feat(propose): the PDF shipped a structure index and the build discarded it unopened

`okf build` recovers a PDF's boundaries from the shape of its page text and
never opens the file's own `/Outlines` bookmark tree. On a 701-page process
code whose publisher also ships a NISO-STS structure for it, measured outside
this repository and reproduced here exactly: the shipped default finds 1967 of
2761 titled sections, 0 of its 28 chapters, and 794 of 794 misses have their
heading text PRESENT in the extracted text. The line was read; the boundary
was never opened. The same file's bookmark tree matches 2761 of 2761 of those
titles exactly after normalisation.

`--pdf-outline`, OFF, cuts a PDF at the boundaries its tree declares.

  boundaries                 1967 of 2761  ->  2759 of 2761  (gate was 2700)
  depth 1                       0 of 28    ->     28 of 28
  titles identical to source        --     ->   2761 of 2761
  false positives             163 of 2182  ->      3 of 2762
  directories with two files  132 of 2050  ->      2 of 2738
  front-matter concepts        72 of 2182  ->      2 of 2762
  consumption fasit present       4 of 7   ->        7 of 7
  hit@1 / hit@8 / hit@50      1/6 2/6 4/6  ->   3/6 5/6 6/6

It is a SEGMENTATION arm, not a reader option: the extracted text is byte for
byte the same either way. A PDF with no tree builds byte-identically with the
flag on -- `diff -r` empty across the pre-change tree, the arm off and the arm
on. An unresolvable `/Dest` is dropped and COUNTED, never fabricated into a
boundary and never a refusal of the file.

The bridge from (page, y) to a line index is the whole risk, so both routes
are measured. `extract_text_lines` splits lines identically to `extract_text`
on 701 of 701 pages, and is CHECKED per page rather than assumed. The y route
and the title route disagree on 0 of 2762 nodes, flat from a 0pt tolerance to
8pt and collapsing at 12pt, so the rule ships with no tolerance constant. The
naive "nearest line" rule was wrong on 1840 of 2762, one line early every time.

The orphan check is not applied to a bookmark mark: it asks whether anything
stands under a candidate's first line, which is the right question for a
heuristic's guess and the wrong one for a publisher's declaration. 683 of 2762
marks are container sections; applying it scores 2079 instead of 2759.

No new dependency and no second parse of the pages: `pdfminer.six` already
ships under `pdfplumber` in `[extract]`. 119.22s -> 183.31s wall, peak RSS
3252 -> 3251 MiB. The default does not move; 1 of the 8 reference PDFs carries
a usable tree at all.

`.pdf` also gains its `_EVIDENCE` row, as `measured` -- it was the row with the
most measurement behind it and no entry in the table.

Report: docs/2026-09-10-k3-runde12-pdf-outlines.md

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
Kjell Tore Guttormsen 2026-09-10 02:27:41 +02:00
commit e1f4faa098
12 changed files with 1261 additions and 15 deletions

