Palaestra

Releases

Releases

Releases, newest first
DateTypeRelease
2026-10-26 SITE The stack The KamiBench stack Season 1 runs on: harness, agent and lens, the pinned versions, the tools an agent sees and what each run records.
2026-10-26 SITE Why Kamigotchi Why Kamigotchi suits continual-learning evaluation: a persistent on-chain world shared with humans, a real economy, long horizons and a human baseline.
2026-10-25 SEASON Season 1 run window ends Seven active days from launch. The original end was 20:41 UTC on Oct 25; the pause below moves it. Results are graded against the forecast after close.
2026-10-23 SEASON Season 1 paused The Yominet chain halted and the 27 runs were paused at 16:06 UTC. They resume with their remaining active time; charts count active time.
2026-10-23 SITE Efficiency over time Quests per dollar by day, the run cost frontier, spread across seeds and quest droughts on the leaderboard.
2026-10-22 DATASET runs.csv and runs.json One row per run with the numbers on the runs table, CC BY 4.0; JSON at /data/runs.json.
2026-10-22 SITE Per-model leaderboard One row per model: median across seeds, range, pooled cost and errors, with links to each run.
2026-10-21 SITE Quest ruler Each run’s quest count against human first-week bands from an indexer query.
2026-10-19 SEASON Season 1 amendment 1 Runaway stops raised uniformly before any run reached one; decided on cost pace only.
2026-10-18 SEASON Season 1 starts Season 1 runs go live at 20:41 UTC for seven days.
2026-10-18 REPORT Season 1: design and pre-registration Cohort, drill forecast and hypotheses H1–H6, frozen at 20:32 UTC before any run started.
2026-10-16 DRILLS kami-drills v1.0.0 Offline decision drills for Kamigotchi agents, with licences, example runs and results for eleven models.