feat(table): detect table regions from text alignment, opt-in

bench/README.md

@@ -13,7 +13,7 @@ Each harness downloads what it needs into a scratch directory.
 | Benchmark | Dataset | Measures | Report |
 | --- | --- | --- | --- |
 | [`icdar2013/`](icdar2013/) | ICDAR 2013 Table Competition (125 PDFs) | table detection + structure | [2026-08-02](../docs/evaluations/2026-08-02-icdar2013-table-structure.md) |
-| [`icdar2013/compare.py`](icdar2013/compare.py) | same, 10 systems | pdfgrab vs the whole field | [2026-09-17](../docs/evaluations/2026-09-17-field-comparison-and-metric-correction.md) |
+| [`icdar2013/compare.py`](icdar2013/compare.py) | same, 13 systems | pdfgrab vs the whole field | [2026-09-17](../docs/evaluations/2026-09-17-field-comparison-and-metric-correction.md) |
 | [`icdar2013/oracle.py`](icdar2013/oracle.py) | same, with ground-truth boundaries | the ceiling a layout model could reach | [2026-08-03](../docs/evaluations/2026-08-03-hybrid-ceiling-oracle-boundaries.md) |
 
 ## Why the numbers live in `docs/evaluations/`

bench/icdar2013/extract.go

@@ -26,6 +26,7 @@ func main() {
 	strategy := flag.String("strategy", "lines",
 		"lines | text | mixed | auto | lines-then-mixed | fallback")
 	merge := flag.Bool("merge", false, "TableSettings.MergeSplitTokens")
+	detect := flag.Bool("detect", false, "TableSettings.DetectRegions (text-alignment region detection)")
 	oracle := flag.String("oracle", "",
 		`JSON of per-page explicit edges: {"1":{"v":[..],"h":[..]}}`)
 	flag.Parse()
@@ -69,6 +70,15 @@ func main() {
 		attempts = []pdfgrab.TableSettings{lines}
 	}
 
+	// Region detection is a property of the run, not of a strategy, so
+	// it is applied to whichever attempts the strategy selected. It only
+	// affects text-derived axes; on a pure "lines" run it is inert.
+	if *detect {
+		for i := range attempts {
+			attempts[i].DetectRegions = true
+		}
+	}
+
 	// Oracle mode: the caller supplies the row/column boundaries and
 	// pdfgrab only fills the cells. This is exactly the shape of the
 	// hybrid a layout model would drive — and fed GROUND-TRUTH edges it

bench/icdar2013/systems.py

@@ -135,7 +135,8 @@ def timed(fn: Callable[[str], Tables], pdf: str, t: Timing) -> Tables:
 # --- adapters ---------------------------------------------------------
 
 
-def pdfgrab(exe: str, strategy: str, merge: bool = False) -> Callable[[str], Tables]:
+def pdfgrab(exe: str, strategy: str, merge: bool = False,
+            detect: bool = False) -> Callable[[str], Tables]:
     """pdfgrab, via the benchmark's Go extractor binary."""
 
     def run(pdf: str) -> Tables:
@@ -144,6 +145,8 @@ def run(pdf: str) -> Tables:
         cmd = [exe, "-strategy", strategy]
         if merge:
             cmd.append("-merge")
+        if detect:
+            cmd.append("-detect")
         cmd.append(pdf)
         out = subprocess.run(cmd, capture_output=True, timeout=120).stdout
         return [t["rows"] for t in json.loads(out or b"[]")]
@@ -261,6 +264,12 @@ def build_adapters(exe: str, gx_exe: str = "") -> list[Adapter]:
                 note="the library under test"),
         Adapter("pdfgrab (auto)", pdfgrab(exe, "auto"),
                 note="one-axis-ruled support, opt-in"),
+        Adapter("pdfgrab (text+detect)", pdfgrab(exe, "text", detect=True),
+                note="HAL-1362: text-alignment region detection"),
+        Adapter("pdfgrab (fallback+detect)", pdfgrab(exe, "fallback", detect=True),
+                note="HAL-1362: lines, then text within detected regions"),
+        Adapter("pdfgrab (text, no detect)", pdfgrab(exe, "text"),
+                note="control: page-wide text strategy"),
         Adapter("pdfplumber (lines)", pdfplumber_tables("lines"),
                 module="pdfplumber", install="pdfplumber",
                 note="the implementation pdfgrab is a port of"),