littlechest/boxoffice-verified-seeds

BoxOffice Verified Seeds

This dataset contains the released BoxOffice seed datasets used in the benchmark pipeline described in the accompanying paper. The release includes ten verified seeds:

• —7
• —11
• —13
• —17
• —19
• —23
• —29
• —31
• —47
• —73

For each seed, we provide:

• —a full JSONL file containing warmup rows plus evaluation rows
• —an eval JSONL file containing only the evaluation rows
• —a manifest JSON file
• —a validation JSON file with directional warmup counts

Layout

• —seeds/
• —boxoffice_filtered_v4_balanced_m4_s<seed>_full.jsonl
• —boxoffice_filtered_v4_balanced_m4_s<seed>_eval.jsonl
• —manifests/
• —boxoffice_filtered_v4_balanced_m4_s<seed>.manifest.json
• —validation/
• —boxoffice_filtered_v4_balanced_m4_s<seed>_direction_counts.json
• —croissant.json
• —Croissant metadata with the minimal Responsible AI fields required for NeurIPS dataset submission

What These Files Represent

Each JSONL row is a fully rendered benchmark example built from a synthetic film corpus. The released seeds were generated by:

1. 1.constructing or reusing warm-retrieval metadata,
2. 2.oversampling candidate rows for the four directional cells,
3. 3.keeping only rows that the filter models solve with full context but fail in the question-only setting, and
4. 4.augmenting the resulting evaluation set with balanced directional warmup rows.

The released seed ids above are the verified outputs of that pipeline.

Intended Use

This dataset is intended for:

• —reproducing the released BoxOffice seed datasets,
• —evaluating cache-aware and retrieval-augmented language model methods, and
• —auditing benchmark behavior across multiple verified seeds.

This dataset is not intended to support broad real-world claims outside the benchmark setting.

Notes

• —The companion code release will be linked here once public.
• —The hosted croissant.json is the file intended for NeurIPS submission after final metadata checks.