PDF accessibility remediation — Sample: a county document library (140 documents, 1,803 pages)

Machine-checkable PDF/UA-1 defects before and after automated remediation. Every figure on this page comes from veraPDF, the reference open-source validator — not from our own checker.

Documents

140

all processed

Pages

1,803

5611.1s total

Failures before

732,765

veraPDF PDF/UA-1

Failures after

25,526

96.5% resolved; 96.6% excluding OCR documents

Fully conformant

40/140

zero remaining failures

Text recovered

630,797

characters via OCR on 33 doc(s)

Left untouched

none made worse

Alt-text cost

$2.72

$0.0015/page

Per document

Before After Bars share one scale; counts are labelled.

Document	Failed checks	Resolved / recovered	Elements	Per page	Conformant
doc92.pdf1 pages · full mode	200,994 2	100.0%	8,569	149.440s	no
doc91.pdf1 pages · full mode	192,609 2	100.0%	6,681	148.300s	no
doc191.pdf385 pages · rebuilt mode	117,795 20,833	82.3%	460	3.209s	no
doc135.pdf1 pages · full mode	27,639 2	100.0%	2,452	24.800s	no
doc139.pdf1 pages · full mode	27,639 2	100.0%	2,452	24.170s	no
doc159.pdf1 pages · full mode	27,639 2	100.0%	2,452	16.730s	no
doc289.pdf206 pages · rebuilt mode	27,310 1,209	95.6%	2,102	0.477s	no
doc290.pdf120 pages · rebuilt mode	19,604 269	98.6%	3,417	0.987s	no
doc154.pdf85 pages · full mode	17,954 2	100.0%	5,760	0.070s	no
doc93.pdf1 pages · full mode	16,736 2	100.0%	528	9.280s	no
doc155.pdf85 pages · full mode	12,807 2	100.0%	4,647	0.049s	no
doc157.pdf64 pages · full mode	9,707 2	100.0%	3,558	0.033s	no
doc153.pdf17 pages · full mode	3,514 2	99.9%	1,091	0.061s	no
doc100.pdf4 pages · rebuilt mode	2,622 115	95.6%	381	0.207s	no
doc168.pdf3 pages · rebuilt mode	2,560 28	98.9%	339	1.530s	no
doc98.pdf4 pages · rebuilt mode	2,319 13	99.4%	316	1.435s	no
doc162.pdf17 pages · metadata mode	2,236 26	98.8%	0	0.115s	no
doc156.pdf12 pages · full mode	2,153 24	98.9%	692	0.168s	no
doc167.pdf2 pages · rebuilt mode	1,641 17	99.0%	197	3.210s	no
doc158.pdf6 pages · full mode	952 2	99.8%	346	0.038s	no
doc101.pdf2 pages · rebuilt mode	913 8	99.1%	88	0.290s	no
doc190.pdf25 pages · metadata mode	808 211	73.9%	0	0.083s	no
doc303.pdf5 pages · full mode	779 4	99.5%	174	0.060s	no
doc102.pdf4 pages · metadata mode	754 27	96.4%	0	0.352s	no
doc108.pdf2 pages · full mode	754 4	99.5%	253	0.545s	no
doc128.pdf2 pages · full mode	754 4	99.5%	253	0.335s	no
doc304.pdf5 pages · full mode	745 4	99.5%	176	0.036s	no
doc99.pdf4 pages · metadata mode	621 26	95.8%	0	1.998s	no
doc149.pdf8 pages · metadata mode	597 592	0.8%	0	0.060s	no
doc117.pdf2 pages · full mode	553 6	98.9%	78	0.200s	no
doc130.pdf2 pages · rebuilt mode	486 15	96.9%	111	2.705s	no
doc181.pdf2 pages · rebuilt mode	486 15	96.9%	111	2.305s	no
doc186.pdf1 pages · rebuilt mode	425 2	99.5%	71	0.070s	no
doc184.pdf3 pages · full mode · OCR	399 3	+725 chars	170	5.840s	no
doc192.pdf4 pages · full mode	384 3	99.2%	52	0.058s	no
doc121.pdf2 pages · rebuilt mode	359 4	98.9%	104	3.290s	no
doc179.pdf2 pages · rebuilt mode	359 4	98.9%	104	2.510s	no
doc123.pdf1 pages · full mode	321 0	100.0%	27	8.110s	yes
doc182.pdf1 pages · full mode	321 0	100.0%	27	5.420s	yes
doc137.pdf1 pages · full mode	290 4	98.6%	23	0.100s	no

Showing the 40 documents with the most failures. The other 100 went from 5,227 to 2,034 failed checks (0 of them started clean). Every total on this page covers all 140 documents.

