Add services/guidance_ingester.py — extracts guidance documents (pdfplumber for
PDF, an HTML stripper otherwise; pdfplumber is imported lazily so the module and
its tests load without it) and tags them as a SEPARATE interpretative source:
source_class=supervisory_guidance / authority_weight=70 / bindingness=
interpretative / use_for_primary=false, with references_out to the binding norms
they interpret (Art. N DSGVO / § N BDSG). Guidance therefore ranks below binding
law for obligation questions yet stays retrievable as interpretation context.
supervisory_guidance is reused deliberately: the live re-ranker already weights
it 70 and 8k+ chunks use it (no classifier change, no schema drift). EDPB is the
first target; technical standards (weight 80) are a later separate class.
Tested: 6 unit tests on the text + metadata path (PDF extraction is exercised in
the container), ruff + mypy clean.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Benjamin Admin