Refactor judge module into judge/ package

Split monolithic src/judge.py into a proper package structure:
- src/judge/batch_providers.py: BatchProvider ABC, AnthropicBatchProvider, XAIBatchProvider
- src/judge/judges.py: RegexChecker, BlackboxChecker, GlassboxChecker
- src/judge/helpers.py: extract_* functions, parse_json_verdict
- src/judge/prompts.py: BLACKBOX_PROMPT, GLASSBOX_PROMPT, constants
- src/judge/judge.py: Judge orchestrator class
- src/judge/__init__.py: Re-exports all public APIs

1 files changed, 47 insertions, 0 deletions

diff --git a/src/judge/__init__.py b/src/judge/__init__.py
new file mode 100644
index 0000000..02f77cb
--- /dev/null
+++ b/src/judge/__init__.py
@@ -0,0 +1,47 @@
+"""
+Judge package - evaluates experiment logs for deceptive behaviour.
+"""
+
+from .batch_providers import (
+    AnthropicBatchProvider,
+    BatchProvider,
+    BatchRequest,
+    BatchResult,
+    XAIBatchProvider,
+)
+from .helpers import (
+    extract_model_output,
+    extract_reasoning_trace,
+    extract_system_prompt,
+    parse_json_verdict,
+)
+from .judge import Judge
+from .judges import BlackboxChecker, GlassboxChecker, RegexChecker
+from .prompts import (
+    BLACKBOX_CATEGORIES,
+    BLACKBOX_PROMPT,
+    GLASSBOX_CATEGORIES,
+    GLASSBOX_PROMPT,
+    SOPHISTICATION_TIERS,
+)
+
+__all__ = [
+    "AnthropicBatchProvider",
+    "BatchProvider",
+    "BatchRequest",
+    "BatchResult",
+    "BlackboxChecker",
+    "GlassboxChecker",
+    "Judge",
+    "RegexChecker",
+    "XAIBatchProvider",
+    "BLACKBOX_CATEGORIES",
+    "BLACKBOX_PROMPT",
+    "GLASSBOX_CATEGORIES",
+    "GLASSBOX_PROMPT",
+    "SOPHISTICATION_TIERS",
+    "extract_model_output",
+    "extract_reasoning_trace",
+    "extract_system_prompt",
+    "parse_json_verdict",
+]