Implementations
Where the working code lives.
| Library | Role | Install | Links |
|---|---|---|---|
| thurstone | The fast ability transform: forward (abilities → win probabilities) and inverse (probabilities → abilities), arbitrary base densities, dead-heat handling, dynamic tracking. | pip install thurstone |
GitHub · docs & demos |
| kinetics | This repository: identity-verification experiments, synthetic KMC testbeds, and the watermark-races study — exploratory code that graduates elsewhere when stable. | git clone |
GitHub · experiments · watermark_races |
| precise | Online covariance estimation and Schur-complement machinery — relevant to the rank-one cavity side and to streaming maintenance (Q5). | pip install precise |
GitHub |
| schur | Sibling research site: the same Schur-complement object in portfolio construction and spatial pseudo-likelihoods. | — | GitHub · site |
Running the experiments
The identity checks in
experiments/
need only numpy:
python experiments/cavity_downdate_demo.py # rank-one cavity: one inverse, all deletions python experiments/race_field_demo.py # multiplicative cavity: one field, all races
The watermark study in
watermark_races/
additionally needs PyTorch and Hugging Face Transformers; see its README.