datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
decision-bench
DecisionBench
DecisionBench tests how well AI models read a situation and answer questions about it. Each model gets
the same information, rules, and answer choices. Questions ask for a yes/no answer, a choice from a
list, or a rating.
There are two sets: medium, with explicit rules, and hard, with more judgment calls.
Each has 80 situations and 293 questions. The chart compares how often each model matches the answer
key and how much the API requests cost.
Accuracy is… See the full description on the dataset page: https://huggingface.co/datasets/akhilaaa3/decision-bench.brazilian_court_decisions
Dataset Card for predicting-brazilian-court-decisions
Dataset Summary
The dataset is a collection of 4043 Ementa (summary) court decisions and their metadata from
the Tribunal de Justiça de Alagoas (TJAL, the State Supreme Court of Alagoas (Brazil). The court decisions are labeled
according to 7 categories and whether the decisions were unanimous on the part of the judges or not. The dataset
supports the task of Legal Judgment Prediction.
Supported Tasks and… See the full description on the dataset page: https://huggingface.co/datasets/joelniklaus/brazilian_court_decisions.typed-decisions-synth
Typed Decisions Synth
This is the synthetic dataset I made for Hmm, a small open model that answers questions about your data with probabilities instead of text.
It has 7,414 cases with 25,859 questions across 149 domains and workflows. Every question has an answer and a soft label (a probability for every option), so you can train a model to be unsure when it should be.
Code and the model: github.com/n4ze3m/hmm
Note: Everything here is written and labelled by an LLM. Nobody… See the full description on the dataset page: https://huggingface.co/datasets/n4ze3m/typed-decisions-synth.typed-decisions
typed-decisions
A typed-decision corpus for training a masked-diffusion LM to emit calibrated
discrete decisions instead of text. Built for fine-tuning
Sebasdi/nanodiff-350m-base
(the LLaDA recipe).
The interface
Every example is a prompt plus a response, and every decision is a single
masked token. The answer is always one option letter A-J:
### State:
<unstructured state text>
### Question:
<the decision to make>
### Options:
A) yes
B) no
### Answer:
A
The… See the full description on the dataset page: https://huggingface.co/datasets/pngwn/typed-decisions.typed-decisions-code-holes
typed-decisions: code-holes
Single-token substitutions mined from the git history of 64 public repositories in the
kotoba-lang organisation, each turned into one choice question
with a gold answer: the token the commit actually put there. Part of
kotoba-lang/typed-decisions (README 第8反復),
whose trained model is com-kotobalabs/open-jev-deberta-v3-large.
Sibling dataset (ungoaled): com-kotobalabs/typed-decisions-repo-governance.
What a record is
state = the commit… See the full description on the dataset page: https://huggingface.co/datasets/com-kotobalabs/typed-decisions-code-holes.granite-decisions-synthetic
Granite Decisions synthetic datasets
Original, deterministic English fixtures for Adam Pippert's personal
Granite Decisions project.
The original default config has 162 examples: 54 train, 54 calibration, and 54 test.
These exercise the pipeline; they are not a representative quality benchmark.
Source and license
The source is the project's original template generator, published here as
make_smoke_data.py, from
release v0.1.0,
commit… See the full description on the dataset page: https://huggingface.co/datasets/adampippert/granite-decisions-synthetic.typed-decision-bench
Typed Decision Bench v0.3
Built by Blobfish AI. A benchmark for one-pass decision models: 5,387 items, 25 tasks, 5 use-case
suites. Blobfish designed the tasks, wrote the typed questions, framed each one as a decision a business actually
delegates (use case, vertical), drew stratified seeded panels, froze them, and built the scoring, the contamination
tiers and the quality scorecard. The underlying records are drawn from 21 openly licensed public datasets plus one
generator of… See the full description on the dataset page: https://huggingface.co/datasets/SamuelChien821/typed-decision-bench.a-s-flc-decisions
A-S-FLC Decision Dataset
Training data for fine-tuning LLMs on Asymmetric Signed Force-Loop-Chain reasoning.
