Fix trace reuse and packaged model assets

trace_analyzer/cli.py

@@ -14,6 +14,9 @@ from .reporting import write_reports
 from .resume_advanced import collect_existing_detail_paths, run_advanced_from_existing
 from .study import parse_input_length_bucket_thresholds, run_study
 
+ANALYSIS_PROVENANCE_FILENAME = "analysis_provenance.json"
+ANALYSIS_PROVENANCE_SCHEMA_VERSION = 1
+
 
 def build_parser():
     parser = argparse.ArgumentParser(description="Analyze coding-agent trace patterns.")
@@ -282,6 +285,70 @@ def _existing_detail_outputs(output_dir):
     return collect_existing_detail_paths(output_dir)
 
 
+def _file_fingerprint(path: str | Path) -> dict:
+    resolved = Path(path).resolve()
+    stat = resolved.stat()
+    return {
+        "path": str(resolved),
+        "size": stat.st_size,
+        "mtime_ns": stat.st_mtime_ns,
+    }
+
+
+def _path_option(value: str | None) -> str | None:
+    if value is None:
+        return None
+    path = Path(value)
+    return str(path.resolve()) if path.exists() else str(value)
+
+
+def _expected_analysis_provenance(args, release_input_path: Path, input_length_bucket_thresholds: list[int]) -> dict:
+    return {
+        "schema_version": ANALYSIS_PROVENANCE_SCHEMA_VERSION,
+        "raw_input": _file_fingerprint(args.input),
+        "release_input": _file_fingerprint(release_input_path),
+        "options": {
+            "block_size": args.block_size,
+            "segment_mode": args.segment_mode,
+            "tokenizer_path": _path_option(args.tokenizer_path),
+            "tokenizer_batch_size": args.tokenizer_batch_size,
+            "model_family": args.model_family,
+            "model_meta_dir": _path_option(args.model_meta_dir),
+            "input_length_buckets": input_length_bucket_thresholds,
+        },
+    }
+
+
+def _analysis_provenance_path(output_dir: Path) -> Path:
+    return output_dir / ANALYSIS_PROVENANCE_FILENAME
+
+
+def _load_analysis_provenance(output_dir: Path) -> dict | None:
+    path = _analysis_provenance_path(output_dir)
+    if not path.exists():
+        return None
+    try:
+        with path.open("r", encoding="utf-8") as handle:
+            data = json.load(handle)
+    except (OSError, json.JSONDecodeError):
+        return None
+    return data if isinstance(data, dict) else None
+
+
+def _analysis_provenance_matches(output_dir: Path, expected: dict) -> bool:
+    return _load_analysis_provenance(output_dir) == expected
+
+
+def _write_analysis_provenance(output_dir: Path, provenance: dict) -> None:
+    output_dir.mkdir(parents=True, exist_ok=True)
+    destination = _analysis_provenance_path(output_dir)
+    temp_path = destination.with_suffix(destination.suffix + ".tmp")
+    with temp_path.open("w", encoding="utf-8") as handle:
+        json.dump(provenance, handle, ensure_ascii=False, indent=2, sort_keys=True)
+        handle.write("\n")
+    temp_path.replace(destination)
+
+
 def _stage_message(progress, step: int, total_steps: int, message: str) -> None:
     tqdm.write(f"Stage {step}/{total_steps}: {message}")
     progress.update(1)
@@ -301,6 +368,8 @@ def main(argv=None):
                 f"Release trace not found for raw trace {args.input}. "
                 "Run `python -m trace_formatter build-release <raw-trace>` first, or pass --release-input."
             )
+        expected_provenance = _expected_analysis_provenance(args, release_input_path, input_length_bucket_thresholds)
+        reuse_allowed = _analysis_provenance_matches(output_dir, expected_provenance)
         total_steps = 4
         progress = tqdm(
             total=total_steps,
@@ -310,7 +379,7 @@ def main(argv=None):
         )
         try:
             prepare_result = None
-            reusable_base = _existing_base_outputs(output_dir)
+            reusable_base = _existing_base_outputs(output_dir) if reuse_allowed else None
             if reusable_base:
                 _stage_message(progress, 1, total_steps, "reuse existing features.csv")
                 prepare_result = {
@@ -320,7 +389,7 @@ def main(argv=None):
             else:
                 _stage_message(progress, 1, total_steps, "prepare features.csv")
                 prepare_result = stream_prepare(args.input, output_dir, show_progress=True)
-            reusable_details = _existing_detail_outputs(output_dir)
+            reusable_details = _existing_detail_outputs(output_dir) if reuse_allowed else None
             if reusable_details:
                 _stage_message(progress, 2, total_steps, "reuse existing details/")
                 advanced_paths = reusable_details
@@ -362,6 +431,7 @@ def main(argv=None):
                 dataset_title=dataset_name,
                 show_progress=True,
             )
+            _write_analysis_provenance(output_dir, expected_provenance)
         finally:
             progress.close()
         print(