View file

@ -82,6 +82,172 @@ NUMBERED_FONT_CONTENT = (
)
# THREE pages and a THREE-LEVEL `/Outlines` tree, which is the shape the
# bookmark arm has to be proved against. A two-level tree cannot tell "the
# level the node declares" apart from "one below the root", and a
# one-page fixture cannot tell a page-local line offset from a document-wide
# one -- the arm's whole risk is the bridge from (page, y) to a line index.
#
# The third page carries FOUR lines and its second bookmark points at the
# THIRD of them, so a bridge that resolved to the page and stopped would put
# the mark two lines early and still look like it worked.
OUTLINED_CONTENTS = (
b"BT /F1 12 Tf 20 170 Td (1 Grunnlag) Tj ET\n"
b"BT /F1 12 Tf 20 150 Td (Innledende tekst om grunnlaget.) Tj ET\n",
b"BT /F1 12 Tf 20 170 Td (1.1 Omfang) Tj ET\n"
b"BT /F1 12 Tf 20 150 Td (Omfanget dekker hele arbeidet.) Tj ET\n",
b"BT /F1 12 Tf 20 170 Td (1.2 Krav) Tj ET\n"
b"BT /F1 12 Tf 20 150 Td (Kravet gjelder alle klasser.) Tj ET\n"
b"BT /F1 12 Tf 20 130 Td (1.2.1 Materialer) Tj ET\n"
b"BT /F1 12 Tf 20 110 Td (Materialene skal vaere godkjente.) Tj ET\n",
)
#: `(title, level, page index, /XYZ top)`. The `top` values are the PDF's own
#: bottom-up user space: 185 sits above the first line of a page and 150 above
#: its third, which is what makes the third-line mark a statement rather than
#: a coincidence.
OUTLINED_TREE = (
("1 Grunnlag", 1, 0, 185),
("1.1 Omfang", 2, 1, 185),
("1.2 Krav", 2, 2, 185),
("1.2.1 Materialer", 3, 2, 150),
)
#: One resolvable bookmark and one whose `/Dest` names an object that is not a
#: page. A PDF in the wild carries these; R761 carries none of them, so
#: without this fixture the "drop it, count it, do not fabricate a boundary"
#: branch would ship having never run.
BROKEN_DEST_CONTENT = (
b"BT /F1 12 Tf 20 170 Td (1 Grunnlag) Tj ET\n"
b"BT /F1 12 Tf 20 150 Td (Innledende tekst om grunnlaget.) Tj ET\n",
)
def build_outlined_pdf(
contents: tuple[bytes, ...],
tree: tuple[tuple[str, int, int, int], ...],
*,
broken_dest: bool = False,
) -> bytes:
"""`build_paged_pdf` plus a hand-laid `/Outlines` tree in the catalog.
The tree is written from the flat `(title, level, page, top)` list rather
than from a nested literal, because `/First`, `/Last`, `/Next`, `/Prev` and
`/Parent` all have to agree with each other and with the level column --
a hand-written nest gets one of them wrong silently and the reader then
reports a level the document never declared.
`broken_dest` appends one extra item whose `/Dest` names the FONT object
instead of a page. It is a valid indirect reference to a real object that
is not a page, which is the failure a reader has to survive.
"""
count = len(contents)
page_numbers = [3 + 2 * index for index in range(count)]
font_number = 3 + 2 * count
outlines_number = font_number + 1
items = list(tree) + ([("Uoppl\u00f8selig", 1, -1, 185)] if broken_dest else [])
item_numbers = [outlines_number + 1 + index for index in range(len(items))]
kids = b" ".join(str(number).encode() + b" 0 R" for number in page_numbers)
objects = [
b"<< /Type /Catalog /Pages 2 0 R /Outlines " + str(outlines_number).encode() + b" 0 R >>",
b"<< /Type /Pages /Kids [" + kids + b"] /Count " + str(count).encode() + b" >>",
]
for index, content in enumerate(contents):
stream_number = page_numbers[index] + 1
objects.append(
b"<< /Type /Page /Parent 2 0 R /MediaBox [0 0 200 200] /Contents "
+ str(stream_number).encode()
+ b" 0 R /Resources << /Font << /F1 "
+ str(font_number).encode()
+ b" 0 R >> >> >>"
)
objects.append(
b"<< /Length " + str(len(content)).encode() + b" >>\nstream\n" + content + b"endstream"
)
objects.append(
b"<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica /Encoding /WinAnsiEncoding >>"
)
# The parent of an item is the last item seen at the level above it; its
# previous sibling is the last item seen at its OWN level under that same
# parent. Both are resolved in one forward pass so the links cannot drift.
parents: list[int | None] = []
last_at_level: dict[int, int] = {}
siblings: list[int | None] = []
children: dict[int, list[int]] = {}
roots: list[int] = []
for position, (_, level, _, _) in enumerate(items):
parent = last_at_level.get(level - 1) if level > 1 else None
parents.append(parent)
previous = None
group = children.setdefault(parent, []) if parent is not None else roots
if group:
previous = group[-1]
siblings.append(previous)
group.append(position)
last_at_level[level] = position
for deeper in [key for key in last_at_level if key > level]:
del last_at_level[deeper]
def _ref(position: int | None) -> bytes:
return b"" if position is None else str(item_numbers[position]).encode() + b" 0 R"
root_body = b"<< /Type /Outlines /Count " + str(len(items)).encode() + b" >>"
if roots:
root_body = (
b"<< /Type /Outlines /First "
+ _ref(roots[0])
+ b" /Last "
+ _ref(roots[-1])
+ b" /Count "
+ str(len(items)).encode()
+ b" >>"
)
objects.append(root_body)
for position, (title, _, page, top) in enumerate(items):
parent = parents[position]
parent_ref = _ref(parent) if parent is not None else str(outlines_number).encode() + b" 0 R"
group = children.get(parent, []) if parent is not None else roots
index_in_group = group.index(position)
body = b"<< /Title (" + _pdf_text(title) + b") /Parent " + parent_ref
if index_in_group > 0:
body += b" /Prev " + _ref(group[index_in_group - 1])
if index_in_group + 1 < len(group):
body += b" /Next " + _ref(group[index_in_group + 1])
own = children.get(position, [])
if own:
body += b" /First " + _ref(own[0]) + b" /Last " + _ref(own[-1])
body += b" /Count " + str(len(own)).encode()
target = str(font_number).encode() if page < 0 else str(page_numbers[page]).encode()
body += b" /Dest [" + target + b" 0 R /XYZ 20 " + str(top).encode() + b" 0] >>"
objects.append(body)
out = bytearray(b"%PDF-1.4\n")
offsets = []
for number, item in enumerate(objects, start=1):
offsets.append(len(out))
out += str(number).encode() + b" 0 obj\n" + item + b"\nendobj\n"
xref_at = len(out)
size = str(len(objects) + 1).encode()
out += b"xref\n0 " + size + b"\n0000000000 65535 f \n"
for offset in offsets:
out += ("%010d 00000 n \n" % offset).encode()
out += b"trailer\n<< /Size " + size + b" /Root 1 0 R >>\n"
out += b"startxref\n" + str(xref_at).encode() + b"\n%%EOF\n"
return bytes(out)
def _pdf_text(value: str) -> bytes:
"""A PDF literal string in WinAnsi, with the three delimiters escaped."""
raw = value.encode("cp1252")
for character in (b"\\", b"(", b")"):
raw = raw.replace(character, b"\\" + character)
return raw
def build_two_font_pdf(content: bytes) -> bytes:
"""A one-page PDF whose resources declare BOTH a regular and a bold font.
@ -439,6 +605,14 @@ if __name__ == "__main__":
(HERE / "numbered-font-krav.pdf").write_bytes(build_two_font_pdf(NUMBERED_FONT_CONTENT))
print("wrote numbered-font-krav.pdf")
(HERE / "outlined-krav.pdf").write_bytes(build_outlined_pdf(OUTLINED_CONTENTS, OUTLINED_TREE))
print("wrote outlined-krav.pdf")
(HERE / "outline-broken-dest.pdf").write_bytes(
build_outlined_pdf(BROKEN_DEST_CONTENT, (("1 Grunnlag", 1, 0, 185),), broken_dest=True)
)
print("wrote outline-broken-dest.pdf")
for name, parts in (
("two-line-krav.docx", _DOCX_PARTS),
("no-styles-krav.docx", _DOCX_NO_STYLES_PARTS),

