Scope entity resolution reasoning effort independently

config/llm.example.toml

@@ -19,6 +19,8 @@ token_parameter = "max_tokens"
 max_output_tokens = 4096
 # Optional; omit for provider defaults. Supported values depend on the model.
 # reasoning_effort = "medium"
+# Optional override for entity resolution planning and assessment only:
+# entity_resolution_reasoning_effort = "low"
 
 [[llm.providers]]
 id = "own-server"

docs/llm-context-efficiency-v1.md

@@ -40,3 +40,24 @@ The output limit is a ceiling, not a request to generate that many tokens. Think
 and output accounting are provider-dependent. More output capacity can increase
 cost; monitor actual usage and finish diagnostics. No raw reasoning is requested
 or persisted by these settings.
+
+
+## Identity-only reasoning override
+
+Each provider optionally accepts `entity_resolution_reasoning_effort` with the
+same validated values as `reasoning_effort`. It applies only to versioned
+`entity_resolution_plan_vN` and `entity_resolution_assessment_vN` requests.
+All interpretation, hypothesis and investigation requests retain the provider's
+normal `reasoning_effort`. Omission preserves previous configuration snapshots.
+
+```toml
+reasoning_effort = "medium"
+entity_resolution_reasoning_effort = "low"
+max_output_tokens = 16384
+```
+
+The override is recorded in configuration snapshots and their hashes. Existing
+pending calls cannot silently change configuration mid-retry. A new entity
+assessment attempt uses the new settings; accepted plan and previous assessment
+artifacts remain unchanged and are checked without repeating their API calls.
+No attempt counters or investigation budgets are reset.

src/viano_bug_analysis/llm/config.py

@@ -33,6 +33,7 @@ class ProviderConfig:
     output_cost_per_million: float | None = None
     cost_currency: str = "USD"
     reasoning_effort: str | None = None
+    entity_resolution_reasoning_effort: str | None = None
 
     def __post_init__(self):
         if not isinstance(self.id, str) or not re.fullmatch(r"[A-Za-z0-9_-]{1,64}", self.id):
@@ -65,8 +66,9 @@ class ProviderConfig:
         for value in (self.enabled, self.allow_insecure_http):
             if not isinstance(value, bool):
                 raise LLMError("LLM_CONFIG")
-        if self.reasoning_effort is not None and (not isinstance(self.reasoning_effort, str) or self.reasoning_effort not in {"none", "minimal", "low", "medium", "high"}):
-            raise LLMError("LLM_CONFIG")
+        for effort in (self.reasoning_effort, self.entity_resolution_reasoning_effort):
+            if effort is not None and (not isinstance(effort, str) or effort not in {"none", "minimal", "low", "medium", "high"}):
+                raise LLMError("LLM_CONFIG")
         positive(self.timeout_seconds)
         positive(self.max_output_tokens, integer=True)
         if not isinstance(self.cost_currency, str) or not re.fullmatch(r"[A-Z]{3}", self.cost_currency):
@@ -115,6 +117,7 @@ class GatewayConfig:
         # Only configuration references; never resolve credentials for persistence.
         result = asdict(self)
         for provider in result["providers"]:
-            if provider["reasoning_effort"] is None:
-                del provider["reasoning_effort"]
+            for name in ("reasoning_effort", "entity_resolution_reasoning_effort"):
+                if provider[name] is None:
+                    del provider[name]
         return result

src/viano_bug_analysis/llm/gateway.py

@@ -3,6 +3,8 @@ import copy
 import logging
 import time
 import math
+import re
+from dataclasses import replace
 
 from jsonschema import Draft202012Validator
 from jsonschema.exceptions import SchemaError, ValidationError
@@ -73,6 +75,9 @@ class LLMGateway:
         for provider in self.config.providers:
             if not provider.enabled or provider.privacy not in (PRIVACY[request["privacy"]] & PRIVACY[self.config.privacy_profile]):
                 continue
+            if (provider.entity_resolution_reasoning_effort is not None
+                    and re.fullmatch(r"entity_resolution_(plan|assessment)_v[1-9][0-9]*", request["prompt_version"])):
+                provider = replace(provider, reasoning_effort=provider.entity_resolution_reasoning_effort)
             started = time.monotonic()
             response = None
             try:

