Program Prototype

MCAP

Minimal Causal Abstraction Principle

Summary

MCAP is a proposed principle for building minimal representations of systems that preserve intervention-relevant causal structure. The approach focuses on compression that maintains causal fidelity under interventions, enabling more efficient and interpretable models while retaining the ability to reason about cause and effect.

In Scope
  • Intervention-relevant causal structure preservation
  • Minimal representation compression
  • Causal fidelity evaluation under interventions
  • Baseline comparisons with existing methods
Out of Scope
  • General-purpose causal discovery
  • Non-interventional causal inference
  • Real-time production deployment
Evidence Capsule
What Exists Now
  • Defined abstraction objective (intervention-relevant compression)
  • Initial framework and evaluation criteria
What Comes Next
  • Planned: causal fidelity checks under interventions
  • Planned: baseline comparisons (causal discovery + representation learning)
  • Planned: identifiability assumption validation
Evaluation Protocol

Evaluation focuses on four key dimensions: (1) identifiability assumptions and their testability, (2) interventional accuracy compared to ground truth, (3) stability under distribution shift, and (4) ablation against simpler baselines. The protocol emphasizes preregistered evaluation where possible to ensure reproducibility and reduce bias.

Benchmarks will include synthetic causal structures with known ground truth, real-world datasets with verified causal relationships, and stress tests under distribution shift. All results will be reported with uncertainty estimates and failure mode analysis.

Access & Governance
  • Public Overview: This page and high-level documentation are publicly accessible.
  • Controlled Demo: Interactive demonstrations available to qualified researchers upon request.
  • Partner Evaluation: Full technical documentation, evaluation protocols, and access to experimental harnesses available to research partners.