Supplementary Material
======================

"Early Symmetry-Breaking Predicts Attractor Selection in a Bistable
 Two-Player Learning Game"

Contents
--------

1. attractor_selection_experiments.py
   Self-contained Python script that reproduces all numerical results
   reported in the paper. Deterministic given the fixed random seeds
   specified within the script.

2. verified_output.txt
   Complete stdout from running the script, for verification. Readers
   can run the script and compare their output against this file to
   confirm reproducibility.

Requirements
------------

- Python 3.9+
- numpy
- scipy
- scikit-learn

Install dependencies:
  pip install numpy scipy scikit-learn

Usage
-----

  python attractor_selection_experiments.py

Runtime: approximately 6-9 minutes on a single CPU core.
Output is printed to stdout.

What the script reproduces
--------------------------

Section 4.1 (Bistability):
  - Cycling fractions across five independent seed banks for the
    two-player anti-coordination game (~11-18% cycling).
  - Confirmation that the canonical N-player Minority Game cycles
    in 100% of runs for N = 3, 5, 7, 11, 21.

Section 4.3 (Predictor):
  - Leave-one-out cross-validated AUC for value-function asymmetry
    at timestep 100 (200 seeds, with bootstrap CI and permutation test).
  - AUC on the undecided subset (undecided-subset control).
  - Replication on an independent bank of 300 seeds.

Section 4.4 (Transfer entropy leakage):
  - Transfer entropy AUC on all runs versus undecided runs, across
    measurement windows of 100, 250, and 500 timesteps.

Notes
-----

- All randomness is seeded. Results are deterministic.
- No external datasets are required. All data are generated
  programmatically within the script.
- The script does not produce figures. Figures in the paper were
  generated from the same underlying simulation code with
  matplotlib; the figure-generation code is available from the
  author on request.