zeekayandHanzo Dev b9235d3ca0 decision: the frozen Kai/Laya/Jev harness, and the Laya and Jev baselines ...

Questions are the Laya 0.3.20 builders, unchanged (seed 13): ten application
suites of 400, typed decisions (2,000 decisions), and MASSIVE at 51 languages x
100. Every backend is scored by the harness's own metric functions.

Laya: typed decisions on hanzoai/kai-1@b50502c2, whose weights have the same
SHA-256 as convaiinnovations/laya@55cf4c4e. 0.7660 accuracy, reproducing the
published table exactly. Jev: every suite through OpenRouter (typesafe/jev-1.13
served as jev-1.13-20260917), 10,799 of 10,800 answered, $0.25. Laya's
full-suite predictions follow when that run completes.

Co-authored-by: Hanzo Dev <[email protected]>

2026-09-24 19:55:39 -07:00

280 KiB

Raw Permalink History

View Raw