den-dynamiske-rettsstaten/CLAUDE.md

den-dynamiske-rettsstaten

Context

Concept note (Norwegian, LaTeX) and its simulation code. The principle: enforcement effort should track how much, and how seriously, laws are actually broken. Published as open, non-commercial material. Readers are Norwegian, so README and paper are in Norwegian (bokmål); CLAUDE.md and code comments may stay as they are.

Stack

• paper/sim.py: Python 3, numpy + scipy (pinned in requirements.txt). Writes paper/data/*.dat, which pgfplots in the LaTeX source reads.
• paper/*.tex: LaTeX. main.tex and notat.tex build with pdflatex, karusell.tex with xelatex (font Inter from TeX Live). main.tex = the note (inputs part1-part4, refs), notat.tex = 2-page summary, karusell.tex = slides. Shared preamble.tex. Bibliography is a hand-written thebibliography in refs.tex.
• pdf/: the published PDFs. They are build output committed on purpose; rebuild and replace them together with source changes.

Conventions

• The core sentence must be quoted verbatim everywhere: «Innsatsen for å håndheve lover og regler bør løpende stå i forhold til hvor mye, og hvor alvorlig, de faktisk brytes.» Short form of the clarification: «Straffebud og strafferammer ligger fast.»
• Any change to sim.py must keep paper/data/ byte-identical unless the change is meant to alter results; then text, figures and PDFs change in the same commit.
• Licensing is split: text (pdf/, paper/*.tex, README.md, CHANGELOG.md, CLAUDE.md) is CC BY 4.0 (LICENSE-CC-BY-4.0.txt); code and data (paper/sim.py, paper/diagnose.py, paper/forlop.py, paper/eksempel_domstol.json, paper/norge_2025.json, requirements.txt, paper/data/) are MIT (LICENSE). New files follow the same split.
• Never commit working documents; private/ is ignored for that purpose.

Commands

• Setup: python3 -m venv .venv && .venv/bin/pip install -r requirements.txt
• Run simulation (about 2 min, from paper/): ../.venv/bin/python sim.py
• Bottleneck diagnosis from own numbers (stdlib only, from paper/): python3 diagnose.py eksempel_domstol.json; must reproduce section 9.5 (utilization 1,28, +19 % capacity). python3 diagnose.py norge_2025.json must match the table in README (1,09 / 1,04 / 0,88 / 1,09).
• Time course and stability check from own numbers (needs the venv, from paper/): ../.venv/bin/python forlop.py eksempel_domstol.json; must reproduce section 9.5 and table 10 (critical delay 7,2; +19 % now, +12 % after 12 months, capacity 137,9 = 1,31; with measures 8,3 / 0,74 / 1,10 / +31 % / 3,5; without 68,7 / 0,65 / 1,26). A copy of the example with restanse 3000 for Domstol must give a course that starts at 3 000 (first decision +119 %, as diagnose.py). forlop.py imports sim.py and must never require a change there. It has its own copy of the simulate_endo loop (simuler, policies koord and sched only) so that the backlog in the file can be put in directly; every run checks that the copy matches simulate_endo bit for bit, so a change to simulate_endo must be mirrored in simuler.
• Reproduction check after a run: git status paper/data shows no changes (verified 2026-11-10: 22/22 files identical with Python 3.14.0, numpy 2.5.4, scipy 1.18.1)
• LaTeX build (from paper/, TeX Live 2026 in /Library/TeX/texbin): pdflatex main.tex and pdflatex notat.tex three times each, xelatex karusell.tex twice; copy to pdf/ (verified 2026-11-10: main 58 pages, notat 2, karusell 16, no undefined references). Build in a scratch copy of paper/ to keep aux files out of the repo, and give each run a time limit: one pdflatex run has been seen to hang.

Workflow

• Published at https://git.fromaitochitta.com/open/den-dynamiske-rettsstaten. Visibility, description and moves are the operator's decisions.
• Run the repo-standard gate before pushing changes to README.md or the files the standard requires.