What is still failing

Rule	Description	Checks
`7.21.8-1`	A PDF/UA compliant document shall not contain a reference to the .notdef glyph from any of the text showing operators, regardless of text rendering mode, in any content stream	20,683
`7.1-3`	Content shall be marked as Artifact or tagged as real content	1,200
`7.21.7-1`	The Font dictionary of all fonts shall define the map of all used character codes to Unicode values, either via a ToUnicode entry, or other mechanisms as defined in ISO 14289-1, 7.21.7	857
`7.3-1`	Figure tags shall include an alternative representation or replacement text that represents the contents marked with the Figure tag as noted in ISO 32000-1:2008, 14.7.2, Table 323	582
`7.21.4.1-1`	The font programs for all fonts used for rendering within a conforming file shall be embedded within that file, as defined in ISO 32000-1:2008, 9.9	479
`7.2-10`	TR element may contain only TH and TD elements	369
`7.1-1`	Content marked as Artifact should not be present inside tagged content	361
`7.1-2`	Tagged content should not be present inside content marked as Artifact	361
`7.21.4.2-2`	If the FontDescriptor dictionary of an embedded CID font contains a CIDSet stream, then it shall identify all CIDs which are present in the font program, regardless of whether a CID in the font is referenced or used by the PDF or not	144
`5-1`	The PDF/UA version and conformance level of a file shall be specified using the PDF/UA Identification extension schema	100
`7.1-5`	All non-standard structure types shall be mapped to the nearest functionally equivalent standard type, as defined in ISO 32000-1:2008, 14.8.4, in the role map dictionary of the structure tree root. This mapping may be indirect; within the role map a non-standard type can map directly to another non-standard type, but eventually the mapping shall terminate at a standard type	82
`7.18.5-1`	Links shall be tagged according to ISO 32000-1:2008, 14.8.4.4.2, Link Element	54

Text recovered by OCR

These documents were images with no text layer — unreadable to a screen reader, unsearchable, impossible to copy from. OCR added 630,797 characters of real text across 33 document(s). Expect the failure count to rise on some of them: a blank scan has almost nothing to fail, and readable text has much more. That is an improvement, not a regression. OCR is a transcription and makes mistakes — spot-check anything legally operative.

Document	Characters	OCR time	Failed checks
doc185.pdf	129,049	39.9s	84 → 81
doc193.pdf	129,049	70.3s	84 → 81
doc287.pdf	91,295	71.1s	53 → 0
doc305.pdf	91,295	33.6s	53 → 0
doc360.pdf	54,135	25.6s	117 → 30
doc366.pdf	16,446	10.5s	14 → 0
doc358.pdf	15,154	11.1s	12 → 0
doc363.pdf	13,518	14.7s	12 → 0
doc351.pdf	11,498	9.3s	12 → 0
doc132.pdf	9,641	9.3s	7 → 0
doc115.pdf	7,184	11.9s	7 → 0
doc364.pdf	5,779	11.8s	7 → 0
doc183.pdf	5,600	8.7s	8 → 0
doc114.pdf	4,977	13.1s	7 → 0
doc120.pdf	4,820	8.5s	7 → 0
doc365.pdf	4,372	6.8s	7 → 0
doc350.pdf	4,260	9.0s	10 → 19
doc343.pdf	4,123	12.0s	11 → 19
doc200.pdf	3,369	8.6s	6 → 0
doc201.pdf	3,224	8.9s	6 → 0

Pages with no text layer

These pages are images of documents. Tagging cannot make them readable, and a validator will not object — a page marked up correctly as a figure passes the structural rules while a screen reader still gets nothing. Any document below needs OCR first; until then, treat a low failure count on it as meaningless.