What is A-S-FLC?
A decision-making framework where:
Positives are trusted exactly (known benefits)
Negatives are estimated with a conservative buffer proportional to uncertainty
Multiple event chains are scored and the highest stable-net path is chosen
This catches "trap" decisions where uncertain downsides are underestimated.
Dataset Details… See the full description on the dataset page: https://huggingface.co/datasets/denialkhmbot/a-s-flc-decisions.this-that-complex-decisions
this-that-complex-decisions
1,710 decisions where the answer follows from a stated policy applied to a state, and where no
single field of that state gives it away.
1,710 questions 19 decision types 40 domains chance rate 0.258
Each row is a state, a question, a closed set of options, and the index of the one option the
policy selects. The answer is determinate: given the state and the policy there is exactly one
correct choice, and it does not depend on anyone's… See the full description on the dataset page: https://huggingface.co/datasets/limberc/this-that-complex-decisions.next-jev-tetris-decisions-10k
Tetris placement decisions with known answers (10k)
10,000 Tetris decisions for collecting reasoning from an LLM teacher: the teacher writes its reasoning and a final \box{X}, and verify.py keeps only answers that match the known best option.
Each item is exactly what the community Jev Tetris game (commit a0b411c) sends to a local model in its "Laya" slot: a one-paragraph board description, the instruction, and the game heuristic's top 6 placements. Option order is shuffled and… See the full description on the dataset page: https://huggingface.co/datasets/JonesLin/next-jev-tetris-decisions-10k.typed-decisions-repo-governance
typed-decisions: repo-governance family
A seed, not a corpus (n=1 as of first publish, 2026-09-19): real governance findings from a
private software monorepo, each run through TypeSafe's Jev (a "System One"
model — it returns typed choice / score / noul decisions, never free text or code) as part of
kotoba-lang/typed-decisions, a reproduction and
measurement of the Jev shape whose trained model is published at
com-kotobalabs/open-jev-deberta-v3-large.
What this is… See the full description on the dataset page: https://huggingface.co/datasets/com-kotobalabs/typed-decisions-repo-governance.PATRA-EVAL
PATRA-EVAL
Evaluation splits for PATRA: Pattern-Aware Alignment and Balanced Reasoning for Time Series Question Answering (ICML 2026).
Code: https://github.com/decisionintelligence/PATRA
Fields
Each row is a JSON object with:
input (str) — the prompt; a <ts><ts/> placeholder marks where the time series is fed in.
timeseries (list[list[float]]) — the numeric series consumed by the multimodal model.
question_format (str) — one of multiple_choice, true/false… See the full description on the dataset page: https://huggingface.co/datasets/DecisionIntelligence/PATRA-EVAL.swiss_leading_decisions
Dataset Card for Swiss Leading Decisions
Dataset Summary
Swiss Leading Decisions is a multilingual, diachronic dataset of 21K Swiss Federal Supreme Court (FSCS) cases. This dataset is part of a challenging text classification task. We also provide additional metadata as the publication year, the law area and the canton of origin per case, to promote robustness and fairness studies on the critical area of legal NLP.
Supported Tasks and Leaderboards
Swiss Leading… See the full description on the dataset page: https://huggingface.co/datasets/rcds/swiss_leading_decisions.tool-decision-training-pool
Tool calling decision training pool
Public tool-calling data from five datasets, read at the pinned revisions named below and laid out
twice. Every row is a user request with the function declarations offered alongside it, and the
answer is a call on some rows and prose on others, so the pool teaches when to call as well as
how. Train on either layer or on both.
pool.jsonl
Every source rewritten into one shape, 237337 rows, one JSON object per line, with these… See the full description on the dataset page: https://huggingface.co/datasets/Emulated-Inc/tool-decision-training-pool.PATRA-TRAIN
PATRA-TRAIN
Training data for PATRA: Pattern-Aware Alignment and Balanced Reasoning for Time Series Question Answering (ICML 2026).
