The .causal Format: Deterministic Inference for AI-Assisted Hypothesis Amplification

Background: Papers I–IV of the Sovereign Discovery Series generated 22 novel Long COVID hypotheses from
5,084 extracted causal triplets across 376 papers. However, traditional relational database storage (SQLite) limits
discovery to explicitly extracted facts, missing convergence patterns hidden below the detection threshold.

Innovation: We present the .causal binary format—a MessagePack + zlib compressed knowledge graph with
embedded deterministic inference rules. The format achieves 72% storage reduction while amplifying fact counts by
1.90x through three-pass transitive inference: exact keyword matching, semantic direction propagation, and Jaro-
Winkler fuzzy entity resolution.

Key Finding: Weak signals invisible in SQLite (3 triplets) become detectable convergence points in .causal (21+
triplets). This amplification revealed three new hypothesis candidates—Vagus Nerve Convergence (7.0x), POTS/
Autonomic Axis (7.7x), and Autonomic-Cardiovascular Link (30x)—that were previously below the pattern recognition
threshold.

Significance: The .causal format does not replace AI-assisted hypothesis generation—it enhances it. By pre-
computing transitive chains deterministically (zero hallucination risk), the format provides Claude with amplified
signal patterns, enabling discovery of convergence hubs that would otherwise remain hidden in noise.