Known limitations

Documentation vs code

The extractor is deterministic and doesn't interpret sentences. It handles migrations, short negations, comparisons and common words (see Documentation that lies), but not:

• long or indirect negations ("at this stage the project does not need a PostgreSQL database" — the negation is only detected up to two words before the name);
• conditional or future sentences ("we will use PostgreSQL in production");
• technologies missing from the detection table;
• a project that legitimately uses two relational databases where only one has signals in the code.

Code analysis

• Only Python, JavaScript and TypeScript are parsed; other languages are counted and their manifests read.
• Route prefixes are resolved inside a file and through local imports (include_router, register_blueprint, app.use, fastify.register); routes built dynamically are not detected.
• Notebook code is counted, not parsed.

Interface

The interface is available in English (default) and Portuguese. Profiles saved by version 0.1.x keep the Portuguese text they were saved with until the project is analysed again. Technology names in the documentation are matched in any language, but the sentences that mark a mention as ambiguous ("migrated from", "instead of", "no longer"…) are recognised only in English and Portuguese.

The full list (in Portuguese) is kept in docs/limitacoes.md.