asingh15/data-curation-sample-browser-qwen35-n8

Filtered Data Curation Candidate Browser

Public research browser for ten reviewed English-language train targets from the filtered Qwen3.5 9B data-curation branch cache. It covers all ten capability axes and contains all eight curator candidates for each target.

Trajectories are read from the original candidates.jsonl journal rather than the 8,000-character materialized evidence field. Assistant reasoning, response content, tool calls, tool results, final submissions, and ID lists are exported without additional visualizer truncation.

The source cache contains 4,991 train and 30 held-out validation targets after 75 examples above the 196,575-token context limit were excluded. This browser publishes reviewed train examples only; held-out validation rows are not included.

Gold is the unclipped privileged normalization used by the branch trainer:

(candidate_nll_delta - random_delta)
-----------------------------------------------------------
(privileged_oracle_delta - random_delta)

Values may be negative because a candidate can improve NLL less than the random baseline. Candidate trajectories contain model-generated content and selected examples from the public base dataset.