What is extracted from every resume. To capture something new, add a field under `additional_information` — it reaches the database with no migration. Fields in the other blocks map to real columns and cannot be added here.
How the model is told to read a resume, and the output contract.
How a plain-English search is turned into SQL.
The skill vocabulary: canonical names, the spellings that mean the same thing, which section a skill belongs to (`category`, e.g. `cloud` — this is what a search for "cloud experience" matches), and which platform it sits under (`parent`, e.g. Amazon EC2 under AWS). Saving re-seeds the skill tables and re-runs matching for skills that previously went unrecognised, so an edit reaches resumes that are already parsed. Removing a line retires that skill from future matching; candidates who already matched it keep it, because those rows record what a document actually said.