Files
SkillFactory/performance_v042/final-source-evidence-review.md
T
chenlin c528dd268f feat: checkpoint Go2 FL simulation performance v0.4.2
Include callable skill and live-GUI prerequisites, epoch sensor reads, packed snapshots, sampled bit-exact kinematics, and verified-tick rendering without changing the physical/control schedule.

Record bounded analytic S1 GUI/headless qualification, original failures and CPU evidence (1557 passed, 6 skipped). Preserve accepted_skill=false and sub-real-time limitations. Exclude native authorities, caches, large media and local delivery copies; retain sealed artifact digest.
2026-10-16 15:46:01 +08:00

2.9 KiB
Raw Blame History

Review

  • Correct — new P1 fix: live_cadence_v042.py:18–46 retains the concrete Kit identity, checks membership/liveness/pause state before and after added rendering, and increments the paint count only after validation. This closes the SDK-swallowed visualizer-removal gap identified in performance_v042/display-code-review.md.
  • Correct — integration: scripts/evaluate_cycle_functional_flow.py:175–179,226–231,291–295,328–331 applies that guard before physics, after controls—including final interval20 rendering—and during/post final viewing. live_cadence_v042.py:63–76 preserves evidence-before-paint ordering and headless no-painter behavior.
  • Correct — regression evidence: tests/test_contact_cycle_live_cadence_v042.py:95–137 covers swallowed removal, original interval20 failure, unhealthy/replaced identities and headless operation. Retained logs and XML agree: targeted 77 passed; full suite 1557 passed, 6 skipped, zero failures/errors. Nothing was rerun.
  • Correct — bounded native evidence: Both review extracts report original campaign/worker success, six cycles and 27000 ticks. Original observer receipts show actual caller wait0; GUI additionally records real WM_DELETE, unchanged hold counters and window absence. Both physical audits report zero new steps, passing functional checks and exact world targets (actual-audit.json:1010–1026).
  • Correct — performance and equality claims: performance-result.json supports GUI 764.863s / 0.07060×, headless 294.679s / 0.18325×, and 25650 added + 1350 original renders. Both comparison JSONs report 70 equal tensors and identical whole-trace SHA256 d0934245e50f3d7e174b24c109f167236c4693a023d1da01bbadfd71e41b4dba. The inspected PNG visibly shows 33.98 FPS, tick2519. Documentation correctly separates render submissions, capture sampling and simulation throughput (docs/contact_cycle_sim_performance_v042_result.md:15–21).
  • Correct — disclosure: The independent headless postexit first failure remains preserved with unknown cause; the second clean observation is not presented as an explanation or physical rerun. Failed baseline01, initial CPU collection failure, historical 30mm RootXY failure, qualification limits and mandatory accepted_skill=false remain explicit (docs/contact_cycle_sim_performance_v042_result.md:43–55).
  • Fixed: None by this read-only reviewer.

No issues found.

  • Merge verdict: OK with notes — no blocker found for proceeding to packaging of the bounded analytic-S1 result. This is not delivery-complete approval.

Residual risks

New-cadence native early-abort, injected visualizer failure, startup-close, safe-stop and 600s timeout remain untested. S2/S3/nominal20/u256, all-input bitwise equivalence and global native-state equality are not qualified. Package checksums, final inventory/protected/process checks and sealing remain pending.