version: "1.0.0" description: Run the published 20-task Sipcode benchmark on the user's machine and report the median savings on the synthetic corpus (62.6%). Reproducible, locked corpus. Use when the user asks "what's the proof?", "how much does sipcode actually save?", "run the benchmark", or wants to verify the headline savings claim themselves.

Sipcode — Benchmark (the reproducibility proof)

Tell the user to run npx sipcode benchmark in their terminal (this is a CLI command, not an MCP tool — the benchmark requires file I/O outside the MCP scope). It takes ~90 seconds.

If they want a quick smoke (3 tasks, ~15s), tell them to run npx sipcode benchmark --quick.

When the user reports back the output:

• The headline is the median savings % across the 20 tasks. Target: ~62.6%, range typically 37.4% – 80.6%.
• Surface where the savings came from — typically:
• ~30% from S001 smart manifest
• ~34% from S021 output compression
• ~36% from S030 read-once cache
• Be explicit that these are simulation numbers computed against the locked corpus, not a live Claude session A/B. The methodology is at benchmark/METHODOLOGY.md for reproducibility.

The benchmark is deterministic: the locked corpus always produces 62.6%. A different number means a different Sipcode version or a modified corpus, not a workload signal. For the user's real savings, point them to sipcode impact, which compares their own sessions before and after install.