Skip to content

Where the research write-ups are

This repository holds the instruments and the evidence. It does not hold the write-ups.

docs/*.json in the repository carries the measured results — fragment membership across six policy languages, the estimator comparison, the threshold and trend analyses, the solver certificate — and scripts/ carries the code that produced each one. Every figure is reproducible from what is here: point an adapter at the corpus commit its artifact records and you get the same numbers.

The prose that explains those numbers is kept outside the repository until the work is submitted. A journal, and the similarity check it runs, treats a publicly posted full text as prior dissemination, and a public repository is publicly posted. That applies to the manuscript and equally to the long-form development it was written from, so the theory write-up, the suite-coverage study, the readiness assessment and the prospectus all live beside it rather than here.

Nothing is hidden by this and nothing is lost. The artifacts are the claims; the write-ups argue about them. A reader who wants the argument before publication should ask the authors for it.

Running the checks that need them

Two guards read the theory write-up, because a hand-copied table in a proof document is a claim like any other and should not go unchecked. They follow it through an environment variable and skip when it is absent:

TRUSTWEAVE_RESEARCH_DIR=/path/to/write-ups pytest tests/test_decision_class_theory.py
TRUSTWEAVE_PAPER=/path/to/main.tex python scripts/check_manuscript.py

Without those set, both report that they have nothing to check and pass. A checkout without the write-ups is not a checkout with wrong ones.

What stays here

  • scripts/fragment_membership*.py, scripts/policy_mutation.py and the rest of the measurement code.
  • scripts/oracle_rego.py and scripts/interpreter_oracle.py, which check those two instruments against something other than their authors' reading of them: the Rego adapter against OPA's own dependency analysis, and the decision map against the engine that ships. Each needs its subject present — opa on the path and the pinned corpora for the first — and each writes an artifact under docs/ that lists what it found.
  • docs/*.json — every artifact, each recording the corpus commit it read.
  • The engineering documents that describe the tool rather than the research.

Cite the artifacts if you need a figure before the paper appears. They are the primary record, and the write-ups quote them rather than the other way round.