What the scan checks
The scan runs 33 checks, in four groups. Each one marks something software will miss or misread, and says where the check comes from: a published standard, a known failure in PDF text extraction, or our own rule.
This list is generated from the scanner itself, scan version 5.
Text a machine can't see
11 checksWords that exist only inside images
AI reading the text never sees them.
Large images with no text over them
Anything written inside them is invisible to text extraction.
Pages with no extractable text
Scanned or flattened pages read as blank.
Characters that extract as codes
Words come out garbled, such as (cid:12).
Icons, bullets, or ligatures that extract as meaningless codes
Some drop letters from words.
Ligature characters
Words like 'final' may not match searches.
Text drawn twice in the same place
Common tools extract it doubled.
Words split across line breaks
They extract as two fragments.
Rotated text
It extracts out of reading order.
Very small text
Qualifiers often hide there.
Text outside the visible page
Machines read it; people never see it.
Tables
5 checksTable headers with no text
The column labels are probably images.
Row labels a standard extractor drops
Numbers lose what they count.
Merged or missing table cells
Values can attach to the wrong column.
Checkmarks or icons that won't read as symbols
A benefit matrix loses its marks.
Table rows separated from their headers
AI search tools retrieve rows without labels.
Rules that invite different readings
10 checksRules that point elsewhere for their details
AI given only this guide can't follow them.
Links out of the guide
The rule may live on another page.
Discretion or right-to-change language
AI tends to state a definite answer anyway.
Exception words (unless, except, notwithstanding)
Conditions are easy to drop.
Rules qualified by 'provided that'
The condition is easy to drop.
Vague phrases (as appropriate, timely)
Open to more than one reading.
Open-ended lists (etc., and/or)
The rule's scope is unclear.
Deadlines in days, not saying business or calendar
A deadline can differ by days.
'Within' a period with no start event
It's unclear when the clock starts.
Footnote markers
The qualifying text sits elsewhere.
File setup
7 checksNot tagged for accessibility
No logical structure for tools to follow.
No document language set
Accessibility tools handle it worse.
No document title in the metadata
Tools show a file name instead.
Title not set to display
Viewers show the file name.
Copying text is restricted
Some tools refuse to read it.
Accessibility extraction is blocked
Screen readers and parsers are refused.
No version or effective date
Readers can't tell which rules are current.
Each finding is a fact about how the file reads to software, not a judgment of the program. Thresholds, such as what counts as very small text, are our own settings.