SEMCA-7 v1.0.0

SEMCA-7 — first public release

Substrate-agnostic operationalization of seven mathematical consciousness theories applied identically to AI transformer activations and human fMRI BOLD signals on the same naturalistic stimuli.

Paper: 10.5281/zenodo.20435290 (Zenodo, CC-BY-4.0). LaTeX source + bibliography in paper/.

Findings

1. Population magnitudes overlap for four of eight theories across AI and human substrates. IIT, GWT, FEP, and the unified score show sub-5-point cross-substrate gaps. AST, HOT, PPT, and QIT show substantial cross-substrate divergence (gaps from +16 to −52).
2. Per-stimulus rankings: six of seven theories show no detectable cross-substrate correlation (Pearson r ∈ [−0.17, +0.18]). Global Workspace Theory is the partial exception (r = +0.365, permutation p = 0.001, split-half stable), though its cross-substrate signal is architecture-dependent.
3. AI-side per-story variance is dominated by architectural rather than stimulus-driven sources for six of seven theories. The apparent magnitude alignment for the four aligned theories is coincidental overlap of fundamentally different variance sources, not evidence of shared signal.

Fisher-Rao information-geometric integration preserves the null cross-substrate correlation (Riemannian-mean r = −0.185, stable across uniform/identity/20 random alternative compatibility-prior matrices).

Theories

IIT, GWT, AST, HOT, PPT, QIT, FEP — each implemented as a substrate-agnostic function of a single abstract Substrate type with a (T × N) activity matrix. The same Python code runs on transformer attention activations and on K-means-parcellated fMRI BOLD signals.

Reproducing

pip install -r requirements.txt
python -m src.population_cross_substrate
python -m src.perstory_cross_substrate
python -m src.robustness_perstory_analysis
python -m src.cross_substrate_geometric
python -m src.geometric_sensitivity

Each analysis reproduces from pre-computed scores in under one minute on a laptop. Full re-derivation from raw OpenNeuro data takes ~3 hours on 2 × H100.

Cite

@misc{travis2026semca7,
  title  = {Architectural Variance Dominates Stimulus Variance in
            Six of Seven Substrate-Agnostic Consciousness Operationalizations},
  author = {Travis, Nate},
  year   = {2026},
  publisher = {Zenodo},
  doi    = {10.5281/zenodo.20435290},
  url    = {https://doi.org/10.5281/zenodo.20435290}
}

License

MIT (code + data). Paper PDF on Zenodo is CC-BY-4.0. The LeBel ds003020 raw dataset is CC0 from OpenNeuro; pre-computed substrate scores derived from it are released under the same terms.