Kokos

13:31:28 Lock acquired for command: linksite:jp-bench-watch

13:31:28 [2026-07-10T13:31:28+00:00] JP bench 10/51 scored (19.6%) · missing 41
orchestrator: not detected
next cell: jp-2 × moonshotai/kimi-k2.6
log: ❌ Cell failed: jp-2 × moonshotai/kimi-k2.6 — generation failed
log: 🔄 [2/41] jp-2 × fugu-ultra → grid-fugu-ultra-jp-2-20260710-123010
log: ❌ Preflight blocked: jp-2 × fugu-ultra — fugu-ultra: pool account(s) configured but all benched/cooldown — fix routing before spending tokens
log: 🔄 [3/41] jp-3 × moonshotai/kimi-k2.6 → grid-moonshotai-kimi-k2.6-jp-3-20260710-123010
gaps (48):
· [info] jp-1: A/B comparison ready (6 challengers + baseline)
· [medium] jp-2: missing 2 challenger(s): moonshotai/kimi-k2.6, fugu-ultra
· [info] jp-2: A/B comparison ready (3 challengers + baseline)
· [medium] jp-3: missing 4 challenger(s): moonshotai/kimi-k2.6, fugu-ultra, anthropic/claude-sonnet-5…
· [high] jp-4: missing 5 challenger(s): deepseek/deepseek-v3.2, moonshotai/kimi-k2.6, fugu-ultra…
· [high] jp-5: missing 5 challenger(s): deepseek/deepseek-v3.2, moonshotai/kimi-k2.6, fugu-ultra…
· … +42 more (see gaps report file)
improvements (7):
· [high] 112 linksite eval/finalize failed_jobs in 24h — inspect failed_jobs table
· [medium] aisubscription: avg 28.303070094188min/cell — consider timeout headroom or faster judge (bench already uses fixed GLM judge)
· [medium] moonshotai/kimi-k2.6: avg 91.828335563342min/cell — consider timeout headroom or faster judge (bench already uses fixed GLM judge)
· [medium] fugu-ultra: avg 68.812861482302min/cell — consider timeout headroom or faster judge (bench already uses fixed GLM judge)
· [info] finalize_generation.skip_auto_eval_bench ×4 in recent logs — Bench double-eval prevented (expected on is_bench_run)

13:31:28 Lock released for linksite:jp-bench-watch

Bug Report Submitted!