PASS
Inspect the accepted demonstration.
ExplorePublic product / Runcap
Control AI coding spend. Require proof before merge.
A local-first control layer with public source, reproducible instructions and inspectable Proof Gate examples.
Real artifact
Open the implementation.
What this proves
A control boundary, not a universal guarantee.
Inspectable
Source, setup instructions, test cases and demonstration pull requests. In one captured OpenAI call, prompt tokens fell from 1,186 to 737 while the changed-line answer remained correct.
Captured call & reproductionLimits
The 37.9% reduction belongs to that one call. It is not a guaranteed saving. Requests must be routed through Runcap for its spend controls to apply. The documented Proof Gate scope is GitHub Actions and Node/npm repositories.
A practical first step
Bring one workflow.
Find the next useful move.
Discuss the bottleneck, the tools you already use, and what a worthwhile result would look like. No system access needed to start.
Review my workflow