Acceleration Measurement

Acceleration Measurement

Compare a baseline and candidate under the same observation protocol, constraint frame, receiver family, resource envelope, fixed horizon, stopping rule, and evidence policy.

Higher is better for verification yield, receiver reuse, and certified capital gain. Lower is better for time-to-verified, residual half-life, resource cost, and absolute error correlation.

Use an accelerator claim only when accepted=true and acceleration_metrics_certified=true. This is a finite protocol result, not proof of real ASI.

Run The Included Comparison

From a source checkout:

pic runtime compare --baseline examples/runtime_acceleration/baseline.json --candidate examples/runtime_acceleration/candidate.json
pic runtime certify-acceleration --baseline examples/runtime_acceleration/baseline.json --candidate examples/runtime_acceleration/candidate.json

The example folder also contains rejection fixtures for resource mismatch, missing evidence, post-selected horizons, and metric regression. Input-side accepted fields are ignored. Every metric must be finite and evidence-backed; there must be no worsening beyond its declared tolerance and at least one improvement beyond tolerance.