Enhancements to LLM concurrency to improve overall throughput

tests/test_new_config.py

@@ -7,6 +7,8 @@ from audita.core.config import (
     DEFAULT_GRAMMAR_CONFIDENCE_THRESHOLD,
     DEFAULT_HOMOPHONES_CONFIDENCE_THRESHOLD,
     DEFAULT_LLM_CONCURRENCY,
+    DEFAULT_MAX_SECTION_TOKENS,
+    DEFAULT_MIN_SECTION_TOKENS,
     DEFAULT_NORMALIZE_MAX_SEGMENT_GAP,
     DEFAULT_SPOKEN_WORD_CONFIDENCE_THRESHOLD,
     DEFAULT_WORK_DIR_RETENTION,
@@ -20,6 +22,8 @@ def test_default_config_allows_missing_api_key():
 
     assert config.api_key is None
     assert config.llm_concurrency == DEFAULT_LLM_CONCURRENCY
+    assert config.max_section_tokens == DEFAULT_MAX_SECTION_TOKENS
+    assert config.min_section_tokens == DEFAULT_MIN_SECTION_TOKENS
     assert config.module_keys == DEFAULT_MODULE_KEYS
     assert config.glossary_confidence_threshold == DEFAULT_GLOSSARY_CONFIDENCE_THRESHOLD
     assert config.grammar_confidence_threshold == DEFAULT_GRAMMAR_CONFIDENCE_THRESHOLD
@@ -32,12 +36,21 @@ def test_default_config_allows_missing_api_key():
 def test_cli_overrides_take_precedence():
     config = AuditaConfig.from_sources(
         env={"AUDITA_MAX_SECTION_TOKENS": "1000"},
-        overrides=ConfigOverrides(max_section_tokens=2000),
+        overrides=ConfigOverrides(max_section_tokens=2000, min_section_tokens=1000),
     )
 
     assert config.max_section_tokens == 2000
 
 
+def test_min_section_tokens_cli_override_takes_precedence():
+    config = AuditaConfig.from_sources(
+        env={"AUDITA_MIN_SECTION_TOKENS": "2000"},
+        overrides=ConfigOverrides(min_section_tokens=6000),
+    )
+
+    assert config.min_section_tokens == 6000
+
+
 def test_llm_concurrency_cli_override_takes_precedence():
     config = AuditaConfig.from_sources(
         env={"AUDITA_LLM_CONCURRENCY": "2"},
@@ -53,6 +66,12 @@ def test_llm_concurrency_env_is_parsed():
     assert config.llm_concurrency == 3
 
 
+def test_min_section_tokens_env_is_parsed():
+    config = AuditaConfig.from_sources(env={"AUDITA_MIN_SECTION_TOKENS": "3000"})
+
+    assert config.min_section_tokens == 3000
+
+
 def test_generic_llm_api_key_env_is_read():
     config = AuditaConfig.from_sources(env={"AUDITA_LLM_API_KEY": "generic-key"})
 
@@ -164,3 +183,19 @@ def test_invalid_thresholds_are_rejected(env_name):
 def test_invalid_llm_concurrency_is_rejected(value):
     with pytest.raises(AuditaConfigError, match="AUDITA_LLM_CONCURRENCY"):
         AuditaConfig.from_sources(env={"AUDITA_LLM_CONCURRENCY": value})
+
+
+@pytest.mark.parametrize("value", ["0", "-1", "many"])
+def test_invalid_min_section_tokens_is_rejected(value):
+    with pytest.raises(AuditaConfigError, match="AUDITA_MIN_SECTION_TOKENS"):
+        AuditaConfig.from_sources(env={"AUDITA_MIN_SECTION_TOKENS": value})
+
+
+def test_min_section_tokens_must_not_exceed_max_section_tokens():
+    with pytest.raises(AuditaConfigError, match="AUDITA_MIN_SECTION_TOKENS"):
+        AuditaConfig.from_sources(
+            env={
+                "AUDITA_MIN_SECTION_TOKENS": "9000",
+                "AUDITA_MAX_SECTION_TOKENS": "8000",
+            }
+        )