AI Demos Research — the structured-intelligence platform. Every verdict on these pages opens to the execution behind it.
Graded 26 October 2026

Does PyMuPDF4LLM keep a table together across a page break?

PyMuPDF4LLM produced two Markdown tables for each document, preserving the repeated-header version but mishandling the headerless continuation. It kept 22 of 22 data rows in the first; in the second, a data row became headings, leaving 25 of 26 data rows and incorrect continuation column names.

1 of 2 test cases passed · 1 failed

Every test case in this scenario has a result.

Pass rate50%1 of 2 gradable
Coverage2 of 2test cases with a result
1PassDid everything it was expected to do.
1FailMissed at least one thing it was expected to do; the reason and proof are with each test case below.
0Not gradableEvery result here could be graded.
0UntestedEvery test case in this scenario has a result.
The pass rate is a summary. The evidence is the 2 test cases below: what we sent, what we checked, what the tool returned, and the proof.

The 2 test cases

Each test case is judged on its own: Pass, Fail, Not gradable, or Untested. The scenario result above is based on these test cases.

Cross-page table — one table, completePassEvidence
What we sent

We converted the monitoring-station PDF to Markdown using the library's standard conversion entry point, with no optional arguments, and saved the result.

InputDocument (PDF)
Input PDF: station table continuing with repeated column headings
Open the PDF ↗
What the tool returned

Copied from Output file (Markdown), lines 89–108

Markdown
|**Station**|**Grid reference**|**Depth (m)**|**Sampling interval**|**Installed**|
|---|---|---|---|---|
|Vardenne Quay|NX 4118 6602|2.4|Fortnightly|March 2011|
|Thrimby Point|NX 4276 6491|1.8|Fortnightly|March 2011|
|Kellow Sands north|NX 4390 6588|0.9|Monthly|June 2013|
|Kellow Sands south|NX 4402 6531|1.1|Monthly|June 2013|
|Marnhead Deep|NX 4025 6714|8.6|Monthly|October 2014|
|Braithe Bridge|NX 3982 6853|3.2|Weekly|April 2015|
|Harrowby Reach|NX 4147 6770|4.5|Fortnightly|April 2015|
|Sandiman Bank|NX 4233 6685|1.3|Monthly|September 2016|
|Vellamy Creek|NX 4318 6438|0.6|Monthly|September 2016|
|Tarbert Jetty|NX 4064 6552|2.9|Fortnightly|May 2018|
|Croyde Gut|NX 4351 6367|1.6|Quarterly|May 2018|
|Prosser Flats|NX 4189 6304|0.7|Quarterly|August 2020|
|Old Lifeboat House|NX 4295 6620|1.9|Fortnightly|August 2020|
|Training Wall east|NX 4126 6459|5.4|Weekly|February 2023|
|Lissamer Point|NX 4208 6742|3.7|Monthly|February 2023|
|Kellow Creek head|NX 4436 6612|0.4|Quarterly|March 2024|
|Vardenne Outfall|NX 4102 6571|2.1|Weekly|March 2024|
|Marnhead Bar|NX 3968 6741|9.3|Monthly|September 2024|

Copied from Output file (Markdown), lines 112–116

Markdown
Page 3 of 4 

Braithe Estuary Partnership  ·  Water quality programme 

### Table 1 (continued) 

Copied from Output file (Markdown), lines 118–123

Markdown
|**Station**|**Grid reference**|**Depth (m)**|**Sampling interval**|**Installed**|
|---|---|---|---|---|
|Thrimby Spit|NX 4301 6467|1.4|Fortnightly|September 2024|
|Harrowby Cut|NX 4163 6799|5.8|Monthly|April 2025|
|Boatyard Steps|NX 4088 6588|1.2|Fortnightly|April 2025|
|Cavenshaw Bight|NX 4372 6521|2.6|Quarterly|November 2025|

