Merge nucleic/upbeat-yarn-seal-ekbe into dev

label_nucleic_prompts.py

@@ -1,5 +1,5 @@
 #!/usr/bin/env python3
-"""Filter and label exported Nucleic prompts with GPT-5.6 Terra via Codex."""
+"""Filter and label prompt-only sources with a configurable Codex teacher."""
 
 from __future__ import annotations
 
@@ -31,8 +31,8 @@ DEFAULT_OUTPUT = (
     / ".artifacts"
     / "nucleic-history-first-prompts.labeled.jsonl"
 )
-MODEL = "gpt-5.6-terra"
-REASONING_EFFORT = "low"
+DEFAULT_MODEL = "gpt-5.6-terra"
+DEFAULT_REASONING_EFFORT = "low"
 STATE_SCHEMA_VERSION = 1
 DEFAULT_BATCH_SIZE = 40
 DEFAULT_BATCH_CHARS = 80_000
@@ -443,6 +443,8 @@ def invoke_codex(
     *,
     codex: str,
     codex_isolation: str,
+    model: str,
+    reasoning_effort: str,
     max_prompt_chars: int,
     timeout_seconds: int,
     max_attempts: int,
@@ -487,9 +489,9 @@ def invoke_codex(
                 "--skip-git-repo-check",
                 *exec_isolation_args,
                 "--model",
-                MODEL,
+                model,
                 "--config",
-                f'model_reasoning_effort="{REASONING_EFFORT}"',
+                f'model_reasoning_effort="{reasoning_effort}"',
                 "--output-schema",
                 str(schema_path),
                 "--output-last-message",
@@ -677,7 +679,8 @@ def label(args: argparse.Namespace) -> dict[str, int]:
     print(
         f"input={len(lines)} prefiltered={len(mechanical_rejections)} "
         f"resumed={len(states) - len(new_mechanical)} pending={len(pending)} "
-        f"batches={len(batch_list)} model={MODEL}",
+        f"batches={len(batch_list)} model={args.model} "
+        f"reasoning={args.reasoning_effort}",
         flush=True,
     )
 
@@ -691,6 +694,8 @@ def label(args: argparse.Namespace) -> dict[str, int]:
             batch,
             codex=args.codex,
             codex_isolation=args.codex_isolation,
+            model=args.model,
+            reasoning_effort=args.reasoning_effort,
             max_prompt_chars=args.max_prompt_chars,
             timeout_seconds=args.timeout_seconds,
             max_attempts=args.max_attempts,
@@ -750,6 +755,8 @@ def build_parser() -> argparse.ArgumentParser:
         help="Audit JSONL for discarded lines (default: derived from --output)",
     )
     parser.add_argument("--codex", default="codex", help="Codex CLI executable")
+    parser.add_argument("--model", default=DEFAULT_MODEL)
+    parser.add_argument("--reasoning-effort", default=DEFAULT_REASONING_EFFORT)
     parser.add_argument("--batch-size", type=int, default=DEFAULT_BATCH_SIZE)
     parser.add_argument("--batch-chars", type=int, default=DEFAULT_BATCH_CHARS)
     parser.add_argument(
@@ -803,6 +810,8 @@ def main(argv: Sequence[str] | None = None) -> int:
         parser.error("--timeout-seconds must be positive")
     if args.max_attempts <= 0:
         parser.error("--max-attempts must be positive")
+    if not args.model.strip() or not args.reasoning_effort.strip():
+        parser.error("--model and --reasoning-effort must be non-empty")
 
     try:
         metrics = label(args)