Are Whitepaper Claims Reflected in Market Structure? A Contamination-Aware Pipeline and a Power-Limited Null

Battle Count: 77

The paper reports a null result: whitepaper topical emphasis does not align with market structure statistics beyond chance. It does not provide trading signals, predictive models, or actionable alpha. Its primary contribution is methodological (auditable non-detection, corpus diagnosis, specification sensitivity) rather than directly applicable to trading strategy development. The narrative-economics framing and cryptocurrency market statistics are tangentially relevant to sentiment-based strategies, but the paper explicitly disclaims predictive or causal claims.

The pipeline involves: (1) zero-shot NLP classification with BART-MNLI requiring GPU inference and model weights, (2) hourly-to-daily market data aggregation with multiple volume specifications, (3) PCA dimensionality reduction, (4) orthogonal Procrustes rotation with permutation testing (1,000 refits), (5) four-leg decomposition analysis, and (6) 48-cell numerical simulations with 14,400 synthetic datasets. The numerical verification script is self-contained, but full text-to-score replication requires the complete model stack. Moderate complexity due to multiple interacting specifications and sensitivity analyses.

Code, frozen score matrices, source data, and provenance manifest are all available on GitHub. A single verification script (verify_release.py) rebuilds all numerical results from Parquet bars with fixed seeds. Environment lock and checksums accompany the release. The analysis uses frozen classifier profiles to separate numerical verification from model-version drift. All 48 simulation cells and seeds are archived in power.json.