feat(benchmarks): reproducible runner for four benchmarks, with the multi-round decider made visible - #425

Open

YueAWu wants to merge 8 commits into

mainfrom

feat/llm-multiround-retrieval

Commits on Aug 27, 2026

feat(search): make the multi-round decider tunable and its fallback visible
juwei.yue
committed
0bc7e3b
fix(embedding): stop retrying permanent provider rejections
juwei.yue
committed
9b5cd29
fix(runtime): bound OME runs, make the pool explicit, allow LLM extras
juwei.yue
committed
0f91c19
refactor(memory): decouple profile extraction from clustering
juwei.yue
committed
cc31dc4
feat(benchmarks): one reproducible runner for all four benchmarks
juwei.yue
committed
d8ee074

Commits on Aug 28, 2026

docs(benchmarks): describe what the decider actually does with no endpoint
juwei.yue
committed
dbc21d3
docs(benchmarks): rewrite the guide, and fold the converter into its adapter
juwei.yue
committed
b7de8bf
feat(benchmarks): ship each benchmark's best configuration, and gate all four
juwei.yue
committed
e50eff5