spectralbranding/verification-bandwidth-derived-results

Verification Bandwidth Under Correlated Evaluators — derived results

Derived numerical results for the paper Verification Bandwidth Under Correlated Evaluators: What an Effective-Sample-Size Statistic Measures in an Acceptance Cascade.

• —Paper concept DOI: 10.5281/zenodo.21891435
• —Paper version DOI (v1.0.0): 10.5281/zenodo.21891436
• —Code: github.com/spectralbranding/orgschema-papers/tree/main/verification-bandwidth/code
• —This dataset's DOI: 10.57967/hf/9953

What this is, and what it is not

This is not an observational dataset, and it should not be cited as evidence. The paper collected no data. It is a theory-and-computation paper whose reported figures come from two places: arithmetic on summary statistics another study published, and seeded simulation of a stated model. This record is the output of that computation — the tables the paper prints, plus the captured stdout of every script that produced them.

Its purpose is auditability rather than reuse. A reader who wants to check a number in the paper against the code that produced it can diff this record instead of re-running anything; a reader who wants to re-run it can, from the repository above, with no network access and no key.

Contents

Provenance and how to reproduce

Every value is produced by reproduce.sh in the repository above, which runs six scripts in dependency order. All fix SEED = 20260811 at file top and exit nonzero if any internal check fails. The whole pipeline runs in well under a minute and requires only Python 3.12 with numpy, scipy and matplotlib.

git clone https://github.com/spectralbranding/orgschema-papers
cd orgschema-papers/verification-bandwidth
./reproduce.sh

Tables 2 and A1 and the worst-case block of Table 5 are seeded Monte Carlo. They appear here only as captured stdout, not as CSV, because emitting them would require a second implementation of a seeded simulation — the exact drift the paper's reproducibility standard exists to prevent. The script is the ground truth for any value the paper calls computed.

Known limits of what these numbers mean

• —Table 3 is arithmetic on another study's published figures, not a re-analysis of its panel. No raw judgements were obtained; the agreement it reports is between a formula and a published number.
• —Table 4 is model-dependent and directional. The point inversion needs a marginal error rate the published record does not carry, so it is bracketed over three values rather than fixed, and no shared-difficulty correction is applied.
• —Table 6 rows are illustrative combinations, not measurements of any organization, and are computed under one-dimensional inspection subspaces.
• —The simulations assume isotropic deviations, rank-one inspection, and a miss-only error model with no false-alarm arm. Each is a stated scope condition of the paper, not a defect of this record.

Citation

Cite the paper, not this record:

Zharnikov, Dmitry (2026), Verification Bandwidth Under Correlated Evaluators: What an Effective-Sample-Size Statistic Measures in an Acceptance Cascade. Working Paper v1.0.0. DOI: 10.5281/zenodo.21891435

License

CC BY 4.0. The code that produced these results is MIT, at the repository above.