[nikolas-sapa]

Documents the RECOGNIZER_REGISTRY_CONF_FILE override offered in #2256 (part 2), where the reporter showed that ItFiscalCodeRecognizer — registered with supported_languages: [it] — is unreachable from the default analyzer image, which loads only the en NLP model.

Adds a section to docs/analyzer/recognizer_registry_provider.md covering:

• how the server selects the registry file via RECOGNIZER_REGISTRY_CONF_FILE (plus a mention of the sibling ANALYZER_CONF_FILE / NLP_CONF_FILE), previously undocumented anywhere in docs/
• a worked example enabling ItFiscalCodeRecognizer for en with a mounted registry file — the exact one-line override the reporter validated end-to-end through LLM gateways
• a cross-link to Filtering recognizers by country for the general supported_countries filter

Docs-only change; no code touched. Related: #2259 covers part 1 of the same issue.

[GiorgioDotcom]

Confirming the example against what I measured for #2256: this is the exact override I ran, a full copy of default_recognizers.yaml with supported_languages changed to [en] on ItFiscalCodeRecognizer and country_code: it left as is, mounted via RECOGNIZER_REGISTRY_CONF_FILE. With it, IT_FISCAL_CODE is detected for language: en, including omocodia variants, and the same request without the override returns nothing.

The note about supported_languages having to match the analyzer configuration is worth keeping: when I set the registry-level supported_languages to [en, it] while the analyzer config still listed only en, the server exited at startup with "Misconfigured engine, supported languages have to be consistent" (exit code 3). The three variants I tried, with the outcome of each, are here: https://github.com/GiorgioDotcom/llm-gateway-privacy-bench/tree/master/presidio-cf/verifica-pr2271

Thanks for picking this up.