wheee happy case is probably slow but probably ok

cluster.py

@@ -20,6 +20,8 @@ class Chars:
       # given median character size
       # drop all above+below first instance
       heights = [i["y1"]-i["y0"] for i in self.chars]
+      if not heights:
+         return [self]
       median_height = sorted(heights)[len(heights)//2]
       at_least_median_height = [i for i in self.chars if i["y1"]-i["y0"] >= median_height]
       at_least_median_height = sorted(at_least_median_height, key=lambda x:x["y0"])
@@ -72,6 +74,14 @@ class Chars:
                result2.append(sub)
          result = result2
 
+      for i in result:
+         i.merge()
+         for j in range(len(i.chars)):
+            i.chars[j]["x0"] -= median_height
+            i.chars[j]["x1"] += median_height
+            i.chars[j]["y0"] -= median_height * 3
+            i.chars[j]["y1"] += median_height * 3
+
       return result
 
    def merge_in(self, other):

main.py

@@ -0,0 +1,22 @@
+import debug
+import cluster
+import config
+import pdfplumber
+
+def main():
+   with pdfplumber.open(config.INPUT) as pdf:
+      for page in pdf.pages:
+         got = cluster.Chars(page.chars, page).divide_into_columns()
+         debug.draw_boxes(page, [
+            {
+               "x0": i.chars[0]["x0"],
+               "x1": i.chars[0]["x1"],
+               "y0": i.chars[0]["y0"],
+               "y1": i.chars[0]["y1"],
+               "debug_label": i.n,
+            }
+            for i in got if i.chars
+         ])
+
+if __name__ == "__main__":
+   main()