All runbooks
runbookuapiredacted view

UAPI NUC Accelerator Benchmark Evidence Hardening Receipt 2026-06-23

Receipt for fixture-only NUC 140T/NPU benchmark evidence hardening in UAPI.

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

UAPI NUC Accelerator Benchmark Evidence Hardening Receipt 2026-06-23

Receipt for fixture-only NUC 140T/NPU benchmark evidence hardening in UAPI.

Promotion Metadata

  • Source feedback id: 257
  • Source feedback ids: 257
  • Promotion request id: 168
  • Feedback category: consider
  • Feedback source: ai-lane
  • Feedback created at: 2026-06-23T06:07:10.749224+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 257

  • Product: uapi
  • Category: consider
  • Related slug: [REDACTED:high-entropy]
  • Source: ai-lane
  • Created at: 2026-06-23T06:07:10.749224+00:00

UAPI NUC Accelerator Benchmark Evidence Hardening Receipt - 2026-06-23

Authority: fixture-only, no execution, no service activation. Product: uapi Kind: runbook receipt Commit: b5c1fcc Implementation commit: f0c8389

Summary: Codex integrated Grok Wave 7 A guidance into the UAPI benchmark-result fixture validator. The NUC 140T/NPU Gemma lanes now require lane-aware benchmark evidence before any route claim can graduate beyond research.

What changed: - Added lane-aware accelerator metric requirements: NPU requires thermal evidence, 140T GPU requires thermal evidence and package_power_watts. - Added required accelerator evidence fields enforced by validator: openvino_version, driver_stack_redacted, device_class, and redacted_device_evidence. - Added device-class validation for NPU and 140T/GPU lanes. - Added 140T positive benchmark fixture and focused negative fixtures for missing package power, missing driver stack, wrong device class, and missing redacted device evidence. - Updated schema error enum and reporting counters. - Collected Wave 7 Grok A/B/C reports as accepted design evidence. GLM Wave 7 A failed with provider 529 and was not accepted.

Verification: - PYTHONPATH=src python3 -m unittest tests.test_benchmark_result tests.test_fixture_schema tests.test_validate_fixtures_cli tests.test_acceptance_summary_cli tests.test_readiness_report tests.test_continuation_packet: PASS, 28 tests. - bash scripts/uapi-fixture-check.sh: PASS, 216 tests. - git diff --check: PASS. - GitNexus impact validate_benchmark_result_case: LOW, 6 impacted symbols, benchmark tests/evaluation only. - GitNexus analyze after implementation: 2,004 nodes / 3,434 edges / 59 clusters / 73 flows.

Current fixture coverage: - schema packets: 23 - schema cases: 129 - strict positive: 21 - expected fail: 96 - structured expectation: 12 - benchmark-result cases: 14

Safety boundary: This is not a hardware benchmark run. It does not execute models, probe the NUC, call providers, activate routes, or change dispatch. NUC routes remain research-only until operator-reviewed measured evidence is attached and all gates are satisfied.

Agent-loop note: Grok Wave 7 A/B/C completed under grok-build effort=max and were committed. GLM Wave 7 A failed with provider 529 and must be handled through the retry/backoff continuation path, not treated as evidence.