as a promise"},"task_latency":{"schema":"bay-run.task-latency.v1","evaluation_evidence_schema":"bay-run.task-evaluation-evidence.v1","quote":{"measurement_scope":"supplied_examples_and_origin_quote_computation","latency_basis":"evaluation_batch_total_estimate","evaluation_batch_total_field":"evaluation_batch_ms","evaluation_batch_average_field":"evaluation_batch_average_ms","hard_latency":{"batch_average_proves_constraint":false,"batch_total_is_admission_estimate":true,"origin_execution_is_authoritative":true},"cache":{"scope":"process_local_and_principal_scoped","ttl_bounded":true,"source_and_age_required":true}},"execution":{"measurement_scope":"quoted_route_origin_model_computation","actual_latency_is_hard_constraint_input":true,"replays_must_be_labeled":true,"cold_load_must_not_be_inferred":true},"warming":{"startup_warm_models":"resident_and_pinned_only","speculative_warm":"bounded_best_effort_after_measured_evidence","unproved_discovery_candidates_are_not_warmed":true},"honesty":"Cached evidence preserves its original measured values and exposes source, age, scope, and deferred candidates. Evaluation batch totals and averages are estimates, not a quality or latency SLA; the origin execution measurement is authoritative."},"paid_measurement_scope":{"mode":"local_fake_no_external_io","includes":["request validation","payment wrapper","model execution"],"excludes":["facilitator network","Cloud SQL","Stripe network","client network","TLS"],"warning":"Do not interpret local fake measurements as live settlement latency."},"compare_only_when":["proof_schema matches","workload_sha256 matches","latency_scope matches","billing_mode matches","requested and observed models match","phase and concurrency match","failed equals zero"]}