fix: coerce low combined_risk_level to medium instead of rejecting

[asteier2026]

Summary

• The LLM occasionally outputs combined_risk_level='low' with a non-leave_as_is
  protection_method_suggestion (e.g. suppress_inference, generalize)
• Previously EntityDispositionSchema._validate_protection_consistency raised a ValidationError, which
  caused the entire record to be skipped (UPDATE - sometimes the record isn't skipped, but the entity silently gets dropped from the sensitivity disposition)
• Now combined_risk_level is promoted to medium — preserving the model's intent to protect the entity
  rather than suppressing it. This is more appropriate for weaker models that apply protection methods
  without fully accounting for contextual risk
• Updated the corresponding test from asserting a ValidationError to asserting that
  combined_risk_level is promoted and the original protection_method_suggestion is retained

UPDATE - the sensitivity disposition prompt is overloaded and doesn't do as well on gpt-oss-120b. We will do another PR to fix that, but above is a quick fix for now

[greptile-apps]

Greptile Summary

This PR changes how EntityDispositionSchema handles the inconsistent combination of combined_risk_level='low' with an active protection method: instead of raising a ValidationError (which could silently drop the entity), the validator now promotes combined_risk_level to 'medium' and emits a logging.warning, preserving the model's protection intent.

• rewrite.py: Imports logging, adds a module-level logger, and replaces the raise ValueError branch with a field mutation (self.combined_risk_level = CombinedRiskLevel.medium.value) plus a structured warning log. The existing high-risk / leave_as_is error path is unchanged.
• test_schemas.py: Replaces the single pytest.raises(ValidationError) case with a @pytest.mark.parametrize test over all four active protection methods (replace, generalize, remove, suppress_inference), asserting both the promoted risk level and the serialization type from model_dump().

Confidence Score: 5/5

• This PR is safe to merge. The change is contained to a single validation branch in one schema class, adds a warning log for observability, and is covered by a parametrized test across all affected protection methods.
• The coercion logic is straightforward — the field mutation uses .value correctly for use_enum_values=True, the second validation branch (high risk + leave_as_is) is untouched, and StrictEntityDispositionSchema is unaffected because it rejects low at field-parse time before the validator runs. All three previously flagged concerns from prior review rounds have been addressed.
• No files require special attention.

Important Files Changed

Filename	Overview
src/anonymizer/engine/schemas/rewrite.py	Replaces the ValueError with a warning log + field coercion in _validate_protection_consistency; assignment uses .value to stay consistent with use_enum_values=True; second validation branch (high + leave_as_is) is unaffected.
tests/engine/test_schemas.py	Old single-case ValidationError test replaced with a parametrized test covering all four active protection methods; also validates serialization type consistency via model_dump().

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[LLM output: combined_risk_level='low'\nprotection_method_suggestion ≠ leave_as_is] --> B{_validate_protection_consistency}
    B --> C[Log warning]
    C --> D[Promote combined_risk_level → 'medium'\nKeep protection_method_suggestion unchanged]
    D --> E{combined_risk_level='high'\n+ leave_as_is?}
    E -- No --> F[Return valid EntityDispositionSchema\nwith medium risk + active protection]
    E -- Yes --> G[raise ValueError]

    H[Normal input: consistent risk+method] --> B
    B --> E

Loading

_{Reviews (7): Last reviewed commit: "fix: suppress type: ignore for use_enum_..." | Re-trigger Greptile}

[lipikaramaswamy]

One non-blocking serialization consistency suggestion.

[lipikaramaswamy]

Non-blocking: because model_config uses use_enum_values=True, normally validated combined_risk_level values are stored and returned by model_dump() as plain strings. This assignment happens after validation, so the coerced path instead stores a CombinedRiskLevel enum object (although it compares equal to "medium"). Could we assign CombinedRiskLevel.medium.value here and add a model_dump() assertion so coerced and non-coerced records preserve the same serialization contract?

[asteier2026]

Made suggested changs.

[asteier2026]

Note that this branch has a different version of rewrite_generation.py than what is being updated in PR 208 - it doesn't have the prereplace logic. How do we handle that?