How CampusMind works

In short

CampusMind rebuilds the structure an inaccessible PDF is missing, scores its own work against PDF/UA-1, and hands back the corrected file.

Validated against: PDF/UA-1 (ISO 14289-1) · 106 rules, checked by VeraPDF
Automation resolves: About 80% of common accessibility issues
Always needs a person: Alt text meaning, ambiguous layout, accountability

What's actually missing

An inaccessible PDF usually isn't missing its content — the words and images are there. What's missing is the tag tree: the machine-readable structure that says "this is a level-2 heading", "this is a data table with these headers", "this image means this".

Without it, a PDF is a picture of a document. A screen reader can often pull the text out, but it can't offer heading navigation, announce table cells with their headers, or describe a figure. Remediation is the work of rebuilding that structure.

Three words you'll see everywhere

Term	What it means
Batch	One remediation run — a single file, or every PDF in a course. Everything on the hub is organized by batch, and each gets an ID like `CM-1715` or `Canvas-1711`.
Source	Where a document came from: Canvas, Blackboard, Brightspace, SharePoint, or Manual for anything you upload yourself.
Group	The set of people a batch is shared with, and how access to its documents is controlled. See Set up your team.

The pipeline

Every document takes the same seven steps.

Extraction
An extraction engine reads the page to find the paragraphs, headings, figures and tables. It returns candidates, each with a confidence score.
Thresholds
How confident the engine has to be before an element counts. This is the step behind a document coming out under- or over-tagged.
Strategy
Either a tag tree is written over the page as it stands, or the page is rebuilt. You choose this per batch.
Tagging
Elements become PDF tags — P, H1–H6, Figure, Formula, Table, List, Link, Caption — in a reading order, the sequence a screen reader announces them in. Figures get generated alt text.
Validation
VeraPDF checks the result against PDF/UA-1's 106 rules. The share that pass is the document's compliance score.
Delivery
The corrected file is stored against its batch, ready to Download. For course documents, Replace original file decides whether it's written back over the original — it's off by default.
Revision
Output lands as version v1, attributed to System (autotag). Edits in the Tag & Caption Editor stack on top, and Revalidate re-runs step 5.

Checking the result

Validation runs against the original as well as the remediated file, which is what makes two before-and-after views possible:

Comparison view — the two PDFs side by side, so you can confirm the page still looks right.
Remediation report — original issues against remaining ones across four categories: document structure, headings and titles, tables and cells, and metadata and tagging. Each carries a severity.

A document can score highly and still have findings. Passing a rule set is not the same as being usable.

The two choices that change the output

Everything else has a sensible default. These two don't, because the right answer depends on the document. Upload & remediate covers where they live on screen.

Strategy

Strategy	What it does	Reach for it when
Preserve Layout (Auto-Tag)	Writes a tag tree over the existing layout. Nothing visual changes.	The document must come back looking identical.
Full Accessibility Reconstruction	Everything above, then rebuilds structure — table layouts, generated captions, color contrast, layout corrections.	Tagging alone can't fix it. Page count may change as content reflows.

These are a choice, not a quality setting: some problems can't be fixed without changing the page, and some documents must not change.

Keeping what's already there

Keep existing tags updates a document's current tags instead of replacing them. Use it when a document was tagged by hand and you want corrections rather than a fresh start.

Thresholds

Engine	Range	Defaults	Confidence to
PaddleX	0.1–1.0	Table `0.50`, Formula `0.50`	detect an element
Document Intelligence (default)	0.1–0.99	Paragraphs `0.95`, Figures `0.99`	keep one
Hybrid	Both	All four sliders	both of the above

Raise a threshold and only confident detections survive: fewer false elements, but genuine content gets dropped. Lower it and more is kept, guesses included.

The Document Intelligence defaults are deliberately high. Change them for a document class that keeps coming out wrong, not as a general setting.

Why a person is still required

Meaning. Whether alt text conveys the purpose of an image is a judgment about the document's argument, not its pixels.
Ambiguity. A sidebar might be an aside or the main point. The engine picks the likelier reading; only a reader knows which is right.
Accountability. Under ADA Title II an institution answers for its content. A conformance score is evidence, not a defense.

See What a person decides.

Standards

PDF/UA-1 (ISO 14289-1) — what remediated documents are validated against, and the number behind every score in the product.
WCAG — the standard the product's compliance framing refers to, but does not score. In-product labeling reads WCAG 2.1; see Accessibility standards for how PDF/UA, WCAG, Section 508 and ADA Title II relate.

Next steps

Quickstart — put this to work on a real document.
Upload & remediate — the settings in practice.
Trust & compliance — the obligations behind the product.