fix(analyzer): stop PhoneRecognizer leaking a match's region into later matches

CHANGELOG.md

@@ -17,6 +17,7 @@ All notable changes to this project will be documented in this file.
 - The `kr` language code is deprecated in favor of the ISO 639-1 code `ko`. `KrRrnRecognizer` and `KrPassportRecognizer` still accept `kr` in `supported_languages` as a backward-compatibility alias for registries configured against their original class defaults; the alias will be removed in <release TBD>. `KrBrnRecognizer`, `KrDriverLicenseRecognizer` and `KrFrnRecognizer` have only ever defaulted to `ko`, so `kr` was removed from their `default_recognizers.yaml` entries in this release. (#2236)
 
 #### Fixed
+- `PhoneRecognizer.analyze` assigned the region detected for a match back to the `region` variable it was iterating over, so an international-format number that parsed successfully overwrote the region used to explain every later match in that pass. With the default recognizer, `"My international number is +44 1234 567890, and my US one is (415) 555-0132"` labelled the US number `Recognized as GB region phone number`; swapping the two numbers hid it. Detected spans, scores and entity types were unaffected — only `analysis_explanation` named a region that never matched the number. The detected region is now held in a per-match `matched_region`, and the `NumberParseException` branch reports the region the match was actually found under.
 - `PhoneRecognizer.DEFAULT_SUPPORTED_REGIONS` used `"UK"`, which is not a valid `phonenumbers` (libphonenumber) region code — region codes are ISO 3166-1 alpha-2, where the United Kingdom is `"GB"`. The `"UK"` entry was a no-op, so UK numbers in national/local format (e.g. `020 7946 0958`) were never detected by default; only international-format `+44 …` numbers matched, because they carry the country code and match under any region. Replaced `"UK"` with `"GB"`.
 - Language model recognizers (`BasicLangExtractRecognizer`, `AzureOpenAILangExtractRecognizer`) configured in a recognizer registry YAML now honour `config_path` (and other recognizer-specific kwargs). Previously these entries were validated by the strict `PredefinedRecognizerConfig` schema, which has no `config_path` field and does not allow extra keys, so `config_path` was silently dropped and the recognizer fell back to its bundled default model configuration. Added a `LangExtractRecognizerConfig` model (`extra="allow"`) and registered both recognizer class names in `CONFIG_MODEL_MAP`.
 - `BasicLangExtractRecognizer` now honours values under `langextract.model.provider.language_model_params` (including `timeout` and `num_ctx`). Previously these were silently dropped because `langextract.extract()` ignores its `language_model_params` argument when a pre-built `ModelConfig` is passed via `config=`, causing Ollama-backed recognizers to fall back to langextract's 120s default regardless of the configured timeout. The recognizer now merges `language_model_params` into `ModelConfig.provider_kwargs`, which is the path that reaches the provider constructor. Explicit entries under `provider.kwargs:` still take precedence. Also fixed a `TypeError` when `kwargs:` or `language_model_params:` is `null` in the YAML. (#1943, Thanks @lsternlicht)

presidio-analyzer/presidio_analyzer/predefined_recognizers/generic/phone_recognizer.py

@@ -70,14 +70,14 @@ def analyze(
             ):
                 try:
                     parsed_number = phonenumbers.parse(text[match.start : match.end])
-                    region = phonenumbers.region_code_for_number(parsed_number)
-                    results += [
-                        self._get_recognizer_result(match, text, region, nlp_artifacts)
-                    ]
+                    matched_region = phonenumbers.region_code_for_number(parsed_number)
                 except NumberParseException:
-                    results += [
-                        self._get_recognizer_result(match, text, region, nlp_artifacts)
-                    ]
+                    matched_region = region
+                results += [
+                    self._get_recognizer_result(
+                        match, text, matched_region, nlp_artifacts
+                    )
+                ]
 
         return EntityRecognizer.remove_duplicates(results)
 

presidio-analyzer/tests/test_phone_recognizer.py

@@ -123,6 +123,14 @@ def test_when_phone_with_leniency_then_succeed(
          2, ["PHONE_NUMBER", "PHONE_NUMBER"],
          ((16, 30), (60, 77),), 0.4,
          ['Recognized as US region phone number, using PhoneRecognizer','Recognized as FR region phone number, using PhoneRecognizer']),
+        # The international number comes first here: it is parsed successfully and
+        # used to be assigned back to the loop variable, so the national-format
+        # number that follows was explained with the previous match's region.
+        # Every case above puts the national number first, which hides that order.
+         ("My international number is +44 1234 567890, and my US one is (415) 555-0132",
+         2, ["PHONE_NUMBER", "PHONE_NUMBER"],
+         ((27, 42), (61, 75),), 0.4,
+         ['Recognized as GB region phone number, using PhoneRecognizer','Recognized as US region phone number, using PhoneRecognizer']),
         # fmt: on
     ],
 )