45
tests/fixtures/outline-broken-dest.pdf vendored Normal file
View file

@ -0,0 +1,45 @@
%PDF-1.4
1 0 obj
<< /Type /Catalog /Pages 2 0 R /Outlines 6 0 R >>
endobj
2 0 obj
<< /Type /Pages /Kids [3 0 R] /Count 1 >>
endobj
3 0 obj
<< /Type /Page /Parent 2 0 R /MediaBox [0 0 200 200] /Contents 4 0 R /Resources << /Font << /F1 5 0 R >> >> >>
endobj
4 0 obj
<< /Length 105 >>
stream
BT /F1 12 Tf 20 170 Td (1 Grunnlag) Tj ET
BT /F1 12 Tf 20 150 Td (Innledende tekst om grunnlaget.) Tj ET
endstream
endobj
5 0 obj
<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica /Encoding /WinAnsiEncoding >>
endobj
6 0 obj
<< /Type /Outlines /First 7 0 R /Last 8 0 R /Count 2 >>
endobj
7 0 obj
<< /Title (1 Grunnlag) /Parent 6 0 R /Next 8 0 R /Dest [3 0 R /XYZ 20 185 0] >>
endobj
8 0 obj
<< /Title (Uoppløselig) /Parent 6 0 R /Prev 7 0 R /Dest [5 0 R /XYZ 20 185 0] >>
endobj
xref
0 9
0000000000 65535 f
0000000009 00000 n
0000000074 00000 n
0000000131 00000 n
0000000257 00000 n
0000000412 00000 n
0000000509 00000 n
0000000580 00000 n
0000000675 00000 n
trailer
<< /Size 9 /Root 1 0 R >>
startxref
771
%%EOF

