perf: reuse extract's file list when writing the traces README

src/teich/cli.py

@@ -373,6 +373,7 @@ def _write_extract_readme(
     *,
     model_filter: str | None = None,
     repo_id: str | None = None,
+    trace_files: list[Path] | None = None,
 ) -> Path:
     cfg = _extract_dataset_config(provider, output, model_filter=model_filter, repo_id=repo_id)
     return write_traces_readme(
@@ -382,6 +383,7 @@ def _write_extract_readme(
         model_id=None,
         repo_id=cfg.get_publish_repo_id(),
         extraction_provider=provider,
+        trace_files=trace_files,
     )
 
 
@@ -465,7 +467,7 @@ def _run_extract_command(
     if stale_readme_path.exists() and stale_readme_path.is_file():
         stale_readme_path.unlink()
     anonymize_report = None if skip_anonymize else anonymize_path(output, output, in_place=True)
-    readme_path = _write_extract_readme(provider, output, model_filter=model_filter)
+    readme_path = _write_extract_readme(provider, output, model_filter=model_filter, trace_files=result.copied_files)
     extracted_message = f"Extracted {result.count} {provider} trace{'s' if result.count != 1 else ''}"
     if model_filter:
         extracted_message += f" with {model_filter}"

src/teich/studio/extraction.py

@@ -152,7 +152,7 @@ def emit_progress(event: dict[str, Any]) -> None:
                         "totals": self.anonymize_totals,
                     }
                 )
-            self._write_readme()
+            self._write_readme(result.copied_files)
             self._emit_status(
                 "completed",
                 f"Extracted {self.result_rows} trace row(s) across {len(self.result_files)} file(s).",
@@ -164,7 +164,7 @@ def emit_progress(event: dict[str, Any]) -> None:
             self.finished_at = datetime.now(timezone.utc)
             self.events.close()
 
-    def _write_readme(self) -> None:
+    def _write_readme(self, trace_files: list[Path] | None = None) -> None:
         cfg = _extract_dataset_config(self.provider, self.output_dir, model_filter=self.model_filter)
         readme_path = write_traces_readme(
             cfg.output.traces_dir,
@@ -173,6 +173,7 @@ def _write_readme(self) -> None:
             model_id=None,
             repo_id=cfg.get_publish_repo_id(),
             extraction_provider=self.provider,
+            trace_files=trace_files,
         )
         self.events.append({"kind": "extract_readme", "text": f"Wrote {readme_path.name}.", "path": str(readme_path)})
 

src/teich/trace_readme.py

@@ -523,10 +523,14 @@ def write_traces_readme(
     tools: list[dict[str, Any]] | None = None,
     excluded_dirs: list[Path] | None = None,
     extraction_provider: str | None = None,
+    trace_files: list[Path] | None = None,
 ) -> Path:
+    # Callers that just wrote the traces can pass the file list to skip the
+    # directory rescan; the exclusion filters still apply either way.
+    candidates = trace_files if trace_files is not None else traces_dir.rglob("*.jsonl")
     trace_files = sorted(
         path
-        for path in traces_dir.rglob("*.jsonl")
+        for path in candidates
         if path.is_file()
         and not {"partials", "failures"}.intersection(path.relative_to(traces_dir).parts)
         and not any(_path_is_relative_to(path, excluded_dir) for excluded_dir in excluded_dirs or [])