# Reproduction and citation

Run `python tools/reproduce_analysis.py` with Python3.10+ and matplotlib to regenerate9CSV datasets and9PNG/SVG pairs. PNGs are2400×1440. Numeric outputs are reproducible; image bytes can vary with font/library versions. Ten arithmetic assertions must pass.

## Model boundaries
M/G/1: Wq=[rho/(1−rho)]×[(1+CV²)/2]×s. Poisson arrivals,independent iid service,finite second moment,one continuously available FCFS processor,rho<1. Wq is waiting; W=Wq+s. A daily load ratio is a separate planning heuristic when staff are not continuously available.

Fluid backlog: B/(mu−lambda),constant rates,mu>lambda. B0returns0; this does not guarantee ongoing stability.

Filing: Tsearch=R×ts; Tfile=N×f+R×tf. Equal success/rework initially. If ts<=tf and preparation is positive,filing cannot win on this time-only model.

AI: manual−assisted−extra review−(extra words×full readings/wpm)−extra followup. Differences may be negative. Preserve meaning and quality. Do not double-count baseline review or price released capacity as guaranteed payroll savings.

## Publication
Use actual publication date, existing verified editorial author, canonical URL,title,and version1.0. Research checking date is not publication date. No fabricated DOI,peer review,independent institute,participant study,or product effect.

Empirical discoveries belong to their authors. Attribute only these editorial calculations and new graphics to InboxCabin. Data tables expose selected reported aggregates,not raw participant data. Source papers are linked,not redistributed.

Before announcing a public reuse license,the owner must decide those terms. This package does not grant rights to third-party papers/logos or declare a Creative Commons license on the owner's behalf. Suggested citation text is descriptive,not permission from researchers.