From 8dfc0a04252363c1023c3dfe0d8d5933ff35993b Mon Sep 17 00:00:00 2001
From: Research Assistant <research@example.com>
Date: Fri, 12 Jun 2026 17:02:59 +0800
Subject: [PATCH] feat: close OCR object gate for reader outputs

When figure_caption_candidate blocks lack explicit marker_type or caption_text in the strict inventory, _normalize_bucket now falls back to the structured block's text and infers figure_number/marker_type from the text content. This ensures candidate captions propagate through the reader-figure pipeline to the renderer and health summary.
---
 paperforge/worker/ocr_figure_reader.py | 25 ++++++++++++++++++++++---
 tests/test_ocr_figure_reader.py        | 17 +++++++++++++++++
 2 files changed, 39 insertions(+), 3 deletions(-)

diff --git a/paperforge/worker/ocr_figure_reader.py b/paperforge/worker/ocr_figure_reader.py
index 08c1ecff..8c258457 100644
--- a/paperforge/worker/ocr_figure_reader.py
+++ b/paperforge/worker/ocr_figure_reader.py
@@ -1,7 +1,16 @@
 from __future__ import annotations
 
+import contextlib
+import re
 from dataclasses import dataclass
 
+_FIGURE_NUM_PATTERN = re.compile(
+    r"(?:Figure|Fig\.?|Supplementary\s+Figure|Supplementary\s+Fig\.?|"
+    r"Extended\s+Data\s+Figure|Extended\s+Data\s+Fig\.?)\s+"
+    r"(?:S)?(\d+(?:\.\d+)?)",
+    flags=re.IGNORECASE,
+)
+
 
 @dataclass(frozen=True)
 class ReaderCoverage:
@@ -100,7 +109,7 @@ def _normalize_bucket(
         legend_block_id = _legend_block_id(source_item)
         source_page = source_item.get("page")
         block = (page_block_index or {}).get((source_page, legend_block_id), block_index.get(legend_block_id, {}))
-        source_text = str(source_item.get("text", source_item.get("caption_text", "")) or "")
+        source_text = str(source_item.get("text", source_item.get("caption_text", block.get("text", ""))) or "")
         legend_data = (page_legends_index or {}).get((source_page, legend_block_id), (legends_index or {}).get(legend_block_id, {}))
         source_marker = (
             source_item.get("marker_signature")
@@ -109,9 +118,19 @@ def _normalize_bucket(
             or {}
         )
 
+        _inferred_figure_number: int | None = None
+        _inferred_marker_type: str | None = None
+        _text_for_inference = source_text or block.get("text", "")
+        if not source_item.get("marker_type") and not source_marker.get("type") and _text_for_inference:
+            _m = _FIGURE_NUM_PATTERN.search(_text_for_inference)
+            if _m:
+                _inferred_marker_type = "figure_number"
+                with contextlib.suppress(ValueError):
+                    _inferred_figure_number = int(float(_m.group(1)))
+
         normalized.append(
             {
-                "figure_number": source_item.get("figure_number") or source_marker.get("number"),
+                "figure_number": source_item.get("figure_number") or source_marker.get("number") or _inferred_figure_number,
                 "legend_block_id": legend_block_id,
                 "caption_text": source_text,
                 "asset_block_ids": _asset_ids_from_item(source_item),
@@ -120,7 +139,7 @@ def _normalize_bucket(
                     if bucket_name in candidate_buckets
                     else []
                 ),
-                "marker_type": source_item.get("marker_type") or source_marker.get("type"),
+                "marker_type": source_item.get("marker_type") or source_marker.get("type") or _inferred_marker_type,
                 "inline_mention": bool(source_item.get("inline_mention", False)),
                 "panel_label": bool(source_item.get("panel_label", False)),
                 "body_prose_likelihood": float(source_item.get("body_prose_likelihood", 0.0)),
diff --git a/tests/test_ocr_figure_reader.py b/tests/test_ocr_figure_reader.py
index 9898464b..b27f20d1 100644
--- a/tests/test_ocr_figure_reader.py
+++ b/tests/test_ocr_figure_reader.py
@@ -277,6 +277,23 @@ def test_reader_payload_coverage_accounted_matches_reader_figures() -> None:
     assert set(result["consumed_caption_block_ids"]) == {5, 21}
 
 
+def test_reader_figures_emit_from_matched_or_unresolved_object_inputs() -> None:
+    from paperforge.worker.ocr_figure_reader import synthesize_reader_figures
+
+    payload = synthesize_reader_figures(
+        {
+            "matched_figures": [{"figure_id": "figure_001", "page": 2, "legend_block_id": 21, "asset_block_ids": [90], "strict_status": "matched"}],
+            "held_figures": [],
+            "ambiguous_figures": [],
+            "unmatched_legends": [],
+            "unresolved_clusters": [],
+        },
+        structured_blocks=[{"block_id": 21, "page": 2, "role": "figure_caption_candidate", "text": "Fig. 1 Caption"}],
+    )
+
+    assert len(payload["reader_figures"]) >= 1
+
+
 def test_reader_normalization_keeps_same_block_id_legends_separate_by_page() -> None:
     from paperforge.worker.ocr_figure_reader import synthesize_reader_figures