Add scanned math pages exercising formula extraction

ATTRIBUTIONS.md

@@ -190,3 +190,19 @@ Vendored WordPerfect-family test files. Per-file provenance (upstream path, revi
 - **License:** CC-BY-SA-4.0. ShareAlike is copyleft, so these are reference only.
 - **Used here:** 178 mathematics articles, reference.
 - **Modifications:** none.
+
+## Synthetic scanned math pages
+
+- **Citation:** none; generated for this corpus.
+- **Source:** rendered with matplotlib mathtext, rasterized at 300 DPI, binarized, and packed as CCITT G4 PDFs with no text layer, to mirror the shape of a scanned textbook page (xberg issue #1385).
+- **License:** MIT, as our own work.
+- **Used here:** `pdf_scanned/synthetic_math_*.pdf`, four pages that carry display equations only as pixels. They exercise formula detection and recognition on scanned input, including a skewed scan.
+- **Modifications:** not applicable.
+
+## xberg benchmark-harness fixtures
+
+- **Citation:** none; part of the xberg repository.
+- **Source:** <https://github.com/xberg-io/xberg>, `tools/benchmark-harness/fixtures/split/memo_marketing_form.pdf`, page 1, rasterized at 300 DPI and packed as a bilevel CCITT G4 scan.
+- **License:** MIT, as our own work.
+- **Used here:** `pdf_scanned/memo_prose_scanned.pdf`, the negative case: a scanned page with no mathematics, which must yield no formulas.
+- **Modifications:** rasterized and binarized from the vector original.

MATH_PROVENANCE.md

@@ -160,7 +160,12 @@ That is real formula-level ground truth, independent of any extractor. `README.m
 | `xml/math/journal_pcbi_1005589_type_manuscript_58c02e.xml` | [PLOS Computational Biology](https://journals.plos.org/ploscompbiol/article/file?id=10.1371/journal.pcbi.1005589&type=manuscript) | CC-BY-4.0 | MathML inside JATS <alternatives>, JATS Journa |
 | `xml/math/manuscript_933c4c.xml` | [Texture](https://raw.githubusercontent.com/substance/texture/master/data/kitchen-sink/manuscript.xml) | MIT | LaTeX in <tex-math> CDATA, JATS disp-formula / |
 
-Committed: 136. The binary formats among them live in the bucket and are pinned by `corpus.lock.json`, following the storage split in `README.md`.
+| `pdf_scanned/synthetic_math_interpolation_scanned.pdf` | Generated for this corpus (matplotlib mathtext, 300 DPI bilevel CCITT G4, no text layer) | MIT | Rasterized display equations (interpolation) |
+| `pdf_scanned/synthetic_math_interpolation_skew_scanned.pdf` | Generated for this corpus; same page rotated 1.2 degrees before binarization | MIT | Rasterized display equations, skewed scan |
+| `pdf_scanned/synthetic_math_matrix_scanned.pdf` | Generated for this corpus (matplotlib mathtext, 300 DPI bilevel CCITT G4, no text layer) | MIT | Rasterized display equations (determinants) |
+| `pdf_scanned/synthetic_math_series_scanned.pdf` | Generated for this corpus (matplotlib mathtext, 300 DPI bilevel CCITT G4, no text layer) | MIT | Rasterized display equations (series) |
+
+Committed: 140. The binary formats among them live in the bucket and are pinned by `corpus.lock.json`, following the storage split in `README.md`.
 
 ## Reference only, not committed
 

corpus.lock.json

@@ -1625,6 +1625,10 @@
       "sha256": "0350f7e061a91df9c11548f061ac308360f0a1a0e8389c74e1b76c1bb70d5184",
       "size": 126609
     },
+    "pdf_scanned/memo_prose_scanned.pdf": {
+      "sha256": "9f75e84ec7e322b71715a626f06df1ce1ebe5bd7d921e6b046807d18a981b484",
+      "size": 42912
+    },
     "pdf_scanned/multi_page_scanned.pdf": {
       "sha256": "cc9455385e3fa2621664ef6fc3afc58f5df29ea37139e0fdd8d5dbb3b74073e1",
       "size": 2527028
@@ -1677,6 +1681,22 @@
       "sha256": "2da8b562f74b778fcba92efe07fb8892aad1b279dbf84daba412ee148a4b56c8",
       "size": 1233386
     },
+    "pdf_scanned/synthetic_math_interpolation_scanned.pdf": {
+      "sha256": "99dd205450e6dfbd638a2a4ce03058ba595436acdafe0e736e126753c5b71834",
+      "size": 14894
+    },
+    "pdf_scanned/synthetic_math_interpolation_skew_scanned.pdf": {
+      "sha256": "553023190b90aa46e6196e74a55bf71939853db478c3109aea3ef1e90c4e4fe7",
+      "size": 15586
+    },
+    "pdf_scanned/synthetic_math_matrix_scanned.pdf": {
+      "sha256": "8505715c5362ce8eb733cd3cc783d27cffa218d51dabbf2a971820cc37b09090",
+      "size": 10464
+    },
+    "pdf_scanned/synthetic_math_series_scanned.pdf": {
+      "sha256": "6dcf181fa25d84409f7d99531bb4f1cd6607f003b73f81de8dbeb34e91fb3074",
+      "size": 11751
+    },
     "ppt/simple.ppt": {
       "sha256": "731b1a5478f36b996e02a0a2151df3de46c1c8293b3f0e10e7eda8aa24f249ab",
       "size": 504320