asingh15/data-curation-sample-browser

1---2title: Data Curation Candidate Browser3colorFrom: green4colorTo: yellow5sdk: static6app_file: index.html7pinned: false8---9 10# Data Curation Candidate Browser11 12Temporary public research browser for reviewed rows from the data-curation13rubric-branch pilot cache. It contains five target tasks with all 64 curator14candidates and their corresponding gold scores.15 16Trajectories are read from the original `candidates.jsonl` journal rather than17the 8,000-character materialized evidence field. Assistant reasoning, response18content, tool calls, tool results, final submissions, and ID lists are exported19without additional visualizer truncation.20 21Gold is the unclipped privileged normalization used by the branch trainer:22 23```text24(candidate_nll_delta - random_delta)25-----------------------------------------------------------26(privileged_oracle_delta - random_delta)27```28 29Values may be negative because a candidate can improve NLL less than the random30baseline. This is the 98-train/30-validation pilot cache derived from the31128-target evaluation run, not the unfinished 5,066-target production cache.32