chore: lower deriver custom instruction cap

README.md

@@ -428,7 +428,7 @@ Examples:
 - `DB_CONNECTION_URI` - Database connection string
 - `AUTH_JWT_SECRET` - JWT secret key
 - `DERIVER_MODEL_CONFIG__TRANSPORT` - Transport for the background deriver
-- `DERIVER_MAX_CUSTOM_INSTRUCTIONS_TOKENS` - Explicit prompt budget cap for deriver custom instructions
+- `DERIVER_MAX_CUSTOM_INSTRUCTIONS_TOKENS` - Explicit prompt budget cap for deriver custom instructions (maximum supported value: `500`)
 - `SUMMARY_MODEL_CONFIG__MODEL` - Summary model override
 - `DIALECTIC_LEVELS__low__MODEL_CONFIG__MODEL` - Model for low reasoning level
 - `LOG_LEVEL` - Application log level

config.toml.example

@@ -86,7 +86,7 @@ STALE_SESSION_TIMEOUT_MINUTES = 5
 DEDUPLICATE = true
 LOG_OBSERVATIONS = false
 MAX_INPUT_TOKENS = 23000
-MAX_CUSTOM_INSTRUCTIONS_TOKENS = 1024  # Required for non-blank reasoning.custom_instructions; over-limit values fail validation
+MAX_CUSTOM_INSTRUCTIONS_TOKENS = 500  # Required for non-blank reasoning.custom_instructions; max supported value is 500
 WORKING_REPRESENTATION_MAX_OBSERVATIONS = 100
 REPRESENTATION_BATCH_MAX_TOKENS = 1024
 FLUSH_ENABLED = false  # Bypass batch token threshold, process work immediately

docs/v3/contributing/configuration.mdx

@@ -532,7 +532,7 @@ DEFAULT_TTL_SECONDS = 300
 [deriver]
 ENABLED = true
 WORKERS = 1
-MAX_CUSTOM_INSTRUCTIONS_TOKENS = 1024
+MAX_CUSTOM_INSTRUCTIONS_TOKENS = 500
 
 [deriver.model_config]
 transport = "openai"

src/config.py

@@ -733,7 +733,7 @@ class DeriverSettings(HonchoSettings):
 
     MAX_INPUT_TOKENS: Annotated[int, Field(default=23000, gt=0, le=23000)] = 23000
     MAX_CUSTOM_INSTRUCTIONS_TOKENS: Annotated[
-        int | None, Field(default=None, gt=0, le=23000)
+        int | None, Field(default=None, gt=0, le=500)
     ] = None
 
     # Maximum number of observations to return in working representation

tests/test_config.py

@@ -7,6 +7,7 @@ def _make_deriver_settings(
     *,
     MAX_INPUT_TOKENS: int = 23000,
     MAX_CUSTOM_INSTRUCTIONS_TOKENS: int | None = None,
+    REPRESENTATION_BATCH_MAX_TOKENS: int = 1024,
 ) -> DeriverSettings:
     return DeriverSettings(
         MODEL_CONFIG=ConfiguredModelSettings(
@@ -15,6 +16,7 @@ def _make_deriver_settings(
         ),
         MAX_INPUT_TOKENS=MAX_INPUT_TOKENS,
         MAX_CUSTOM_INSTRUCTIONS_TOKENS=MAX_CUSTOM_INSTRUCTIONS_TOKENS,
+        REPRESENTATION_BATCH_MAX_TOKENS=REPRESENTATION_BATCH_MAX_TOKENS,
     )
 
 
@@ -29,9 +31,9 @@ def test_effective_custom_instructions_tokens_requires_explicit_limit() -> None:
 
 
 def test_effective_custom_instructions_tokens_uses_explicit_limit() -> None:
-    settings = _make_deriver_settings(MAX_CUSTOM_INSTRUCTIONS_TOKENS=2048)
+    settings = _make_deriver_settings(MAX_CUSTOM_INSTRUCTIONS_TOKENS=500)
 
-    assert settings.effective_max_custom_instructions_tokens == 2048
+    assert settings.effective_max_custom_instructions_tokens == 500
 
 
 def test_custom_instructions_tokens_cannot_exceed_input_budget() -> None:
@@ -40,6 +42,12 @@ def test_custom_instructions_tokens_cannot_exceed_input_budget() -> None:
         match=r"MAX_CUSTOM_INSTRUCTIONS_TOKENS.*cannot exceed max deriver input tokens",
     ):
         _make_deriver_settings(
-            MAX_INPUT_TOKENS=1024,
-            MAX_CUSTOM_INSTRUCTIONS_TOKENS=2048,
+            MAX_INPUT_TOKENS=400,
+            MAX_CUSTOM_INSTRUCTIONS_TOKENS=500,
+            REPRESENTATION_BATCH_MAX_TOKENS=128,
         )
+
+
+def test_custom_instructions_tokens_cannot_exceed_supported_cap() -> None:
+    with pytest.raises(ValueError, match="less than or equal to 500"):
+        _make_deriver_settings(MAX_CUSTOM_INSTRUCTIONS_TOKENS=501)