feat: persist role_span_profiles.json and wire into rebuild pipeline

paperforge/worker/ocr.py

@@ -1754,6 +1754,9 @@ def postprocess_ocr_result(vault: Path, key: str, all_results: list[dict]) -> tu
     write_raw_blocks_jsonl(artifacts.blocks_raw, all_raw_blocks)
     structured = build_structured_blocks(all_raw_blocks)
     write_structured_blocks_jsonl(artifacts.blocks_structured, structured)
+    # Write role-level span profiles
+    from paperforge.worker.ocr_profiles import write_role_span_profiles
+    write_role_span_profiles(structured, artifacts.blocks_structured.parent)
 
     # --- Phase 2: metadata resolution ---
     metadata_dir = ocr_root / "metadata"

paperforge/worker/ocr_profiles.py

@@ -6,6 +6,9 @@ profile-quality scoring, span cross-validation, role-family comparison.
 
 from __future__ import annotations
 
+import json
+from pathlib import Path
+
 
 def extract_block_span_profile(block: dict) -> dict | None:
     """Extract a normalized style profile dict from a block's span_metadata.
@@ -231,3 +234,18 @@ def cross_validate_with_span(
         "suggested_roles": suggested_roles,
         "match_details": {tentative_role: current_match},
     }
+
+
+def write_role_span_profiles(
+    blocks: list[dict],
+    output_dir: str | Path,
+) -> Path:
+    """Build and write role_span_profiles.json.
+
+    Returns the path to the written file.
+    """
+    profiles = build_role_span_profiles(blocks)
+    output_path = Path(output_dir) / "role_span_profiles.json"
+    with open(output_path, "w", encoding="utf-8") as f:
+        json.dump(profiles, f, indent=2, ensure_ascii=False)
+    return output_path

paperforge/worker/ocr_rebuild.py

@@ -60,6 +60,9 @@ def run_derived_rebuild_for_keys(vault: Path, keys: list[str]) -> dict:
 
         structured = build_structured_blocks(all_raw_blocks)
         write_structured_blocks_jsonl(artifacts.blocks_structured, structured)
+        # Write role-level span profiles
+        from paperforge.worker.ocr_profiles import write_role_span_profiles
+        write_role_span_profiles(structured, artifacts.blocks_structured.parent)
 
         # Read source metadata
         source_meta = read_json(artifacts.source_metadata) if artifacts.source_metadata.exists() else {}

tests/test_ocr_blocks.py

@@ -103,3 +103,19 @@ def test_build_structured_blocks_carries_span_metadata() -> None:
     ]
     rows = build_structured_blocks(raw_blocks)
     assert rows[0].get("span_metadata") == span_data
+
+
+def test_role_span_profiles_written_to_output() -> None:
+    """Verify that role_span_profiles.json is written during rebuild."""
+    import json
+    from paperforge.worker.ocr_profiles import build_role_span_profiles
+
+    blocks = [
+        {"role": "section_heading", "span_metadata": {"size": 16.0, "flags": "bold"}},
+        {"role": "body_paragraph", "span_metadata": {"size": 10.0, "flags": 0}},
+    ]
+    profiles = build_role_span_profiles(blocks)
+    # Must be JSON-serializable
+    dumped = json.dumps(profiles)
+    assert "section_heading" in dumped
+    assert "body_paragraph" in dumped