Returned Markdown: 22 station rows split between two tables

Open the Markdown file ↗
Expected vs. Found
✓Expected: Retain all 22 data rows.Found: The output contains 22 data rows, matching the PDF's 22: 18 in the first table and 4 in the second.
✓Expected: Preserve Thrimby Spit's grid reference, depth, sampling interval and installation date.Found: Thrimby Spit retains grid reference NX 4301 6467, depth 1.4 m, fortnightly sampling and installation in September 2024.
✓Expected: Preserve Harrowby Cut's grid reference, depth, sampling interval and installation date.Found: Harrowby Cut retains grid reference NX 4163 6799, depth 5.8 m, monthly sampling and installation in April 2025.
✓Expected: Preserve Boatyard Steps' grid reference, depth, sampling interval and installation date.Found: Boatyard Steps retains grid reference NX 4088 6588, depth 1.2 m, fortnightly sampling and installation in April 2025.
✓Expected: Preserve Cavenshaw Bight's grid reference, depth, sampling interval and installation date.Found: Cavenshaw Bight retains grid reference NX 4372 6521, depth 2.6 m, quarterly sampling and installation in November 2025.
✓Expected: Keep Marnhead Bar before Thrimby Spit across the page break.Found: Marnhead Bar ends the first table; Thrimby Spit starts the second.
✓Must not appear: Do not treat repeated column headings as data; a second table with those headings is allowed.Found: The repeated headings became the second table's header row, not a data row.
✓Must not appear: Keep page numbers, the running header and the continuation caption outside table bodies.Found: Page numbers, the running header and the continuation caption remain outside both table bodies.
Supporting proof
Proof 1Screenshot
Execution screenshot: conversion details, continuation output and the repeated-header source page
Open original ↗
Why this result

Two Markdown tables were allowed if the continuation retained its headers. The output preserved the required rows and column relationships, so the split did not prevent a pass.

Also observed in this test case

Continuation caption formatting. The continuation caption "Table 1 (continued)" became a third-level Markdown heading rather than plain text.

Tested by Vamshidhar Rao Thumula · evidence dated 24 October 2026
Cross-page table — continuation without a repeated headerFailEvidence
What we sent

We converted the sampling-visits PDF to Markdown using the library's standard conversion entry point, with no optional arguments, and saved the result.

InputDocument (PDF)
Input PDF: sampling visits continuing without repeated column headings
Open the PDF ↗
What the tool returned

Copied from Output file (Markdown), lines 17–19

Markdown
|Visit|Station|Date|Time|Determinations taken|Sampler|
|---|---|---|---|---|---|
|V-01|Vardenne Quay|2026-01-06|09:40|Suspended solids, chloride|J. Marren|

Copied from Output file (Markdown), lines 32–52

Markdown
|V-14|Training Wall east|2026-02-05|07:45|Suspended solids, chloride, nitrogen|J. Marren|



Page 2 of 2 

Braithe Estuary Partnership  ·  Water quality programme 

|V-15|Lissamer Point|2026-02-05|08:35|Suspended solids, chloride, nitrogen|J. Marren|
|---|---|---|---|---|---|
|V-16|Kellow Creek head|2026-02-06|11:20|Suspended solids, chloride, orthophosphate|R. Ashby|
|V-17|Vardenne Outfall|2026-02-09|07:30|Suspended solids, chloride, DOC|P. Sowerby|
|V-18|Marnhead Bar|2026-02-09|08:50|Suspended solids, chloride, DOC|P. Sowerby|
|V-19|Thrimby Spit|2026-02-10|09:25|Suspended solids, chloride|J. Marren|
|V-20|Harrowby Cut|2026-02-11|10:15|Suspended solids, chloride, nitrogen|R. Ashby|
|V-21|Boatyard Steps|2026-02-11|11:00|Suspended solids, chloride|R. Ashby|
|V-22|Cavenshaw Bight|2026-02-12|08:10|Suspended solids, chloride, orthophosphate|P. Sowerby|
|V-23|Braithe Bridge|2026-02-16|07:55|Suspended solids, chloride, nitrogen|J. Marren|
|V-24|Vardenne Outfall|2026-02-16|08:40|Suspended solids, chloride, DOC|J. Marren|
|V-25|Training Wall east|2026-02-19|07:50|Suspended solids, chloride, nitrogen|P. Sowerby|
|V-26|Harrowby Reach|2026-02-19|08:45|Suspended solids, chloride, nitrogen|P. Sowerby|

Returned Markdown: a visit used as headings and 25 remaining data rows

Open the Markdown file ↗
Expected vs. Found
✗Expected: Retain all 26 data rows.Found: The output has 25 data rows against 26 in the PDF: 14 in the first table and 11 in the second.
✗Expected: Keep visit 15 as a data row with its original values.Found: Visit 15's values became the second table's headings; no data row remains for that visit.
✓Expected: Preserve visit 22's station, date, time, measurements and sampler.Found: Visit 22 retains Cavenshaw Bight, 12 February 2026, 08:10, suspended solids, chloride, orthophosphate and its original sampler, under incorrect headings.
✓Expected: Preserve visit 26's station, date, time, measurements and sampler.Found: Visit 26 retains Harrowby Reach, 19 February 2026, 08:45, suspended solids, chloride, nitrogen and its original sampler, under incorrect headings.
✓Expected: Keep visit 14 before visit 15 across the page break.Found: Visit 14 ends the first table; visit 15's content appears next, as the second table's headings.
✗Expected: Keep continuation rows under the original six column headings, in one table or a second table.Found: Visits 16 through 26 appear under headings taken from visit 15's values, rather than the original six column headings.
✓Must not appear: Keep page numbers and the running header outside table bodies.Found: The page number and running header sit between the two tables, outside both table bodies.
Supporting proof
Proof 1Screenshot
Execution screenshot: conversion details, the misplaced visit headings and the headerless source page
Open original ↗
Why this result

