AI Demos Research — the structured-intelligence platform. Every verdict on these pages opens to the execution behind it.
Graded 22 October 2026

Does MarkItDown preserve equations as math markup or an honest fallback?

MarkItDown corrupted all six equations rather than preserving them as math or providing an honest fallback. Fractions, exponents and subscripts became disordered fragments, while the surrounding prose remained accurate.

1 of 1 test case failed

Every test case in this scenario has a result.

Pass rate0%0 of 1 gradable
Coverage1 of 1test case with a result
0PassNo test case passed.
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 test case below: what we sent, what we checked, what the tool returned, and the proof.

The test case

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

Equations — math markup or honest fallbackFailEvidence
What we sent

We ran MarkItDown’s conversion method once on the complete PDF and saved the returned Markdown to a file.

InputDocument (PDF)
Input PDF: six typeset equations with fractions, superscripts and subscripts across three pages.
Open the PDF ↗
What the tool returned

Copied from Output file (Markdown), lines 16–19, 22–29, 36–41, 46–50, 66–70, 88–92

Markdownthe whole file · cited lines 16–19, 22–29, 36–41, 46–50, 66–70 and 88–92 marked
Tarnbeck Institute of Hydrology TIH/TN/18
A one-dimensional model of tidal attenuation in the Braithe channel
Technical note TIH/TN/18  ·  January 2026
R. K. Sandiman and H. Vale
1  Introduction
The tide entering the Braithe is attenuated over the eleven kilometres between the bar and Braithe
Bridge, and the attenuation is large enough to matter to anyone predicting water levels in the upper
estuary. This note sets out the one-dimensional model the Institute has used since 2018, states the
approximation on which the working formula rests, and compares the result with the gauge record at
three stations.
Nothing here is new. The intention is to have the derivation, the assumptions and the coefficients in one
place, because the formula has been quoted in Partnership papers without them.
2  Governing equations
Take the  channel  as  a  single  reach  of  slowly  varying  cross-section  and  neglect  lateral  inflow.
Conservation of mass gives
|     | ∂A  | ∂Q  |     |
| --- | --- | --- | --- |
|     | +   | =0  | (1) |
|     | ∂t  | ∂x  |     |
where A is the cross-sectional area below the free surface and Q the discharge through it. Conservation of
momentum, with friction represented in the Manning form, gives
|      | 2   |       | 2    |
| ---- | --- | ----- | ---- |
| ∂Q ∂ | Q   | ∂η gn | Q|Q| |
(2)
| +     | +gA | +   | =0  |
| ----- | --- | --- | --- |
|       | ( ) |     | 4/3 |
| ∂t ∂x | A   | ∂x  | AR  |
in which η is the surface elevation above mean sea level, n is Manning's coefficient, R the hydraulic
radius R=A/P for wetted perimeter P, and g the acceleration due to gravity. The friction term is the only
nonlinear one that matters over the range of interest.
3  Attenuation of the leading harmonic
For a channel of nearly uniform depth the leading semidiurnal harmonic decays close to exponentially
with distance upstream, so that
−μx (3)
| η(x,t)=η | e   | cos(ωt−kx) |     |
| -------- | --- | ---------- | --- |
0
with η  the amplitude at the bar, ω the angular frequency of the harmonic, k the wavenumber and μ the
0
attenuation coefficient. Linearising the friction term by the Lorentz method and collecting terms gives
Page 1 of 3

