Context

During v0.0.7 benchmarking, Kimi K2.5 took ~48 minutes per target while Claude Opus 4 took ~4 minutes. We had no way to know this because the script doesn't log start/completion times.

What's needed

• Log start time and elapsed time for each target in scripts/run_full_benchmark.py
• Add per-target timing to the summary table at the end
• Optionally: write a timing.json alongside the results for programmatic analysis

Flagged during v0.0.7 runs but deferred.