The continuation needed its original columns. The default conversion completed but turned its first row into headings, so this was a table-extraction failure rather than a blocked conversion. This result concerns the default pipeline, not every available table-detection option.

Also observed in this test case

Remaining visit values. Visits 16 through 26 retained their original cell values.

Tested by Vamshidhar Rao Thumula · evidence dated 24 October 2026

Configuration and setup

How this tool was set up for the run and what the test needed in place. Each row is a fact from the run's records; a fact the records do not hold is left out, not guessed.

Software that produced the output
PyMuPDF4LLM
Build or version
PyMuPDF4LLM 1.28.2
Surface
SDK / library
Conversion
Default pipeline, no keyword arguments
Tested
24 October 2026 · Vamshidhar Rao Thumula

How this scenario is graded

How we decide Pass, Fail and Not gradable. The same rules apply to every tool tested on this scenario.

How results are decided

Each test case gets one result per tool: Pass, Fail or Not gradable. A test case we haven't run yet shows Untested. There are no partial results.

The rules
  • Pass: the tool did everything the test expected, and nothing it said contradicts the correct answer.
  • Fail: at least one expected behaviour clearly didn't happen; the test case's result says which one and quotes the tool.
  • Not gradable: our evidence couldn't settle the outcome (for example a record we needed is missing). It is never counted as a fail, and the test case's result says what's missing.

Where this sits in the benchmark

This page is one cell of a larger study: one tool, one scenario. Only this benchmark's frame appears here.

History of this result

What has happened to this result since it was first published. Runs and grades are never overwritten: a retest or a re-grade publishes a new result and keeps the earlier one readable.

from the publication record
16 November 2026First publishedConverting a complex PDF into clean Markdown with an open-source library v1

Act on this result

Nothing filed here edits the run or the grade. A challenge opens a review, and a review can produce a new run or a re-grade — which becomes the current result and leaves this one in the history.

This matches what I see

You run the same kind of test against your own setup and get the same behaviour.

Agree →
This does not match

Yours behaves differently. Tell us what you got, with a screenshot if you have one.

Disagree →
Point out an issue

Something here is wrong — a reference value, a transcription, a grade.

Report an issue →
Request a retest

On a newer build, a larger dataset, or your own setup.

Request a retest →
We have fixed this

Tell us what changed and we schedule a rerun of the failing test case. The old result stays as history.

Vendor notice →
Filed against this evidenceNothing yet. Challenges, counter-evidence and fix notices appear here with their outcome, and stay on the page after they are resolved.
Cite this result
aidemos.com/benchmarks/pdf-to-markdown-open-source/results/pymupdf4llm/a-table-that-continues-across-a-page-break · 1 pass · 1 fail · coverage 2/2 · graded 2026-10-26

The same record is available as structured data through the AI Demos MCP server, with the counts, the coverage and every per-test-case reason carried as fields.

Verify the proof files

These files support this result. Open a file to inspect the original evidence.

File fingerprints (SHA-256)

A fingerprint identifies the exact file used for this result.

Cross-page table — one table, complete · Input · Document (PDF)8d903e9d03bc804cb94d363c9f7fc7db871e22a0f9a0998d409f1eea845a3209
Cross-page table — one table, complete · Output file (Markdown)b8411145b6c570fa5e6e0fb68ebeda56cb23c045e1c04d2298175bb49a028ea8
Cross-page table — one table, complete · Proof 1 · Screenshot21989b50d00ec48f42ef2d2c2575104ecbd00f4d965ceaf71327caa190f6038c
Cross-page table — continuation without a repeated header · Input · Document (PDF)17f7972834b1eb0e4bf5ce81b9d9f5e8b21fce710084efc9a01d92ef15469a41
Cross-page table — continuation without a repeated header · Output file (Markdown)4623fdf740ed774e2345add39e14b28ced91c4772e8a60cb273991f0286ba7dc
Cross-page table — continuation without a repeated header · Proof 1 · Screenshot39d83ae79e2899bf7a918ffc86e10b643654d0fe363e5d7b5ccd2fa68b7687e4