Document accessibility inventory · ADA Title II

Most agencies cannot answer the first question: how many documents are we actually on the hook for?

Recordstake counts them, tests them, and tells you which parts of the answer a machine decided, which parts a person checked, and which parts nobody assessed.

Measured
Judged
Not assessed

Get the free scan

The deadline, stated plainly

26 Apr 2027

public entities of 50,000 people or more
26 Apr 2028

smaller entities and special districts

The technical standard is WCAG 2.1 Level AA. Those are the extended dates — DOJ pushed both back by a year in an interim final rule effective 20 April 2026; the originals were 24 April 2026 and 26 April 2027. If a vendor quotes you the old dates, they have not read the rule since it changed.

The rule does not cover everything you have ever published. It excepts archived material meeting four specific conditions, and preexisting documents that are not currently used to apply for, gain access to, or participate in a service. Which of your documents those exceptions remove is the single biggest factor in what this costs you — and it is the question we start with, because nobody can price the work without it.

An exception is not a permanent pass. Even where content is excepted, an entity still owes effective communication under the ADA — if a person asks for an excepted document in a form they can use, it generally has to be provided. Out of scope means "not required proactively", not "never has to be readable". Any vendor who tells you otherwise is selling you a smaller number, not a smaller obligation.

We are not attorneys and this is not legal advice. We produce evidence and a recommendation you can hand to your counsel, who decides.

What you get first, at no charge: the free scan

We read what your website already publishes, using a labelled crawler at one request per second that honours your robots.txt. You send us nothing, and we ask for no login, no credentials and no access to your systems. Then you get a report that says:

How many documents you publish, how many pages that is, and what kinds they are.
How many of them a validator finds faults in, and which kinds of fault, counted across the library.
What share of your sheets are photographs of pages rather than text — usually the signed ones, and the expensive case.
How much of your library the exceptions plausibly remove, how much they plausibly do not, and how much is undetermined — as shares of the whole, including an honest count of what nobody can settle from outside.

That is the size and shape of your problem, measured on your own library rather than estimated from someone else's. There is no obligation attached and no meeting required to receive it.

Request the free scan by email
One email with your agency's website address is enough. We reply with a link to your report — nothing to install, nothing to sign.

What the free scan does not do is name the documents. Which specific files carry which fault, whether each fault is a real barrier or only bookkeeping, and which side of the exception each document falls on: that is the detailed pass, it is paid work, and the scan is what tells you whether it is worth buying.

Every statement we make sits in one of three buckets

This is the part that makes the rest worth reading. Any claim about any document is labelled as exactly one of these, and never blurred between them.

Measured: An instrument produced this — a named tool, at a named version, on a named date. Reproducible by you or by anyone else. It settles what is in the file, and nothing about whether that matters to someone trying to read it.
Judged: A person examined it, against a stated question, on a stated sample. We say how many, not just that we looked. No tool decides whether a failure is a real barrier or only bookkeeping.
Not assessed: Nobody looked, or no method could settle it. Stated as plainly as the other two.

Automated tools decide only part of this standard. Published estimates put automated coverage at roughly 30–40% of accessibility problems, and our own reading of the Matterhorn Protocol — the industry checklist for PDF/UA — puts about a third of its failure conditions beyond any machine's reach.

Whether a heading is really a heading, whether alt text describes the picture, whether the reading order makes sense aloud: no validator decides those. Any vendor whose report has no "not assessed" section is not telling you what they did not check.

Where the real cost hides

Measured across the entire published library of one mid-sized American city — roughly 2,700 documents and more than 60,000 sheets, every sheet opened and examined. Figures are rounded and the agency is not named. We have written to them; until they choose to be named, publishing a named audit of someone would be discourteous. Nothing here is rounded in a direction that flatters us.

8%

of sheets are pictures of text, not text
1 in 5

documents holds at least one such sheet
~50%

of budget documents contain a scanned page
900,000+

words recovered from those sheets by recognition

Those scanned sheets are overwhelmingly the ones that carry a signature — resolutions, certifications, executed agreements. They are the legally operative pages, and they are the ones a screen reader cannot read at all. They are also where text recognition is least reliable: on that corpus our recogniser misread a city's own motto, and a signature page's own heading, while reporting complete confidence in both. That is exactly why a person has to check recognised text, and why we say so rather than quietly shipping it.

