clm-0088
measured-herehigh ●●●
citable URL: https://halobench.com/records/clm-0088/ — this address never moves; the anchor /records/#clm-0088 keeps resolving
HG-002 Phase C's corrected matched plain control completed the five-task tau2 airline smoke at 4/5, mean 0.80, with six valid nonempty tool calls, zero empty turns, zero infrastructure errors and zero max-step cuts. MTP n1 scored 1.0 on each of tasks 0-3 with 15 valid calls, but task 4 returned an empty AssistantMessage and tau2 closed it as an infrastructure error; the arm is invalid and has no aggregate mean. Therefore the Phase A 1.1569x bounded throughput lift does not earn full tau2 promotion. Plain remains the recommendation and HG-002 closes without promoting n1.
verified 2026-08-19 · volatility medium
Note — the record's own working
The first plain root independently produced the same 0.80 scientific result, but its parser false-negative closed rc 1 and is not represented as a formal pass. Exact artifact roots and hashes are preserved on the run records.
Cited by — computed at build time, never stored
candidate gate history nemotron35-lightning-30b