This directory contains published benchmark artifacts and, during local development, may also contain intermediate run outputs.
Use these artifacts as the public source of truth:
- compact tool-call benchmark:
zcp_mcp_tool_call_benchmark.mdzcp_mcp_tool_call_benchmark.json
- Excel semantic workflow benchmark:
full_semantic_compare_v5/excel_llm_token_benchmark.mdfull_semantic_compare_v5/excel_llm_token_benchmark.jsonfull_semantic_compare_v5/semantic_benchmark_summary.md
Do not treat checkpoint files, smoke runs, or older comparison directories as official release evidence. Those are for local iteration only and should be removed or archived before a public release.
Use the helper scripts in scripts
or the benchmark entrypoints under examples.