benchmarks/README.md

LoT H3 benchmarks

• Experiment history: runs, results, limits, and what was tested
• Current r4 A/B summary
• OpenVid step-0 layout comparison
• Runtime microbenchmarks
• Step-0/250/500 OpenVid pilot CSVs
• r4 control and preserve validation CSVs
• r4 untouched confirmation CSVs and preserve

Pilot CSVs

• Step 0 baseline and aggregate JSON
• Step 250
• Step 500

r4 validation CSVs

• Control step 250 · control step 500
• Preserve step 250 · preserve step 500

The r4 run ended without a promoted winner. Metric definitions, sample sizes, and limits are documented in the experiment history.