BEV 200K Decision Mix

200,000 English-language decision rows drawn from existing datasets, public records, text-based game environments, and procedural generators. Each row presents source-grounded input state and one or more bounded decisions inspired by the JEV CHOICE, NOUL, and SCORE contracts. The mixture aims to train models that select an option, judge a proposition, or assign an ordered score without generating a long free-form answer.

There is no validation split. The first 150,000 rows (125,614 train / 24,386 test) are the original release, unchanged and in the same order; the hard 50K batch (42,000 train / 8,000 test) is appended after them. When packaging, 169 rows of the original release were reassigned: 166 Upworthy rows moved from test to train and three rows moved from train to test, to keep repeated input states and related source groups on one side. The release contains 185,637 distinct normalized input states; repeated states within a split can carry different decision questions. No state appears in both splits.

Decision formats

Each decision row has a state string and a questions_json string containing a questions object keyed by meaningful decision names. Question types are:

The number of questions exceeds the number of rows because a state can have several decision targets. For CHOICE, criteria maps stable answer keys to option descriptions and label is one key. For NOUL, label is a boolean. For SCORE, criteria is an ordered list of anchored levels and label is its zero-based index. A SCORE label is an ordinal target; it is not a calibrated probability or an unconstrained numeric prediction.

Domains and skills

The table uses one primary domain per row. It describes the actual released mixture, not the sizes of the upstream datasets. Each link leads to an original source or source collection.

Construction and provenance

Five independently built batches contribute 50,000; 19,997; 50,000; 30,003; and 50,000 rows. Most labels come from upstream annotations, recorded outcomes, test transitions, simulator replays, executed programs, or declared deterministic policies. In one 20,000-row slice of the original release, GPT-6 Luna rewrote bounded decision questions and selected source quotes from existing records; source-backed labels were kept. Other smaller Luna steps named existing response options or produced date-checkable deadline questions.

The fifth batch (hard 50K) targets multi-hop lookup, confusable routing, temporal and numeric reasoning, trap cases, unsettled evidence, and answer adequacy. About 60% of its rows are purely procedural, and none of its labels come from a model. In 9,576 of its rows, GPT-6 Luna wrote text that the row contains:

• policy prose (2,077 rows), checked by a second call to have the same rules as a code-generated template;
• a correct or planted-flaw answer (3,000 rows), where a blind judge must recover both verdicts;
• a rewritten or annotated request (1,999 rows);
• handler catalogs and requests (2,500 rows), where a blind router must pick the intended handler.

Outputs that failed a check were dropped. Luna also wrote the one-line option definitions for routing intents.

Some upstream corpora are themselves synthetic, template-generated, or model-generated. This is not a collection of exclusively human-written prompts.

The published rows include a primary domain category. The domain table above links the upstream sources. Row-level source identifiers, raw source downloads, and generation scripts are not included in this release.

Loading the data

data/train.parquet and data/test.parquet are the Hub-compatible viewer files. Their five columns are state, questions_json, domain, question_types, and question_count. The JSON column contains the flexible JEV-like question object. Read the decision questions with:

import json
from datasets import load_dataset

ds = load_dataset("avbiswas/bev-decision-150K")
example = ds["train"][0]
state = example["state"]
questions = json.loads(example["questions_json"])

The release contains only the two Parquet data files. Parquet is used because the heterogeneous nested question objects cannot be reliably inferred as one Arrow JSON schema.

Intended use and limits

This mixture is for training and studying bounded decision models. Its test split is a held-out portion of the mixture, not an independent benchmark suite. It includes existing benchmark families, so evaluating on those same public benchmarks requires checking record and source-family overlap first. Some choice alternatives are sampled or constructed, and answer frequencies were sometimes rebalanced; neither the test split nor SCORE outputs imply real-world calibration.

Several labels have narrower meanings than their names may suggest: a bank subscription is an observed outcome, not a causal recommendation to contact that customer; CVSS base severity is not asset-specific risk; game scores describe immediate transitions, not long-run expected return; and Taskmaster service choices are derived from annotated dialogues, not logs of executed tools. Some source judgments and LLM-rewritten questions may be noisy. English filtering was automated and should not be read as a guarantee that every excerpt is perfect English.

Hard-50K caveats:

• Deletion twins (MuSiQue unanswerable, 2Wiki and ContractNLI deleted evidence) are checked only for the absence of the deciding value, not for every possible implication.
• PRM800K step ratings and CondaQA answers are human judgments.
• The rulebook, temporal, and access-log generators use fixed templates, so their wording varies less than natural text.
• Luna checks are model judgments. They filter noise but may favor cases that model resolves easily.
• A MuSiQue sub-question can appear in both splits.

Attribution and redistribution

No single blanket license is asserted for this mixed-source release. The Hub metadata uses license: unknown for that reason. Consult the linked upstream sources, their papers, and their current terms before public redistribution or commercial use, and attribute every source used. In particular, the ProofWriter mirror leaves upstream redistribution rights to the user to confirm; NVD descriptions may be authored by third-party CVE reporters; and the GoEmotions release includes Reddit text. The hard 50K batch adds MuSiQue, TAT-QA, ContractNLI, MASSIVE and MBPP (CC BY 4.0), CLINC150 (CC BY 3.0), 2WikiMultihopQA and CondaQA (Apache-2.0), and FinQA and PRM800K (MIT). No share-alike or non-commercial source was added. The generated question wording does not erase those upstream obligations.

Please cite this dataset's Hub page and the relevant upstream datasets or papers linked in the domain table above when using a subset.