Tarnbeck Institute of Hydrology TIH/TN/18
2
ω χ 1/2 8n cη
0
μ= , χ= (4)
c(2(1+χ 2 )) 3πh 4/3
where c is the frictionless wave speed and h the mean depth. The second expression is the friction
parameter; it is small in the lower reach and approaches unity above Harrowby Reach, which is where
the exponential form begins to fail.
4 Application to the Braithe
Taking the mean depth as eleven metres below the bar and four metres at the bridge, with Manning's
coefficient at the value fitted in 2018, the formula reproduces the observed amplitude at Vardenne Quay
to within four per cent and at Harrowby Reach to within nine. At Braithe Bridge it overestimates the
amplitude, and the discrepancy grows through the spring tides, which is the behaviour the friction
parameter predicts.
A two-reach treatment with separate depths would probably remove most of the error at the bridge. It has
not been attempted here because the gauge record above Harrowby Reach is too short to fit a second
coefficient with any confidence.
5 Coefficients
Manning's coefficient is not constant with depth in a channel of this kind, and the single fitted value used
above is a compromise. Fitting the 2018 record with a depth-dependent form gives
b
n(h)=n + (5)
∞
1/3
h
with n the deep-water value and b a bed constant. The fit is better in the lower reach and no better at
∞
the bridge, which suggests that the error there is not in the friction term at all.
The two constants were fitted to the 2018 record and have not been refitted since. Refitting on the four
years now available would be worth doing; the Institute's expectation is that the deep-water value will
move very little and the bed constant appreciably, because it is carrying the shallow reaches where the
record has improved most.
6 Limitations
Three limitations should be stated plainly. The first is the one-dimensional assumption, which fails where
the channel divides above Harrowby Reach. The second is the neglect of the freshwater inflow, which is
small in summer and not small after rain. The third is the linearisation itself, which is a poor
approximation once the friction parameter approaches unity - that is, in exactly the reach where the
model is least accurate.
The error measure quoted in the previous section is
Page 2 of 3

Tarnbeck Institute of Hydrology TIH/TN/18
η −η
| o m|
ε= (6)
η
o
taken over the spring-neap cycle at each gauge, with η the observed amplitude and η the modelled
o m
one. It is a crude measure and it flatters the model at the bar, where the amplitude is large and the
absolute error is not.
None of this is an argument against using the formula, which is quick, transparent and good enough for
the purposes the Partnership puts it to. It is an argument for quoting it with the reach and the tidal range
attached, which the papers that quote it have not always done.
References
Sandiman, R. K. (2018). Tidal propagation in the Braithe: a first fit. Tarnbeck Institute internal report TIH/IR/09.
Vale, H. and Croyde, P. (2021). The Harrowby Reach gauge, 2015 to 2020. Journal of Estuarine Hydrology 44,
218-31.
Prosser, A. (2009). Friction coefficients for the Braithe channel. Braithe Harbour Board, unpublished.
Page 3 of 3

Returned Markdown: broken equations, with Windows line endings; LF normalisation reproduces the receipt’s output fingerprint.

Open the Markdown file ↗
Expected vs. Found
✗Expected: Preserve equations as LaTeX or MathML, or provide an honest fallback rather than garbled text.Found: Of six source equations, none became LaTeX, MathML or an honest fallback; they broke into tables or loose fragments, with no image references or placeholders.
Supporting proof
Proof 1Screenshot
Terminal screenshot: the complete PDF was processed and conversion finished normally.
Open original ↗
Why this result

The default pipeline finished converting the complete PDF, so this is not an execution failure. It failed the math requirement: neither usable equations nor an honest fallback survived.

Also observed in this test case

Success without a warning. The conversion reported success without warning that the equations in the returned Markdown were corrupted.

Tested by Vamshidhar Rao Thumula · evidence dated 22 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
MarkItDown
Build or version
markitdown 0.1.7 (PyPI, markitdown[pdf])
Surface
SDK / library
Set up before the run
The technical note on tidal attenuation, including typeset fractions, exponents and subscripts, was to be converted in full with the default PDF pipeline, once, without tool-specific tuning.
Pipeline
file conversion · default PDF pipeline, no options, one pass
Tested
22 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
12 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/markitdown/a-document-containing-mathematical-equations · 0 pass · 1 fail · coverage 1/1 · graded 2026-10-22

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.

Input · Document (PDF) ↗application/pdf · 236 KB
Output file (Markdown) ↗text/markdown; charset=utf-8 · 5 KB
Proof 1 · Screenshot ↗image/png · 71 KB
File fingerprints (SHA-256)

A fingerprint identifies the exact file used for this result.

Input · Document (PDF)f3d69b6d9d2191c5806a426fd4ea5e209dcd7cf693ad6abe67a5d18a014b9fd6
Output file (Markdown)72597026833ac8059d3841596dbc28c7a83b837e9baffd0a9d561a3b39a6df91
Proof 1 · Screenshote951b17830889a554d18b2009fdeb63931695f4c4c556cc8c701b49fd319f3be