Document	Scanned pages	Share	Action
doc140.pdf	1 / 1	100%	needs OCR before tagging is meaningful
doc160.pdf	1 / 1	100%	needs OCR before tagging is meaningful
doc95.pdf	1 / 2	50%	needs OCR before tagging is meaningful
doc234.pdf	26 / 76	34%	needs OCR before tagging is meaningful
doc366.pdf	1 / 10	10%	scanned inserts; OCR those pages
doc190.pdf	1 / 25	4%	scanned inserts; OCR those pages
doc141.pdf	1 / 30	3%	scanned inserts; OCR those pages
doc289.pdf	5 / 206	2%	scanned inserts; OCR those pages
doc191.pdf	7 / 385	2%	scanned inserts; OCR those pages
doc290.pdf	2 / 120	2%	scanned inserts; OCR those pages

Needs human review

Automated tagging is a first pass. These documents are not ready to publish until a person has confirmed reading order, heading levels, table headers, and every generated alt-text string. Nothing here should be described to an auditor as certified compliant.

doc100.pdf — p1: Heading level changed from H3 to H1 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('DECLARATION SCHEDULE')
doc100.pdf — p1: Heading level changed from H3 to H1 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('INSTRUCTION SHEET FOR FILING ATD 39 and/or 40')
doc100.pdf — p1: Heading level changed from H3 to H1 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('GENERAL INFORMATION')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('DECLARATION SCHEDULE')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('NAICS CODE:___________________')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Operator ID #')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Contact Person')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('API Number')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Unit Name')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Group Number')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Field Name')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Basin')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Well Type')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Well Status')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Oil and / or Gas Field Equipment Description')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Model')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Units or Footage')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Size')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Pressure')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Condition')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Year')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Installed')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Capacity')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Installed')
doc100.pdf — p2: Heading level changed from H3 to H2 so the outline is sequential and starts at H1 (WCAG 1.3.1 disallows skipped levels). Confirm this matches the document's intended hierarchy. ('Cost')

…and 2655 more.

Known limitations

1 content stream(s) were only partly marked by their producer. Existing marked content was adopted rather than rewritten, so its tag semantics come from the original file and were not reviewed.
1 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
1 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
10 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
11 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
110 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
116 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
119 content stream(s) were only partly marked by their producer. Existing marked content was adopted rather than rewritten, so its tag semantics come from the original file and were not reviewed.
12 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
120 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
13 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
16 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
166 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
19 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
2 content stream(s) were only partly marked by their producer. Existing marked content was adopted rather than rewritten, so its tag semantics come from the original file and were not reviewed.
2 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
20 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
206 content stream(s) were only partly marked by their producer. Existing marked content was adopted rather than rewritten, so its tag semantics come from the original file and were not reviewed.
21 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
22 content stream(s) were only partly marked by their producer. Existing marked content was adopted rather than rewritten, so its tag semantics come from the original file and were not reviewed.
22 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
23 figure(s) carry alt text that is an internal code rather than a description (for example "P1TB1#y1"), written by whatever produced the original file. These pass the validator but tell a screen reader user nothing, and need a person to describe them.
258 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
26 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
3 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
3 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
383 content stream(s) were only partly marked by their producer. Existing marked content was adopted rather than rewritten, so its tag semantics come from the original file and were not reviewed.
4 content stream(s) were only partly marked by their producer. Existing marked content was adopted rather than rewritten, so its tag semantics come from the original file and were not reviewed.
4 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
4 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
40 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
45 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
49 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
5 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
5 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
587 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
6 content stream(s) were only partly marked by their producer. Existing marked content was adopted rather than rewritten, so its tag semantics come from the original file and were not reviewed.
6 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
60 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
61 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
635 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
640 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
7 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
7 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
8 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
80 content stream(s) were only partly marked by their producer. Existing marked content was adopted rather than rewritten, so its tag semantics come from the original file and were not reviewed.
851 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
855 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
866 text run(s) could not be matched to a layout block and were marked as artifacts, so they will not be read aloud. Verify none of them carry meaning.
9 table region(s) were detected but their rows disagreed on column count, so their content is presented as text rather than as a table. A screen reader announcing wrong row/column positions is worse than none, but these need manual table tagging to be properly navigable.
Input was already tagged, so the existing structure tree was preserved and only document-level defects were fixed. Reading order and tag semantics were not reviewed.
The failure count rose against the original because OCR added a text layer: a blank scan has almost nothing to fail, readable text has much more. Measured against the OCR'd file, this document improved.