79
tests/fixtures/outlined-krav.pdf vendored Normal file
View file

@ -0,0 +1,79 @@
%PDF-1.4
1 0 obj
<< /Type /Catalog /Pages 2 0 R /Outlines 10 0 R >>
endobj
2 0 obj
<< /Type /Pages /Kids [3 0 R 5 0 R 7 0 R] /Count 3 >>
endobj
3 0 obj
<< /Type /Page /Parent 2 0 R /MediaBox [0 0 200 200] /Contents 4 0 R /Resources << /Font << /F1 9 0 R >> >> >>
endobj
4 0 obj
<< /Length 105 >>
stream
BT /F1 12 Tf 20 170 Td (1 Grunnlag) Tj ET
BT /F1 12 Tf 20 150 Td (Innledende tekst om grunnlaget.) Tj ET
endstream
endobj
5 0 obj
<< /Type /Page /Parent 2 0 R /MediaBox [0 0 200 200] /Contents 6 0 R /Resources << /Font << /F1 9 0 R >> >> >>
endobj
6 0 obj
<< /Length 104 >>
stream
BT /F1 12 Tf 20 170 Td (1.1 Omfang) Tj ET
BT /F1 12 Tf 20 150 Td (Omfanget dekker hele arbeidet.) Tj ET
endstream
endobj
7 0 obj
<< /Type /Page /Parent 2 0 R /MediaBox [0 0 200 200] /Contents 8 0 R /Resources << /Font << /F1 9 0 R >> >> >>
endobj
8 0 obj
<< /Length 213 >>
stream
BT /F1 12 Tf 20 170 Td (1.2 Krav) Tj ET
BT /F1 12 Tf 20 150 Td (Kravet gjelder alle klasser.) Tj ET
BT /F1 12 Tf 20 130 Td (1.2.1 Materialer) Tj ET
BT /F1 12 Tf 20 110 Td (Materialene skal vaere godkjente.) Tj ET
endstream
endobj
9 0 obj
<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica /Encoding /WinAnsiEncoding >>
endobj
10 0 obj
<< /Type /Outlines /First 11 0 R /Last 11 0 R /Count 4 >>
endobj
11 0 obj
<< /Title (1 Grunnlag) /Parent 10 0 R /First 12 0 R /Last 13 0 R /Count 2 /Dest [3 0 R /XYZ 20 185 0] >>
endobj
12 0 obj
<< /Title (1.1 Omfang) /Parent 11 0 R /Next 13 0 R /Dest [5 0 R /XYZ 20 185 0] >>
endobj
13 0 obj
<< /Title (1.2 Krav) /Parent 11 0 R /Prev 12 0 R /First 14 0 R /Last 14 0 R /Count 1 /Dest [7 0 R /XYZ 20 185 0] >>
endobj
14 0 obj
<< /Title (1.2.1 Materialer) /Parent 13 0 R /Dest [7 0 R /XYZ 20 150 0] >>
endobj
xref
0 15
0000000000 65535 f
0000000009 00000 n
0000000075 00000 n
0000000144 00000 n
0000000270 00000 n
0000000425 00000 n
0000000551 00000 n
0000000705 00000 n
0000000831 00000 n
0000001094 00000 n
0000001191 00000 n
0000001265 00000 n
0000001386 00000 n
0000001484 00000 n
0000001616 00000 n
trailer
<< /Size 15 /Root 1 0 R >>
startxref
1707
%%EOF