Skip to content

Seed 1 traces (2026-11-06)

Choose a tag to compare

@samarmstrong samarmstrong released this 09 Nov 18:57

The four run directories of seed 1 (outputs/vl/{a-plain,b-conveyor,c-review,d-tests}, Sonnet 5.5 medium on Vertex, 20 problems, 101 checkpoints per arm): every checkpoint's workspace snapshot, Claude Code trace, reviewer and shadow-reviewer verdicts and hidden-test results, as a tar.zst (32 MB, 411 MB unpacked). Scrubbed and generated with dedicated single-purpose keys; the results table and CSV are in results/seed-1/.