ROOMFORCHANCE LAB — REPRODUCTION RECORD
Published: 22 September 2026

These are synthetic model experiments, not historical lottery results.
Download run_experiments.py and results.json. Use the Python and NumPy versions
recorded in results.json for exact replay. Install NumPy 2.3.5, then run:
  python run_experiments.py
The program creates lab-data beside itself and writes eleven CSV files and a
results manifest. It runs ten million draws in its largest experiment.
Each experiment initializes an independent PCG64 stream with the recorded seed.

CSV types:
- hot-cold-holdout.csv: one row per independent trial (20,000 rows).
- overdue-gaps.csv: prespecified bins of completed waiting times.
- other CSVs: complete aggregate frequency or summary tables.
No CSV is described as a full raw-draw archive. Raw simulations can be recreated
from the code, seed and environment. CSV SHA-256 values appear in results.json.

Figure reconstruction:
Download science-figures.py. It expects results.json at public/lab/data/results.json
relative to its working directory, and writes SVG and PNG files to
public/science/figures. Install matplotlib and numpy, then run the figure script.
The source uses mathematical reference curves or published aggregate results;
figure captions distinguish them. For identical font rendering, use DejaVu Sans.

The browser simulator uses Mulberry32 v1 and is deliberately a different stream.
Its CSV includes inputs and seed. It is not an exact replay of the Python studies.

DATA AND ORIGINAL FIGURES: CC BY 4.0.
Attribute RoomForChance and link to https://www.roomforchance.com/lab/ or the
specific experiment. See https://creativecommons.org/licenses/by/4.0/.
ORIGINAL CODE: MIT license; see LICENSE.txt.