Let callers set the critic's request parameters

VLMQualityJudge built its chat-completions request with fixed parameters, so the
reasoning and sampling settings used for the critic could not be configured: callers
reproducing a specific evaluation setup had no way to pin e.g. reasoning_effort.

Add an extra_body passthrough (default None, so the request is unchanged unless a
caller sets it) and raise the critic's max_tokens default to 16384. With reasoning
models the thinking tokens come out of the same budget (measured up to ~6.7k tokens
for gemini-3.1-pro at reasoning_effort=high), which leaves little headroom under 8192
for the analysis JSON itself.

agentic_upsampling/clients.py

@@ -477,8 +477,9 @@ class VLMQualityJudge:
         api_token: str,
         endpoint_url: str = DEFAULT_CRITIC_ENDPOINT_URL,
         model: str = DEFAULT_CRITIC_MODEL,
-        max_tokens: int = 8192,
+        max_tokens: int = 16384,
         image_jpeg_quality: int | None = DEFAULT_JPEG_QUALITY,
+        extra_body: dict[str, Any] | None = None,
     ) -> None:
         self.chat_client = OpenAIChatClient(
             ChatClientConfig(
@@ -487,6 +488,7 @@ class VLMQualityJudge:
                 api_token=api_token,
                 max_tokens=max_tokens,
                 max_retries=3,
+                extra_body=extra_body,
             )
         )
         self.image_jpeg_quality = image_jpeg_quality

tests/test_agentic_upsampling.py

@@ -10,7 +10,7 @@ from typing import Any
 
 from PIL import Image
 
-from agentic_upsampling.clients import ImageGenerationClient, PromptRewriterClient
+from agentic_upsampling.clients import ImageGenerationClient, PromptRewriterClient, VLMQualityJudge
 from agentic_upsampling.constants import (
     DEFAULT_CRITIC_ENDPOINT_URL,
     DEFAULT_CRITIC_MODEL,
@@ -521,3 +521,14 @@ def test_extract_best_images_copies_images_and_writes_manifests(tmp_path: Path)
     assert (tmp_path / "export" / "best_generations.jsonl").exists()
     assert (tmp_path / "export" / "best_generations.csv").exists()
 
+
+def test_vlm_quality_judge_sends_no_extra_body_by_default() -> None:
+    judge = VLMQualityJudge(api_token="token")
+
+    assert judge.chat_client.config.extra_body is None
+
+
+def test_vlm_quality_judge_forwards_extra_body() -> None:
+    judge = VLMQualityJudge(api_token="token", extra_body={"reasoning_effort": "high"})
+
+    assert judge.chat_client.config.extra_body == {"reasoning_effort": "high"}