Complete IMU execution comparison

These are already-spent synthetic development records. The whole-API paired median is 1.452x, below the declared 2x target. No independent accuracy claim is made. See the measurement report.

• complete-evidence.tar.gz: every worker input/options/full fitted and native result, log and timing; original toolchain; both complete normalized source capsules; schemas, source regression tests and original packaging producer.
• receipt.json: archive size/digest, member count, index/summary digests and both source commits with producer provenance.
• summary.json and summary.schema.json: the measured target miss and its schema. No observations were discarded or rerun.

The archive's index.json binds every other member by size and SHA-256. The CPU replay command in the report verifies this index, both source manifests, the complete six-worker cohort, provenance bindings, input/options and whole outcome equality, and reconstructs the summary. It never executes archived source. Its exit status is 2 for these valid records because the 2x target was missed. A new run uses a new directory and retains all attempts.