Build provenance, typed rendering, safety-task leaks, answer-first MC fix, privacy-200k - #19

Merged

sileod merged 5 commits into

mainfrom

claude/tender-hamilton-kj7xgv

Sep 25, 2026

Commits on Sep 25, 2026

Build fingerprints and exact source revisions; typed renderer kinds; catalog API fixes
claude
committed
f8f5740
WildGuardMix split by prompt; BeaverTails one row per pair with its majority, plus its vote share as a graded family
claude
committed
bb9241b
load_task: multiple-choice options no longer put the gold answer first
claude
committed
dc04ccb
Add privacy-200k-Mistral-Large-3: privacy sensitivity rated 1-5, as an ordinal task
claude
committed
02267a6
Regenerate task catalogs
claude
committed
d6a7f1a