GammaSB

GammaSB is a public reference implementation of the GammaSB access-ranking score used in the hidden-spectator Ramsey study. This repository keeps the reusable method code, synthetic benchmarks, and a public-data validation entry point.

Zenodo concept DOI for all versions: 10.5281/zenodo.19645047

Repository scope

This public release includes:

• the reduced hidden-spectator model;
• the resource-bundle and ranking utilities;
• synthetic ranking benchmarks;
• a dual-rail Ramsey validation script that reads public external data supplied by the user;
• a public same-device blind validation package for the fixed-device no-refit access test.

This public release does not include:

• raw private data beyond the released same-device validation package;
• generated result files;
• manuscript sources, review materials, or confidential notes;
• local configuration files, credentials, tokens, or machine-specific paths.

Installation

python -m pip install -e .

Quick smoke test

Run a short local check before any larger benchmark:

gammasb smoke

Synthetic benchmarks

Hard-regime ranking benchmark:

gammasb hard-regime --episodes 160 --output-dir outputs/hard_regime

Response-zero rescue benchmark:

gammasb response-zero --draws 48 --noise-seeds 6 --workers 8 --output-dir outputs/response_zero

Dual-rail Ramsey validation

The dual-rail validation expects the public Zenodo files from DOI 10.5281/zenodo.11099521.

Place these two files in a local directory:

• ExtendedDataFig3_ramsey_short_time.csv
• ExtendedDataFig3_ramsey_short_time_phases.csv

Then run:

gammasb dualrail --data-dir /path/to/zenodo_11099521 --output-dir outputs/dualrail

Same-device public validation package

The repository also ships a compact public package for the fixed-device same-device headline:

• data/same_device_blind_package/low_access_only/low_access_data.csv
• data/same_device_blind_package/low_access_only/candidate_probe_ranking_locked.csv
• data/same_device_blind_package/held_out_probe/held_out_probe_data.csv
• data/same_device_blind_package/negative_controls/same_family_negative_controls.csv
• data/same_device_blind_package/manifest.json

This package exposes the frozen low-access block, the locked candidate ranking, the held-out probe values, and the same-family negative controls used in the no-refit audit order.

Python API

from gammasb import ResourceBundle, HiddenAncillaSolver
from gammasb.hard_regime_rank_benchmark import build_hard_regime_rank_benchmark

bundle = ResourceBundle(time_samples=4, observable_count=1, control_depth=0, correlation_order=1)
outputs = build_hard_regime_rank_benchmark(episode_count=8, seed=20260415)
print(outputs.summary)

Citation

If you use this repository, cite the concept DOI:

Lu, Y. GammaSB. Zenodo. https://doi.org/10.5281/zenodo.19645047

Layout

• data/same_device_blind_package/
• src/gammasb/
• pyproject.toml
• README.md
• CITATION.cff
• .zenodo.json