AI Demos Research — the structured-intelligence platform. Every verdict on these pages opens to the execution behind it.
Graded 28 September 2026

Can Mistral AI handle an ordinary digital text document without missing, garbling, or inventing content?

Mistral AI did not keep an ordinary digital text document complete and unchanged. In the graded output for briefing_note.pdf, running headers and page numbers were spliced into body text at the cited line ranges, including 33-38, 56-64, 110-118, and 138-146.

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 on the row.
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 counts this row.

Ordinary digital text — complete and unchangedFailEvidence
What we sent
InputDocument (PDF)
Seven-page briefing-note PDF with running headers and page footers. From outside the graded run; shown for context, not graded.
Open the PDF ↗
What the tool returned
Markdown
Braithe Estuary Partnership

Briefing note BEP/BN/2026/04

# Sediment recharge at Kellow Sands: the 2025/26 placement

Prepared by the technical secretariat for the Partnership Board · 12 February 2026

## Background

The Partnership has placed dredged sand on the upper foreshore at Kellow Sands in each of the last three winters, under the standing consent granted in 2021 and the method agreed after the 2023 trial. The purpose is to slow the loss of intertidal in front of the saltmarsh, which the 2017 study put at rather more than a metre a year at the northern end.

The material is a by-product of maintenance dredging the Harbour Board carries out anyway. Placing it on the foreshore rather than at the licensed offshore site is cheaper for the Board and useful to the Partnership, and the arrangement has run without difficulty since 2021.

## The consent position

The standing consent granted in 2021 runs to March 2029 and covers placement of up to 25,000 cubic metres a winter at the two locations shown in the consent drawing. It carries eleven conditions, of which two bear on the operation: placement is not permitted between 1 March and 31 August, and the Partnership must notify the regulator seven days before a campaign begins.

Both conditions were met this winter. The notification for the November start was sent on 28 October and acknowledged, and the January loads fell within the same campaign and needed no further notice. The regulator inspected once, on 4 December, and raised nothing.

## The dredging programme

The material comes from the Harbour Board's maintenance dredging, which runs on a three-year cycle: the Vardenne Channel and the approach to the quay every year, the bar every second year, and the berths at Marnhead in the third. The volume available to the Partnership therefore varies with the cycle as well as with the winter, and the largest volumes fall in the years the bar is cut.

This was a channel-and-approach year, which is the smaller of the three. The Board's dredger worked twenty-two days in the estuary between October and January against thirty-one in the previous season, and the eight loads placed at Kellow Sands were what the programme could offer rather than what the Partnership asked for. The secretariat does not regard the difference as material at this stage of the trial.

## The winter's placement

The Partnership placed 18,400 cubic metres of dredged sand on the upper foreshore at Kellow Sands between 6 November 2025 and 22 January 2026. The material came from maintenance dredging of the Vardenne Channel and was placed in eight loads under the standing consent granted in 2021. This note records what was placed, what it cost, and what the winter surveys

Page 1 of 7

Braithe Estuary Partnership  
have shown so far.

Briefing note BEP/BN/2026/04

Placement followed the method agreed after the 2023 trial. Each load was discharged at high water on a rising spring tide, at a single point some forty metres seaward of the saltmarsh edge, and left to redistribute without machinery working on the flat. The intention is that the material moves landward under wave action over two or three winters rather than being spread mechanically, which the trial showed disturbs the buried peat horizon at Thrimby Point.

### **How the material is placed**

Each load is brought up on the flood and held off the placement point until high water. Discharge takes between forty minutes and an hour and a quarter depending on the load, and the vessel is off the flat and back in the channel within the hour after. The Partnership's own staff attend every discharge from the shore and record the time, the state of the tide, the wind and anything of note; the records are held with the placement file.

Markdown output of the briefing note, with headers and page numbers retained in body text. From outside the graded run; shown for context, not graded.

Copied from Output file (Markdown), lines 1–42 · an excerpt

Open the whole Markdown file ↗
Expected vs. Found
What this scenario evaluates

These are scenario-level criteria. Each test case's Expected and Found are listed separately.

  • Whether every paragraph is preserved complete and unchanged.
  • Whether nothing is silently missing, garbled, or invented.
  • Whether the result remains recognisably plain text without unintended changes to the content.
How this scenario is judged →
✗Found: Page headers and page numbers were spliced into body sentences at several page breaks.
Supporting proof
Proof 1Screenshot
Run metadata card showing the OCR run configuration and result.
Open original ↗
Why this result

The Markdown kept the source text, but page headers and page numbers were carried into the middle of sentences at several page breaks. Under the text-fidelity check, that is not unchanged output, so the result fails.

Also observed on this row

Nested bold headings. Headings appear at multiple depths, and many keep bold markup inside the heading text.

Character count mismatch. The run metadata reports 19,083 characters; the stored file is 19,096 bytes.

Tested by Chandresh Bisht · evidence dated 10 September 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
Mistral AI
Build or version
mistral-ocr-latest (model id echoed by the API; dated version not exposed)
Plan or tier
Mistral La Plateforme (metered OCR endpoint)
Model
mistral-ocr-latest
Surface
Mistral La Plateforme
Set up before the run
A short digital-text PDF of plain paragraphs was provided, and the complete PDF was converted to Markdown with the product’s default pipeline in one pass, never from a cropped page.
pipeline
Document AI default OCR pipeline, single pass, no flag tuning.
polling
synchronous call
request scope
the complete PDF, 7 page(s), in one request
endpoint
https://api.mistral.ai/v1/ocr
Tested
10 September 2026 · Chandresh Bisht

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 row says which 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 row 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
6 October 2026First publishedConverting a complex PDF into clean Markdown with a hosted API 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-apis/results/mistral-ai/an-ordinary-digital-text-document · 0 pass · 1 fail · coverage 1/1 · graded 2026-09-28

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 · 103 KB
Output file (Markdown) ↗text/markdown · 19 KB
Proof 1 · Screenshot ↗image/png · 87 KB
File fingerprints (SHA-256)

A fingerprint identifies the exact file used for this result.

Input · Document (PDF)291669466fefcdd19a890091f82c0fba09f8ddce2882ef5755c52d8abe28ffa1
Output file (Markdown)481e5bb2bdf5d126b6da9dedf36950a3a376ea0be21f9940999e62b8de4696e5
Proof 1 · Screenshot5c0ce978df7cc1a44eea2a85406c898082c47114de3e02cfa8c339c77a406d8d