tests/test_entity_resolution.py

@@ -367,3 +367,39 @@ def test_missing_capture_feedback_resumes_and_replays_without_repeating_research
     assert PackageValidator().validate(result.package).valid
     assert run(store, model).reused
     assert len(model.requests) == 4
+
+
+def test_third_assessment_can_use_lower_reasoning_without_resetting_history(built_case):
+    from dataclasses import replace
+    from viano_bug_analysis.llm import LLMError, LLMGateway
+    class TruncatedModel(Model):
+        failures = 2
+        efforts = []
+        async def generate(self, config, request, *, max_response_bytes):
+            self.efforts.append((request['prompt_version'], config.reasoning_effort))
+            if request['prompt_version']=='entity_resolution_assessment_v3' and self.failures:
+                self.failures -= 1
+                raise LLMError('LLM_INCOMPLETE_OUTPUT')
+            return await super().generate(config, request, max_response_bytes=max_response_bytes)
+    store = built_case[0]
+    model = TruncatedModel()
+    service = service_for(store)
+    old = gateway(model)
+    old.config = replace(old.config, providers=tuple(replace(p, reasoning_effort='medium') for p in old.config.providers))
+    owner = CaseInterpreter(store, old, project_research=service)
+    for _ in range(2):
+        result = asyncio.run(owner.interpret('case-A','stable-package'))
+        assert result.status == 'REJECTED'
+    adjusted = replace(old.config, providers=tuple(replace(p, entity_resolution_reasoning_effort='low') for p in old.config.providers))
+    owner = CaseInterpreter(store, LLMGateway(adjusted, {'fixture': model}), project_research=service)
+    result = asyncio.run(owner.interpret('case-A','stable-package'))
+    assert result.status == 'COMPLETED'
+    assert len(service.database.calls) == 1
+    assert model.efforts == [('entity_resolution_plan_v3','medium'),
+        ('entity_resolution_assessment_v3','medium'),('entity_resolution_assessment_v3','medium'),
+        ('entity_resolution_assessment_v3','low'),('case_interpretation_resolved_v3','medium')]
+    with PackageReader(result.package) as reader:
+        saved = json.loads(reader.read_bytes('analysis/interpretations/0001/entity-resolution/assessment.json'))
+        assert saved['payload']['attempt'] == 3
+    assert asyncio.run(owner.interpret('case-A','stable-package')).reused
+    assert len(model.efforts) == 5

tests/test_llm_context_efficiency.py

@@ -50,3 +50,37 @@ def test_identity_projection_keeps_readable_evidence_and_legacy_requests(stage):
         assert value == context["sources"][key]
     assert len(new["messages"][1]["content"]) < len(old["messages"][1]["content"]) / 2
     assert data["output_schema"] == new["output_schema"]
+
+
+@pytest.mark.parametrize("version,expected", [
+    ("entity_resolution_plan_v1", "low"),
+    ("entity_resolution_assessment_v3", "low"),
+    ("case_interpretation_resolved_v3", "medium"),
+    ("hypothesis_generation_v1", "medium"),
+    ("investigation_v1", "medium"),
+    ("entity_resolution_assessment_v3_suffix", "medium"),
+])
+def test_identity_override_is_scoped_at_http_transport(http_endpoint, version, expected):
+    url, seen, _ = http_endpoint
+    provider = ProviderConfig(id="test", model="fixture", base_url=url,
+        reasoning_effort="medium", entity_resolution_reasoning_effort="low", max_output_tokens=16384)
+    config = GatewayConfig((provider,))
+    query = request()
+    query['prompt_version'] = version
+    asyncio.run(LLMGateway(config).complete_round(query))
+    assert seen[0]['body']['reasoning_effort'] == expected
+    assert seen[0]['body'][provider.token_parameter] == 16384
+    assert provider.reasoning_effort == "medium"
+    assert config.snapshot()['providers'][0]['entity_resolution_reasoning_effort'] == "low"
+
+
+@pytest.mark.parametrize("effort", [True, 8192, [], "unlimited"])
+def test_invalid_identity_reasoning_override(effort):
+    with pytest.raises(LLMError, match="LLM_CONFIG"):
+        ProviderConfig(id="test", model="fixture", base_url="http://localhost",
+            entity_resolution_reasoning_effort=effort)
+
+
+def test_omitted_identity_override_preserves_previous_snapshot():
+    provider = ProviderConfig(id="test", model="fixture", base_url="http://localhost", reasoning_effort="medium")
+    assert 'entity_resolution_reasoning_effort' not in GatewayConfig((provider,)).snapshot()['providers'][0]