All runbooks
runbookuapiredacted view

UAPI Thirty Seventh Wedge Benchmark Harness Receipt 2026-06-22

Receipt for the fixture-only UAPI benchmark harness wedge covering NUC NPU, 140T, CPU reservation, and cloud burst guard cases.

Updated 2026-06-23·Freshness: Reference·Sensitive tokens or credentials were hidden before display.

UAPI Thirty Seventh Wedge Benchmark Harness Receipt 2026-06-22

Receipt for the fixture-only UAPI benchmark harness wedge covering NUC NPU, 140T, CPU reservation, and cloud burst guard cases.

Promotion Metadata

  • Source feedback id: 208
  • Source feedback ids: 208
  • Promotion request id: 137
  • Feedback category: beneficial
  • Feedback source: ai-lane
  • Feedback created at: 2026-06-23T01:21:32.604422+00:00
  • Target slug: [REDACTED:high-entropy]
  • Review state: review
  • Reviewer note: AI lane auto-promoted redacted feedback into internal catalog context.

Source Feedback Body

Feedback 208

  • Product: uapi
  • Category: beneficial
  • Related slug: [REDACTED:high-entropy]
  • Source: ai-lane
  • Created at: 2026-06-23T01:21:32.604422+00:00

UAPI Thirty-Seventh Wedge Benchmark Harness Receipt

Date: 2026-06-22 Project: uapi Authority: fixture_only_no_execution Repo: /[REDACTED:high-entropy] Implementation commit: da0c69c Feature commit: 254f966

Summary

Codex added a fixture-only BenchmarkHarnessPacket for the UAPI NUC 140T/NPU and cloud-burst comparison stream derived from Grok long-pass B. This makes accelerator benchmark readiness testable without running hardware, downloading models, calling providers, activating routes, or claiming production readiness.

Files

  • fixtures/benchmark-harness.v0.json
  • src/uapi_service/benchmark_harness.py
  • tests/test_benchmark_harness.py
  • schemas/uapi-first-wedge-contracts-v0.2.schema.json
  • src/uapi_service/fixture_schema.py
  • src/uapi_service/continuation_packet.py
  • [REDACTED:high-entropy].md

Cases

  • [REDACTED:high-entropy]: pass, fixture-only NPU benchmark envelope.
  • [REDACTED:high-entropy]: pass, fixture-only 140T benchmark envelope.
  • [REDACTED:high-entropy]: fail with E_CPU_RESERVED_FOR_CONDUCTOR.
  • cloud_burst_missing_broker_rejected: fail with E_BROKER_REQUIRED.
  • benchmark_execution_request_rejected: fail with E_EXECUTION_REQUESTED.
  • missing_no_fallback_metric_rejected: fail with E_NO_FALLBACK_METRIC_REQUIRED.

Safety

No live benchmark execution was performed. The fixture explicitly keeps model_download_requested=false, hardware_run_requested=false, provider_call_requested=false, route_activation_requested=false, credential_use_requested=false, and redaction_boundary=server-side. Dispatch, backend calls, ledger writes, service listener activation, and Explorer export from service remain blocked.

Verification

Command: bash scripts/uapi-fixture-check.sh

Result: - Fixture validation: PASS - Acceptance summary: PASS - Readiness report: PASS - Continuation packet: PASS - Unit tests: 133 OK - Schema packets: PASS, packets=14 - Schema coverage: cases=44, strict_positive=7, expected_fail=25, structured=12, unknown=0 - Readiness: fixture_ready_no_activation - Checks: 10 passed, 0 failed - GitNexus refreshed: indexed commit da0c69c, 1,046 nodes, 2,087 edges, 41 clusters, 64 flows

GitNexus

build_schema_packet_report context was checked before wiring the new packet target. build_continuation_packet upstream impact was checked before updating the unit test floor; risk was LOW with four direct test callers. No HIGH or CRITICAL GitNexus warning was ignored. The local GitNexus CLI did not expose detect-changes, so verification used GitNexus status/analyze, git diff --check, and the full fixture gate.

Next

Attach real BenchmarkResult evidence only after operator-reviewed hardware runs. Keep nuc_140t_npu_benchmark blocked until measured metrics and no-fallback proof exist. Continue with Grok/GLM splits for placement evidence attachment and identity-gated accelerator routing.