Page by page
Every page has a clear boundary and a stable PDF page number, even when the printed page label differs.
Turn reports into
structured text.
Ready for LLM pipelinesPage-level traceability
Overview
Sustainability reports are hard to work with: dense text, complex tables, charts and images, all locked inside PDFs. We convert them into text you can search, parse and feed to a model, and we keep track of which page everything came from. The original PDF comes along, so you can always check.
Typed content
Headings, tables, figures, captions and footnotes are labelled, so your code can treat each one differently.
Source included
Every order pairs the Markdown with the original PDF and a 50-field metadata record.
Explore a report
Choose a page to compare the source PDF with the Markdown produced by our pipeline.
HMC Capital · 2022 Sustainability Report
Previewing PDF pages 1–5 of 52
Original PDF
1 / 5
Processed Markdownpage 1.md
:::: {.page n=1}
# 2022 Sustainability Report
Creating Healthy Communities
::: {#1-i-1 .image}
A group of diverse people standing in a circle outdoors, holding each other's shoulders, looking towards the horizon. The image is bathed in warm, golden sunlight, suggesting a community or team effort.
:::
::::Process
Our conversion process is built specifically for sustainability reports. Every page goes through vision OCR. Every figure then gets a second look from a separate vision model, and every table gets its own correction pass.
Read every page
Our vision OCR process reads each page and identifies text, reading order, tables and visual regions across complex report layouts.
Reprocess the visuals
Each detected figure receives a separate vision-model pass for a useful description and type. Tables receive targeted transcription and correction, preserving their evidence and units where available.
Render and validate
We turn the enriched result into page-addressable Markdown, then check page sequence, typed elements, table retention and source binding. Files that don’t pass are held back for review instead of being delivered.
Good to know: we also use these Markdown files in our own data extraction jobs. Our extraction pipeline achieved 98.7% value accuracy in SRC-Bench v1: 226 of 229 reported values matched expert annotations across seven indicators in 50 reports.
Taxonomy
Every element is labelled so your code can select the content it needs: charts, tables, images or document structure. These classes are part of the file, rather than inferred again each time you use it.
We use Pandoc fenced divs such as ::: {#7-bc-1 .bar-chart}. The class identifies the element type; the ID identifies the first bar chart on PDF page 7. Page containers use :::: {.page n=7 label=5}, where n is the PDF page position and label, when available, is the printed page number.
Charts
.bar-chart- Bar, column and stacked-bar charts.
.pie-chart- Pie, donut and ring charts.
.trend-chart- Line charts and time series.
.scatter-plot- Scatter and bubble plots.
Other visualisations
.materiality-matrix- Materiality grids and stakeholder/impact plots.
.map- Geographic maps and regional breakdowns.
.infographic- Mixed visual and text compositions.
.diagram- Process flows, organisation charts and conceptual diagrams.
.chemical- Chemical structures and molecular diagrams.
.other- Visualisations outside the other categories.
Tables and images
.table- Data tables, with units and year ranges where available.
.image- Photographs, logos, icons and decorative graphics.
Document structure
.page- PDF page containers with a sequential page number.
.toc- The document’s table of contents.
.caption- Figure or image captions.
.footnote- Footnote text, including lists and multiple paragraphs.
File structure and formatting
- Document provenance. YAML frontmatter identifies the report and conversion. Certified V3 files also carry the source OCR, converter and validator versions and hashes. The companion metadata includes the original PDF’s SHA-256.
- Stable references. Visual IDs follow
{page}-{type}-{counter}; the counter resets for each type on each page. Captions, footnotes and the table of contents use their class without a visual ID. Inline images use spans such as[company logo]{#7-i-2 .image}. - Tables. Standard Markdown pipe tables retain units and year ranges as attributes where available. Merged cells carry explicit row/column-span annotations, with blank placeholders for covered cells.
- Text and equations. UTF-8 preserves original-language characters. Headings, lists and links retain their structure; equations use
$…$or$$…$$. Repeated page headers and footers are removed by default, and columns are linearised in reading order. - Tool compatibility. Pandoc can parse the typed structure into an abstract syntax tree (AST). Other Markdown tools may require an extension or conversion to preserve attributes.
Customisation and quality checks
On request, we can retain headers and footers, provide tables as embedded HTML, omit visual tags, or add element geometry and enhancement metadata. Custom transformations are quoted separately.
Certified files pass checks for provenance, page sequence, typed elements, table evidence, duplication, character safety and Pandoc structure. OCR body text is not editorially rewritten; model-generated figure descriptions and difficult table transcriptions remain explicitly typed.
Complex layouts can still introduce errors, particularly in rotated tables, stylised charts and ambiguous reading order. If you find systematic issues, share examples so we can investigate and agree a correction.
Delivery
Structured Markdown
One UTF-8 file per report, with document provenance, page containers and typed content. Standard Markdown is enriched with Pandoc attributes for precise structure.
Original PDFs
The unmodified source reports accompany the conversion so your team can inspect layouts, figures and quoted passages.
50-field metadata
Report and company details, identifiers and filenames connect the Markdown to the right source. See the metadata field guide.
Pricing
Prices per report, in USD. Markdown is an add-on to Bulk Reports.
Bulk Reports price per report, in USD
- Tier
- 1
- Order size
- Up to 1,000
- $0.60
- Markdown (PDF included)*
- $1.20
- How to order
- Download Portal
- Tier
- 2
- Order size
- 1,001–10,000
- $0.45
- Markdown (PDF included)*
- $0.90
- How to order
- Request quote
- Tier
- 3
- Order size
- More than 10,000
- $0.30
- Markdown (PDF included)*
- $0.60
- How to order
- Request quote
| Tier | Order size | Markdown (PDF included)* | How to order | |
|---|---|---|---|---|
| 1 | Up to 1,000 | $0.60 | $1.20 | Download Portal |
| 2 | 1,001–10,000 | $0.45 | $0.90 | Request quote |
| 3 | More than 10,000 | $0.30 | $0.60 | Request quote |
* Markdown can’t be ordered in the Download Portal yet. Contact us and we’ll set it up.
Academic and non-profit: 30% off for eligible academic and non-profit organisations.
Getting started
Request a quote
Build your Bulk Reports enquiry and select Markdown files in the Add-ons step.
Request quote