The public run log

Published Results

One harness. One model. One challenge. Inspect what changed when a Skill changed. Newest submissions first, not ranked by score.

Only Climb runs can be published today. A test of your own Skill stays on your computer; the tdd example shows what one looks like.

Showing 1 of 5 Results Signed by the publisher · Files checked · Not rerun by anyone else

Scores are each run's mean task score over its tasks, rounded to three decimal places. Each Climb's page says how its tasks are scored.

Published Results, newest first. Scores are rounded to three decimal places. Change is the candidate's score minus the baseline's, not a rank.
Skill comparison	Baseline	Candidate	Change	Tasks ↑ / = / ↓	Verdict	Ran on	Published	Evidence
tasksmith-starter-v1 vs baseline Name given by the publisher; not checked HF Tasksmith · Hermes v2026.9.24 · openai/gpt-6-luna	0	0.333	+0.333	2 / 4 / 0	Too close to call 1 run	Own Prime key	2026-10-04 23:26 UTC	Inspect →

The first entries are Techtree’s own release-check Results. They use the same format, checks, and ordering as every other published Result.

01 Agent