Guide
How to review remediation someone else delivered
Accepting a batch you did not remediate. The four checks that find the common ways an outsourced file passes a validator without being usable.
About 12 minutes · moderate
What you are actually checking for
You did not do this work, so you cannot verify it step by step. What you can do is check for the specific ways remediation goes wrong when someone is paid per document and measured by whether the checker is green.
None of these are accusations of bad faith. They are what a process optimises toward when the acceptance criterion is a validator’s output.
Check 1 — The artifact test, thirty seconds
This is the most valuable check in this guide. Run it on every file in the batch.
Open the file in Acrobat Pro → All tools → Prepare for accessibility → Fix reading order → tick Show Background/Artifact Content at the bottom of the panel.
Everything marked as an artifact is now overlaid in grey. Look at what is grey.
Should be grey: running headers and footers, page numbers, decorative rules, background tints, printer’s marks.
Should not be grey: body text, headings, table content, charts, signature blocks, and any watermark that says something — “DRAFT”, “SUPERSEDED”, “CONFIDENTIAL”.
Body text in grey is over-artifacting: content declared decorative to make it disappear from the checker. It produces a document that passes every automated test and is unreadable, and it is worse than the untagged original because it now looks remediated.
Check 2 — The two-keystroke test, ten seconds
Select all, copy, paste into a plain-text editor.
Copy order follows structure order in a tagged document, so the paste is a rough transcript of what a screen reader will encounter and in what order.
- Jumbled → the reading order was not fixed.
- Missing content → it was artifacted or never tagged.
- Hebrew or Arabic mirrored → the visual-order problem was not detected, which is unsurprising because no mainstream tool detects it. See TG-RTL-001.
Run this on everything too. It costs nothing.
Check 3 — Read the alt text, not the alt-text count
Every checker reports whether alt text is present. None report whether it is useful, and this is where per-document pricing shows up most clearly.
Open the Tags panel, find the figures, read what is there. Or faster, run a script over the batch:
# Every /Alt string in the file
qpdf --qdf --object-streams=disable in.pdf - | grep -a -o '/Alt ([^)]*)'Red flags: “image”, “graphic”, “photo”, “chart”, a filename, the figure number alone (“Figure 3”), the same string on every image in the document, or a description of the image’s appearance where the image carries data.
What good looks like: the information the image carries. For a chart, the finding. For a photograph in a report, what it shows and why it is there. For a decorative flourish, no alt text at all — it should be an artifact.
An hour reading alt text across a batch tells you more about a vendor’s work than any report they will send you.
Check 4 — Listen to a sample
Ten minutes per sampled document with NVDA or VoiceOver. Use the testing script rather than reading the whole thing.
Sample by type, not by count: one of each document shape in the batch, plus everything with tables, forms or multiple columns. A systematic error — say, every two-column page tagged left to right — survives random sampling and does not survive sampling by type.
Reading the report they sent
If they sent a remediation report, three questions decide how much weight it carries.
Does it record human decisions with justifications? Look for the judgement calls — reading order, alt text, table classification, what was accepted as-is. A report with no decisions in it documented only the machine-decidable half. See what a deliverable should contain.
Does the “what remains” section have anything in it? An empty one usually means nobody looked rather than that nothing remains. Legitimate remainders exist in most real documents.
Does it hash both files? Without a SHA-256 of source and output, the report is about a filename, and you cannot demonstrate later that the file you hold is the file it describes.
What to do with what you find
Over-artifacting — reject and return. It is not a partial result; the document is less usable than before.
Poor alt text — return with specifics. “Figures 2, 4 and 7 describe the image rather than its content” is actionable. “The alt text is weak” is not.
Reading order — return with the pages. This is the expensive one to fix and the one most likely to be argued about, which is exactly why naming the pages matters.
No report — ask for one before accepting. If it was not in the contract, put it in the next one. It is the only part of the deliverable that keeps its value.
Set the acceptance criteria first, next time
The reason this review is difficult is that “make these accessible” is not a specification. Three sentences in the contract change what you get:
- The standard and level: PDF/UA-1, WCAG 2.1 Level AA.
- A remediation report per document, with human decisions recorded, justifications, and the SHA-256 of both files.
- A statement that content may not be artifacted to achieve a pass, and that the rendered appearance must be unchanged.
That third one is the sentence that prevents check 1 from ever finding anything.
Frequently asked questions
The vendor sent a checker report showing zero errors. Is that enough?
It is evidence about the 87 machine-decidable Matterhorn conditions and silent about the 47 that need a human. Those 47 include reading order, alt text quality, heading correctness and whether a table is really a table — which are the failures a reader meets first. A zero-error report is the beginning of the review.
What is the single most common way a remediated file is bad?
Over-artifacting. Content that was awkward to tag is marked as decorative instead, which removes it from the reading order and from the validator's view at the same time. The file passes every automated check and a screen-reader user cannot read half of it. It takes about thirty seconds to detect and almost nobody looks.
How much of a batch should I review?
All of it for the two-keystroke test and the artifact check, which are fast enough to run on everything. A deeper pass on a sample: every document type in the batch, plus anything with tables, forms or multiple columns. Sampling by count rather than by type is how a systematic error survives review.
What if I find problems after I have paid?
That is what the remediation report is for, which is why it should be a contractual deliverable rather than a courtesy. A report that records which decisions were made and by whom lets you show that a specific decision was wrong. Forty PDFs and no report leaves you arguing about taste.
Related failure conditions
- 01-005 Content is neither marked as Artifact nor tagged as real content Every text-showing, image-painting and path-painting operator on a page must sit either inside a marked-content sequence that maps to the structure tree, or inside an artifact sequence. Content that is in neither is orphaned.
- 09-001 Tags are not in logical reading order The order of elements in the structure tree does not match the order a human would read the page. This is the single most common substantive defect in tagged PDFs and the one Acrobat's own tooling handles worst.
- 13-004 Figure tag alternative or replacement text missing A Figure element has neither /Alt nor /ActualText. Every Figure must have one or the other; if the image is decorative it should be an artifact instead of a Figure.
- 18-001 Headers and footers are not marked as pagination artifacts Running heads, footers and page numbers are tagged as real content, so they repeat in the reading order on every page.
- 15-003 A TH cell does not contain a Scope attribute Header cells must declare whether they head a row, a column, or both, unless the table instead uses explicit /Headers ID references on every data cell.
Check your own file. Taggart's validator is free and unlimited — no page cap, no watermark, and no account at all up to 60 pages.
Last updated .