CoolFace
Datasetpublic

anon123312/retrieval-conditional-neurips2026

Dataset Release — Retrieval-Conditional NeurIPS 2026 This bundle accompanies the NeurIPS 2026 D&B Track submission "To Retrieve or Not to Retrieve? Most of the Benefit is Structural, Not Semantic." Contents File Config name Description data/per_task_outcomes.csv per_task_outcomes (default) Per-(backbone × env × condition × task) success/failure labels. 3,064 rows. data/stats_per_cell.csv stats_per_cell 54-cell aggregate success rates and pairwise… See the full description on the dataset page: https://huggingface.co/datasets/anon123312/retrieval-conditional-neurips2026.

sourceHugging Facecc-by-4.0updated 5mo agoView on Hugging Face
0likes28downloads
Dataset Card

Dataset Release — Retrieval-Conditional NeurIPS 2026

This bundle accompanies the NeurIPS 2026 D&B Track submission "To Retrieve or Not to Retrieve? Most of the Benefit is Structural, Not Semantic."

Contents

FileConfig nameDescription
data/per_task_outcomes.csvper_task_outcomes (default)Per-(backbone × env × condition × task) success/failure labels. 3,064 rows.
data/stats_per_cell.csvstats_per_cell54-cell aggregate success rates and pairwise contrasts.
data/feat_env_aggregate.csvfeat_env_aggregatePer-environment task-feature aggregates used in TFA correlations.
data/multibackbone_summary.csvmultibackbone_summaryPer-backbone summary (success rate, mean steps, tokens).
trajectories.tar.gz—Trajectory bundle (3.0 MB compressed, 1,940 files): 390 NoRet evaluation JSON trajectories + 1,387 demonstration-pool trajectories across ALFWorld/ScienceWorld/WebShop + 163 raw server-log files for the FDR/Coupled/Random/NoRet runs. See README.txt inside the archive.
croissant_metadata.json—Croissant 1.0 metadata (validated with mlcroissant 1.1.0 + 7 NeurIPS RAI fields).

The four CSV files have heterogeneous schemas, so each is exposed as a separate dataset config. Load any of them with:

python
from datasets import load_dataset
ds = load_dataset("anon123312/retrieval-conditional-neurips2026", "per_task_outcomes")

Reproducing analysis

The CSVs are sufficient to reproduce every number in the paper body. Trajectory JSONs are provided for qualitative inspection / extending the TFA features.

License

Released under CC-BY-4.0. See Croissant metadata license field.

Verification

bash
sha256sum data/per_task_outcomes.csv  # 78fdbbe70384...
sha256sum trajectories.tar.gz         # 88b013d0e27e...
python -c "import mlcroissant as mlc; mlc.Dataset(jsonld='croissant_metadata.json')"