COREX — Causal Origin Resolution and Empirical eXamination

Decision Logic

Three-Class Causal Classification

Every relationship evaluated by COREX receives a calibrated score and a categorical label with full technical justification.

≥ 0.80

🟢 CAUSAL

Survives all four evaluation axes. Intervention consistency confirmed. Strong evidence of a genuine causal mechanism.

0.50 – 0.79

🟡 SPURIOUS

Breaks under domain shift or intervention instability. Statistical association exists without causal mechanism.

< 0.50

🔴 ARTIFACT

Representation invariance fails. The relationship exists only as a function of the data encoding choice — not reality.

Evaluation Pipeline

Four-Axis Robustness Architecture

No prior causal graph required. COREX operates as a meta-evaluation layer atop any learned relationship.

MODULE 01

Statistical Stability

Evaluates whether P(Y|X) remains invariant across independently drawn subpopulations using KL divergence and KS tests.

S = 1 − mean KL[P(Y|X,D₁) ‖ P(Y|X,D₂)]

MODULE 02

Representation Invariance

Tests stability under linear projections, nonlinear embeddings, noise injection, feature dropout, and PCA compression.

R = 1 − (1/|Φ|) Σ ‖P(Y|X) − P(Y|φ(X))‖₁

MODULE 03

Intervention Consistency

Simulates causal interventions via propensity-score matching, counterfactual generation, and direct feature perturbation.

I = Consistency(do(X=x₁)→Y₁, do(X=x₂)→Y₂)

MODULE 04

Domain Robustness

Evaluates cross-environment generalization using coefficient of variation across distinct distributional regimes.

D = 1 − CV(P(Y|X, e)) over e ∈ E

Quick Start

Up and Running in 60 Seconds

pip install corex-causal

from corex import CausalEvaluator
# Initialize the evaluator
evaluator = CausalEvaluator()

# X: features (n_samples, n_features) | Y: target (n_samples,)
result = evaluator.evaluate(X, Y)

print(result.label)        # "CAUSAL" | "SPURIOUS" | "REPRESENTATION_ARTIFACT"
print(result.corex_score)  # float in [0, 1]
print(result.breakdown)    # {"S": 0.91, "R": 0.88, "I": 0.85, "D": 0.90}
print(result.failure_modes) # list of detected failure modes

from corex import CausalEvaluator

evaluator = CausalEvaluator(
    weights={
        "statistical":    0.25,
        "representation": 0.25,
        "intervention":   0.30,  # highest: closest to true causality
        "domain":         0.20
    }
)
result = evaluator.evaluate(X, Y, environments=env_labels)
print(result.corex_score, result.label)

from corex import CausalEvaluator
from corex.learnable import MetaScorer

# Load pre-trained adaptive weight estimator
meta = MetaScorer.from_pretrained("default")
evaluator = CausalEvaluator(meta_scorer=meta)

# Meta-scorer calibrates weights based on dataset properties
result = evaluator.evaluate(X, Y)
print(result.meta_confidence)  # P(Causal | score_vector)

Citation

Cite COREX in Your Research

If COREX contributes to your research, please use one of the citation formats below.

@software{baladi2026corex_pypi,
  author    = {Baladi, Samir},
  title     = {{COREX}: Causal Origin Resolution and Empirical eXamination},
  year      = {2026},
  version   = {1.0.0},
  publisher = {Python Package Index},
  url       = {https://pypi.org/project/corex-causal-causal-causal},
  note      = {Python package, MIT License}
}

@dataset{baladi2026corex_zenodo,
  author    = {Baladi, Samir},
  title     = {{COREX}: Causal Origin Resolution and Empirical eXamination},
  year      = {2026},
  publisher = {Zenodo},
  version   = {1.0.0},
  doi       = {10.5281/zenodo.20351233},
  url       = {https://doi.org/10.5281/zenodo.20351233},
  series    = {BIO-MED-02}
}

@misc{baladi2026corex_osf,
  author    = {Baladi, Samir},
  title     = {{COREX} Framework: Pre-registered Study Protocol},
  year      = {2026},
  publisher = {Open Science Framework},
  doi       = {10.17605/OSF.IO/XXXXX},
  url       = {https://doi.org/10.17605/OSF.IO/XXXXX},
  note      = {OSF Preregistration}
}

@article{baladi2026corex,
  author  = {Baladi, Samir},
  title   = {{COREX}: An Autonomous Multi-Stage Framework for Robust
             Causal Discrimination in Data-Driven {AI} Systems},
  year    = {2026},
  month   = {May},
  series  = {BIO-MED-02},
  version = {1.0.0},
  doi     = {10.5281/zenodo.20351233},
  url     = {https://doi.org/10.5281/zenodo.20351233}
}

Baladi, S. (2026). COREX: An Autonomous Multi-Stage Framework for
Robust Causal Discrimination in Data-Driven AI Systems
(Version 1.0.0, Series BIO-MED-02). Zenodo.
https://doi.org/10.5281/zenodo.20351233