Datasets:

chkmie
/

dotcausal

license: mit
language:
  - en
tags:
  - knowledge-graph
  - causal-inference
  - rag
  - zero-hallucination
  - triplets
pretty_name: dotcausal Dataset Loader
size_categories:
  - n<1K

dotcausal - HuggingFace Dataset Loader

Load .causal binary knowledge graph files as HuggingFace Datasets.

What is .causal?

The .causal format is a binary knowledge graph with embedded deterministic inference. It solves the fundamental problem of AI-assisted discovery: LLMs hallucinate, databases don't reason.

Technology	What it does	What's missing
SQLite	Stores facts	No reasoning
Vector RAG	Finds similar text	No logic
LLMs	Reasons creatively	Hallucination risk
.causal	Stores + Reasons	Zero hallucination

Key Features

30-40x faster queries than SQLite
50-200% fact amplification through transitive chains
Zero hallucination - pure deterministic logic
Full provenance - trace every inference

Installation

pip install datasets dotcausal

Usage

Load from local .causal file

from datasets import load_dataset

# Load your .causal file
ds = load_dataset("chkmie/dotcausal", data_files="knowledge.causal")

print(ds["train"][0])
# {'trigger': 'SARS-CoV-2', 'mechanism': 'damages', 'outcome': 'mitochondria',
#  'confidence': 0.9, 'is_inferred': False, 'source': 'paper_A.pdf', 'provenance': []}

With configuration

# Only explicit triplets (no inferred)
ds = load_dataset(
    "chkmie/dotcausal",
    "explicit_only",
    data_files="knowledge.causal",
)

# High confidence only (>= 0.8)
ds = load_dataset(
    "chkmie/dotcausal",
    "high_confidence",
    data_files="knowledge.causal",
)

Multiple files / splits

ds = load_dataset(
    "chkmie/dotcausal",
    data_files={
        "train": "train_knowledge.causal",
        "test": "test_knowledge.causal",
    },
)

Dataset Schema

Field	Type	Description
`trigger`	string	The cause/trigger entity
`mechanism`	string	The relationship type
`outcome`	string	The effect/outcome entity
`confidence`	float32	Confidence score (0-1)
`is_inferred`	bool	Whether derived or explicit
`source`	string	Original source (e.g., paper)
`provenance`	list[string]	Source triplets for inferred facts

Creating .causal Files

from dotcausal import CausalWriter

writer = CausalWriter()
writer.add_triplet(
    trigger="SARS-CoV-2",
    mechanism="damages",
    outcome="mitochondria",
    confidence=0.9,
    source="paper_A.pdf",
)
writer.save("knowledge.causal")

References

PyPI: https://pypi.org/project/dotcausal/
GitHub: https://github.com/DT-Foss/dotcausal
Whitepaper: https://doi.org/10.5281/zenodo.18326222

Citation

@article{foss2026causal,
  author = {Foss, David Tom},
  title = {The .causal Format: Deterministic Inference for AI-Assisted Hypothesis Amplification},
  journal = {Zenodo},
  year = {2026},
  doi = {10.5281/zenodo.18326222}
}

License

MIT