Add GLM 5.3 Flash scored artifacts for 97 CPU tasks
#1
by HJH2CMD - opened
No description provided.
Updated this PR with the 2026-10-16 glm-5.3-flash rerun (reasoning_effort=max, simulator/judge gpt-6-luna).
Added previously excluded tasks:
- swepro/swepro-r0021-a2ce046925-student-needing-explanation: invalid, persona_requirement_checks_missing
- tb3/freecad-spring-clip: infrastructure_error, api_interrupted
Replaced the four other SWE-Pro student-needing-explanation trials (r0064, r0240, r0407, r0638). Each ended invalid because the agent returned empty visible text before persona checks were written.
manifest.json now lists 99 tasks: 13 pass, 80 fail, 5 invalid, 1 infrastructure_error. tb3/exam-pdf-eval remains excluded (GPU).