Code: https://github.com/decisionintelligence/PATRA
Model: DecisionIntelligence/PATRA-7B
Eval data: DecisionIntelligence/PATRA-EVAL
Splits
File
# samples
Stage
sft.jsonl
27,906
Alignment stage — supervised fine-tuning
grpo.jsonl
27,906
Reasoning-enhanced stage — GRPO
Fields
sft.jsonl (columns… See the full description on the dataset page: https://huggingface.co/datasets/DecisionIntelligence/PATRA-TRAIN.morocco-cassation-court-decisions
Morocco Cassation Court Decisions
29,000+ full-text decisions from the Moroccan Court of Cassation (محكمة النقض)Source: juriscassation.cspj.ma — Official portal of the Supreme Council of the Judiciary (CSPJ)License: CC BY 4.0
Why this dataset exists
In 2026, accessing the jurisprudence of the Court of Cassation in Morocco requires being physically located in Morocco and armed with patience. The official website does not allow searching by date range, imposes a… See the full description on the dataset page: https://huggingface.co/datasets/OpenDataMoroccanLaw/morocco-cassation-court-decisions.huggingface_filesystem_terminal_12679_q7v2m9_triage_decisionsmdmp-staff-planning-pairs
mdmp-staff-planning-pairs
Leak-reviewed instruction-tuning pairs for MDMP staff-planning coaching. Public doctrine summaries and fictional scenarios only — no proprietary algorithms, customer data, or classified content.
Disclaimer: Unofficial educational dataset. Not affiliated with the U.S. Army.
Dataset description
324 human-reviewed {instruction, input, output} pairs for fine-tuning a Mistral-7B instruct model on Military Decision-Making Process vocabulary… See the full description on the dataset page: https://huggingface.co/datasets/decisionlens/mdmp-staff-planning-pairs.scopeguard-decisions
ScopeGuard Decisions
ScopeGuard Decisions is a deterministic synthetic instruction dataset for training an LLM to classify an agent request before tools execute.
Splits
Split
Rows
Exact prompt overlap
Train
400
0
Validation
60
0
Test
100
0
Each row uses chat-style messages with a system policy, user request, and compact JSON assistant decision.
Output schema
{
"intent": "send_message",
"constraints":… See the full description on the dataset page: https://huggingface.co/datasets/praveenkumarpranjal/scopeguard-decisions.certo-synthetic-decisions
certo — synthetic decision dataset
Synthetic decisions with a known, exact answer distribution, for training and evaluating
calibrated decision models. Each example is generated from a conditional naive-Bayes evidence
world, so the posterior over the answer is computed in closed form — you can grade a model against
the true probabilities (posterior fidelity), not just accuracy.
Part of certo · project page:
https://altslate-labs.github.io/certo/
Schema (JSONL, one… See the full description on the dataset page: https://huggingface.co/datasets/rajpdus/certo-synthetic-decisions.csc-decision-intelligence-dataset
CSC Decision Intelligence Dataset
Deterministic decision intelligence seeds and cryptographic verification samples for multi-dimensional evaluation protocols.
Dataset Description
This dataset provides deterministic baseline seeds used by the CSC Protocol (@csc-protocol/core) to evaluate institutional, corporate, and healthcare entities under autonomous AI governance rules.
Supported Domains
Healthcare (medical): Facility operational efficiency… See the full description on the dataset page: https://huggingface.co/datasets/csc-architecture/csc-decision-intelligence-dataset.typed-decisions-pt-es
Typed Decisions PT/ES
Unofficial Brazilian Portuguese and Spanish translations of
LocalLLaMA/typed-decisions.
