[experiment] Validity Set Labelled: 76-150 Vaisakhi
Labelled Sheet - Vaisakhi 76 to150
🤖 Auto-triage: tagged as an experiment (confidence 0.95).
Reason: The PR uploads a CSV file containing evaluation labels and model responses for a judge reliability run, which constitutes reporting results from an experiment.
This comment and the [experiment] title prefix were added automatically by a local model (qwen3.5:4b). If this isn't an experiment, remove the prefix or reply here.
🤖 Task triage: the closest task to this PR is Add a multiturn question (add-multiturn-question), confidence 0.35. The PR adds a CSV file containing labels for an evaluation run (eval/judge_reliability/validity_v1/labels/sheet_vaisakhi.csv). While the content includes multi-turn examples (e.g., u002), the task is explicitly about building evaluation infrastructure to test judge reliability, not adding new questions to the corpus. The file path indicates this is part of an experiment or validation run rather than a contribution to the question bank itself.
This is recorded in the platform's backend, not in the title.
This comment was added automatically by a local model (qwen3.5:4b). If it got this wrong, reply here and a maintainer will fix the record.
🤖 Posted with Claude Code.
Thanks @vaisakhi ! I checked the sheet: 74 of your 75 units are labelled, all valid. u094 has no label. Could you add it? If not, it still has two labels from the others, so it isn't lost.
A small note: the file was saved in Mac Roman rather than UTF-8 (probably by Excel or Numbers). The scoring script now reads that fine (#323), so there's no need to re-export.