Prepare four isolated bilingual minimal prompt benchmarks

tests/test_minimal_prompt_test.py

@@ -0,0 +1,55 @@
+import copy
+import pytest
+from viano_bug_analysis.llm.config import ProviderConfig
+from viano_bug_analysis.llm.errors import ProviderUnavailable
+from viano_bug_analysis.minimal_prompt_test import make_plan,measure
+
+
+def provider():
+    return ProviderConfig(id='fixture',model='fixture-model',base_url='https://example.invalid/v1beta',
+        adapter='gemini',privacy='cloud',reasoning_effort='low',max_output_tokens=8192)
+
+
+def test_four_requests_differ_only_in_language_and_native_format():
+    p=provider();plan=make_plan(p)
+    assert [c['id'] for c in plan['calls']]==['de_plain','de_json','en_plain','en_json']
+    for plain,structured in (plan['calls'][:2],plan['calls'][2:]):
+        assert plain['payload']['contents']==structured['payload']['contents']
+        assert plain['prompt_characters']<500
+        assert set(plain['payload'])=={'contents','generationConfig'}
+        generation=copy.deepcopy(structured['payload']['generationConfig'])
+        generation.pop('responseMimeType');generation.pop('responseJsonSchema')
+        assert generation==plain['payload']['generationConfig']
+        assert generation=={'maxOutputTokens':8192,'thinkingConfig':{'thinkingLevel':'low'}}
+
+
+@pytest.mark.parametrize('index',[0,1,2,3])
+def test_usage_reports_candidates_and_thoughts_separately(index):
+    case=make_plan(provider())['calls'][index]
+    def post(url,payload,headers,limit,deadline):
+        assert payload==case['payload']
+        return {'modelVersion':'fixture','candidates':[{'finishReason':'STOP','content':{'parts':[
+            {'text':'hidden','thought':True},{'text':'{"step":"lookup","purpose":"test"}' if index%2 else 'Look up spawns.'}]}}],
+            'usageMetadata':{'promptTokenCount':123,'candidatesTokenCount':12,'thoughtsTokenCount':30,'totalTokenCount':165}}
+    result=measure(provider(),case,10000,post)
+    assert result['status']=='SUCCESS'
+    assert result['usage']['input_tokens']==123
+    assert result['usage']['candidate_output_tokens']==12
+    assert result['usage']['reasoning_tokens']==30
+    assert result['usage']['cached_input_tokens'] is None
+    assert 'hidden' not in result['answer']
+
+
+def test_no_retries_and_no_provider_body_leak():
+    calls=[]
+    def post(*args):
+        calls.append(1);raise ProviderUnavailable('LLM_HTTP_TEMPORARY',http_status=503)
+    result=measure(provider(),make_plan(provider())['calls'][0],1000,post)
+    assert len(calls)==1 and result['status']=='ERROR' and result['http_status']==503
+
+
+def test_incomplete_retains_usage():
+    def post(*args):return {'candidates':[{'finishReason':'MAX_TOKENS','content':{'parts':[]}}],
+        'usageMetadata':{'promptTokenCount':99,'thoughtsTokenCount':100}}
+    result=measure(provider(),make_plan(provider())['calls'][0],1000,post)
+    assert result['status']=='INCOMPLETE' and result['usage']['input_tokens']==99