Read the full findings — every figure with its method and its caveat, the two places we were wrong, and what was not assessed at all.

Four questions that make your answer much better

When we inventory a library from outside, roughly three quarters of it comes back undetermined — not because the analysis is weak, but because whether a document is still used to reach a service is a fact about your organisation, not about the file. A crawler cannot see it. You can.

These four answers close most of that gap. None of them is required, none of them needs preparing, and each is usually a five-minute conversation with somebody who already knows.

Which parts of your website do you consider archived — kept for reference or recordkeeping rather than for people to use today?

The archived-content exception asks for material in a dedicated area clearly identified as archived. Where you have labelled a section yourself, that is your evidence rather than our inference. This one usually settles the most. Ask: whoever maintains the website day to day.
Can you share page-view figures for your documents over the last year or two?

The exception turns on whether a document is currently used to apply for, gain access to, or participate in a service. Views are the closest thing to a direct answer, and you already hold them. A file nobody has opened in two years is a strong candidate. Ask: whoever administers your analytics.
Does your records-retention schedule say which of these materials are kept for recordkeeping?

It uses almost the same words the regulation does, and it is an adopted document rather than a judgement made for this exercise — which means it carries more weight with your own counsel than anything we could produce. Ask: the city recorder, clerk or records manager.
Which documents do residents actually ask for at the counter or by phone?

A document people request is being used to reach a service, whatever section it sits in and however old it is. This catches the ones that look archived and are not, and it is the answer no crawler will ever find. Ask: front-counter and permit-desk staff.

None of this is required. You will get the free scan either way, and it will say plainly what it could not determine. Answering any of them narrows that gap — the first one usually narrows it most.

We would rather hand you a report with an honest gap in it than one that quietly guessed and looked tidier.

What it costs

Quoted per engagement

for remediation work, after the scan tells us what is actually there. We do not publish a per-page rate, because the work is not per-page: rebuilding a document's structure, transcribing a map's legend, and producing a companion version of a signed record are three different jobs at three different costs. A rate that charged the same for a page of prose and a page carrying a chart would be a number we could not stand behind. The scan that comes before it is free.

$2,000

for the full Document Inventory: a fresh crawl of your public document library, tested, with a person's review before release, delivered within five business days of your card payment or your purchase order. Paid by card, or against your agency's purchase order invoiced net 30 if you write to us first. For libraries up to 5,000 documents; larger agencies do the same.

Fully credited against remediation. If you go on to remediation work with us, the $2,000 comes off that engagement in full — the Inventory is the scoping document for the work, and you do not pay for it twice.

Buy the Inventory

We publish that number because almost nobody else in this field will. We looked for transacted public-sector prices in cooperative contract rate cards, state contract databases, county bid tabulations and council agenda packets, and found none — the established vendors quote per file, on request, after reviewing your documents.

We can publish a price because we are specific about what we will not do.

What we will not do

We will not tell you that you are covered by the law. That is a legal conclusion, we are not qualified to reach it, and a vendor in this category was fined $1,000,000 by the FTC in 2025 for making claims of that shape.
We will not report a document as fine because a validator found nothing. Absence of a detected fault is not evidence of quality, and we have found rules that pass because the thing they test for is missing entirely.
We will not take on work a machine cannot finish and quietly do it badly. Where a document needs human hands, the report says so and prices it separately — or says we are not the right people for it.
We will not pretend our own limits away. On the one library we have measured end to end — 2,691 documents — our automated pass refused 376 of them and said why: 359 were scanned images with no text to work from, 9 the tagger declined, and 8 were encrypted. Figures drawn as vector artwork, and documents needing their structure rebuilt rather than repaired, are also beyond what we can currently do without a person. We would rather you read that here than discover it after paying us.

Who this is

Recordstake is built and sold by Northline Standard LLC, in West Jordan, Utah. Northline Standard is a small operation, deliberately — which is why the method is written down and the tooling is repeatable rather than depending on anyone's memory.

Our own website is audited against the same standard we report on, and that audit is published here, including the parts we have not assessed. It would be strange to ask you to accept a report we would not publish about ourselves.

Getting the free scan

Email contact@recordstake.com with the name of your agency and the address of your website. If we have already sent you one, reply to it — the same person reads both.

Email for your free scan