Andrew Hundt 4950cc7268 feat(benchmark): emit versioned experiment fact tables ...

Write runs.json, steps.jsonl, results.json, artifacts.json, facts.json, and hashed manifests from scripts/benchmark-incremental-speed.py. Record candidate revision/build context from scripts/run-benchmark-experiments.py and keep absent legacy metadata as explicit unknown facts.

Use experiment terminology for new runner, composition, autotune, provenance, and documentation records while retaining legacy campaign keys, flags, paths, and the shim at compatibility boundaries.

Verified by 182 Python tests (1 skipped, 25 subtests), a rebuilt-binary smoke with 18 emitted step rows, and Draft 2020-12 validation of the generated facts bundle against docs/schema/benchmark-facts-v1.schema.json.

Signed-off-by: Andrew Hundt <ATHundt@gmail.com>

2026-07-22 04:54:18 -04:00

..

benchmark-facts-v1.schema.json

feat(benchmark): emit versioned experiment fact tables

2026-07-22 04:54:18 -04:00