The dataset covers the choice, score, and noul typed-decision tasks. It preserves the
original case IDs, workflows, question IDs, task types, option keys, splits, gold labels, and
soft target distributions.
Contents
Config
Train
Test
pt
1,200
400
es
1,200
400
Each row contains:
id and workflow from the source case;
language, either pt or… See the full description on the dataset page: https://huggingface.co/datasets/telepatia-ai/typed-decisions-pt-es.adaption-preference-trace-decisions
PreferenceTrace — Source Corpus and Adaption Export
PreferenceTrace tests exact decision-making under competing preferences, evidence, approvals, abstention requirements, temporal/contextual precedence, and machine-readable citation contracts.
Two explicit lineage artifacts
File
Rows
Role
SHA-256
preferencetrace-source-96.jsonl
96
Canonical PreferenceTrace source corpus
7a447f9bf47c3ea455ed96ec36860360aa0e7b9e2dc604450e3a1c665b52363e… See the full description on the dataset page: https://huggingface.co/datasets/darthludious/adaption-preference-trace-decisions.decision-model-scenarios
Decision Model Scenarios
2000 synthetic English scenarios with 9716 typed questions and calibrated soft labels. It is the
training data for mghafiri/qwen3.5-0.8B-decision-model, a small model that answers
typed questions about a text "state" with probability distributions instead of generated text.
Each scenario has three parts:
a state: a message, a record, an email thread, a policy plus a ticket, a log, and so on
3–6 independent questions of three types:
Choice: pick one… See the full description on the dataset page: https://huggingface.co/datasets/mghafiri/decision-model-scenarios.cs_czech-court-decisions-ner
Dataset Card for Czech Court Decisions NER
Dataset Description
Czech Court Decisions NER is a dataset of 300 court decisions published by The Supreme Court of the Czech Republic and the Constitutional Court of the Czech Republic.
In the documents, 4 types of named entities are selected.
Dataset Features
Each sample contains:
filename: file name in the original dataset
text: court decision document in plain text
entities: list of selected entities. Each entity… See the full description on the dataset page: https://huggingface.co/datasets/fewshot-goes-multilingual/cs_czech-court-decisions-ner.production-ai-decision-rules
Production AI Decision Rules v0.1
Public v0.1 candidate · reviewed and approved for publication by Dmytro Nasyrov on September 17, 2026.
This public candidate contains 42 decision records and their native method registries. The companion Decision Lab applies scenario inputs, shows a candidate path and retains missing evidence, exclusions and source boundaries.
Pharos Production's published RAG and fine-tuning decision matrix makes selection criteria and stopping conditions… See the full description on the dataset page: https://huggingface.co/datasets/pharosproduction/production-ai-decision-rules.C3R-DecisionMix-v1-preview
C3R DecisionMix v1 Preview
This is a deterministic synthetic schema preview, not the empirical DecisionMix v1 training
corpus and not a benchmark result. It exists so researchers can inspect, load, and validate the
record contract before governed traces are collected and released.
Each record includes the compiled state, typed questions, candidate and counterfactual actions,
baseline action, verifier and task outcomes, latency, cost, risk class, provider identity, and
source… See the full description on the dataset page: https://huggingface.co/datasets/ColomboAI/C3R-DecisionMix-v1-preview.decision-model-hidden-cacheinsurance_decision_boundaries_v1
Dataset Card for insurance_decision_boundaries_v1
Dataset Summary
insurance_decision_boundaries_v1 is a documentation dataset that captures decision boundaries in governed insurance decision support systems. This dataset demonstrates how AI capabilities can support—but never replace—human decision-making in regulated insurance domains.
Each record represents a single decision instance where:
Multiple information sources (rules, data, optional AI signals) are considered… See the full description on the dataset page: https://huggingface.co/datasets/BDR-AI/insurance_decision_boundaries_v1.repro-markov-decision-contests-traces
Agent traces
Agent sessions published from a Trackio Logbook.