model-wtf¶

Compliance as code for Model W projects. The tool reads a repository, builds an inventory of what the application does with personal data — the data items, the entry points that touch them, where the data goes, the processing activities — and a threat model on top of it. Every judgement a human or an agent makes is written next to the code as a small YAML file and checked on every pull request: the gate fails a change that introduces a gap, and a challenger agent re-opens the reviews a change undermines.

The result is a register of processing activities (GDPR Art. 30), a rights coverage matrix, a threat model with weighed findings, and a to-do list that is exactly the distance between the code and the declarations.

uv add --dev model-wtf          # or: pip install model-wtf
uv run model-wtf compliance init
uv run model-wtf compliance check

Where to go¶

Principles¶

• Facts from the code, decisions in files. Introspection produces the facts (models, routes, schemas, auth, calls); nothing is guessed from names alone. Every decision — a classification, a declaration, a stamp — is a YAML file under version control, with who made it and why.
• Deterministic first, agents second. Whatever can be decided from a fact is decided by a rule; agents get the questions that need reading, one narrow question at a time, with one write tool each. Their output is checked by the same rules as a human's.
• Nothing silent. Unknown keys are errors. A value nobody knows is !todo, an established gap is !missing; both show up in check until someone answers.
• The gate is a diff. A pull request is compared to its base by finding identity; it fails only on what it introduces. Pre-existing findings are listed, never blocking.