Document Inventory and Triage: Two Gates, Then Rank

Nobody remediates everything. Pretending otherwise is how programs die.

At some point in your first few months, somebody is going to hand you a number. Eleven thousand PDFs. Forty thousand. Whatever it is, it's going to feel like the whole job just became impossible, and I want to tell you right now: that number is a lie. Not because it's wrong — it's probably accurate — but because it's not the number that matters. The number that matters is a lot smaller, and getting there is the actual work.

Here's how you get there.

Gate one: should this even be a document?

Before anybody touches a remediation tool, ask a dumber question first: does this thing need to exist at all?

Retire whatever's superseded or nobody's opened in two years. Nobody wants to hear "just delete it," but I promise you the fastest, cheapest remediation available to you is deletion, and it's sitting right there, unglamorous and free. Replace with an actual web page whatever never needed to be a PDF in the first place — most "documents" on a government site are a form or a policy that somebody printed to PDF out of habit, not necessity. And where the source file still exists, recreate from it instead of remediating the output. Fixing a scanned PDF by hand is always, always more expensive than going back to whatever made it and exporting it properly the second time.

In a typical document estate, this gate alone clears out a real chunk of the total before you've spent a dollar on remediation. And here's the thing worth saying twice: these decisions are permanent. Once you've retired something or rebuilt it as a page, it's off your list for good. That's not a shortcut. That's the whole point.

Gate two: does an exception actually apply?

The federal rule (opens in new window) exempts archived content and certain preexisting documents. Some states carve out their own version of the same idea. Both exist, and both get abused constantly, so here's the one rule that keeps you honest: test exceptions item by item, never by folder sweep. "Everything in this directory is archived" is not a determination. It's a hope. Write down who made the call, what they decided, and why, for each item — not the folder.

And remember what an exception actually buys you: it removes the default obligation to have already fixed it. It does not remove your obligation to provide that content accessibly the moment one specific person asks. Those are different promises. Confusing them is how coordinators end up explaining, later, to someone with a lot more leverage than they have, why an exception they claimed didn't actually cover the thing a resident needed.

Then, and only then, you rank what's left

Whatever survives both gates is your real number — smaller than the one somebody scared you with, and this is the part where you stop panicking and start working. Rank the survivors by what actually matters, weighted by what it costs to fix:

Highest traffic first. How many people actually touch this thing, counting navigation elements and templates that repeat across the whole site — a broken template outranks a single broken page every time, because you're not fixing one thing, you're fixing every page that inherits from it.

Highest stakes first. Does this document gate someone applying for something, accessing something, or participating in something? A form that decides whether someone gets a permit outranks a historical newsletter from 2019, full stop.

Highest complaint volume first, divided by effort. If it's cheap to fix and people keep hitting it, that's your best return on an hour of work you'll ever get.

One thing overrides all of that math, every time: a named complaint. If a real person told you a specific document is a barrier for them, that document goes to the top of the queue regardless of what the scoring says. I don't care how the spreadsheet ranks it. The person in front of you outranks the spreadsheet. Always.

And one category skips your queue entirely: system-generated documents. Receipts, auto-generated notices, anything a vendor platform spits out on its own. Those aren't an editing job — they're a code change, and they go to your vendor, not your triage list. Don't burn your own hours hand-fixing something the vendor needs to fix at the source, or you'll be fixing the same category of document forever, one instance at a time, while the platform keeps generating new broken ones behind you.

What the ranking is actually weighing

If you want it as a formula instead of a paragraph: criticality (does it gate applying, accessing, or participating — read from your own service map, not a guess), reach (how many people touch it, weighted for templates and navigation), legal posture (is there a named complaint or accommodation request attached — this is why your intake log exists), substitutability (is there already an accessible way to get the same outcome today), severity (how badly does it actually fail — unreadable is not the same as merely untidy, and that distinction needs a human looking at it, not a scanner count), longevity (will this even still be in service next year — check before you spend money on something you're about to retire anyway), and effort as the divisor underneath all of it — recreate from source, light repair, or full remediation, using published government effort estimates, not whatever number a vendor's sales deck offered you.

The honest version of this chapter

Nobody remediates everything, and the coordinators who burn out fastest are the ones who tried to. The defensible position was never "we fixed it all." It's a dated, prioritized queue that a stranger could pick up and audit, showing you knew what you had, decided reasonably what mattered most, and were working the list in an order you could explain out loud, measured against real WCAG 2.1 AA (opens in new window) criteria rather than a gut feeling. That's not a lower bar than perfection. It's the actual bar, and it's one you can clear starting this week with the pile you already have, not the one somebody scared you with on day one.

This isn't legal advice on what your specific exceptions cover — verify your own document estate against your counsel's reading of the rule. If you want a fast first pass on where a given page or document actually stands before you triage it by hand, ask Luke to audit it — it won't replace the manual review chapter 4 of this series walks through, but it's a reasonable place to start a queue. See also our document remediation research and methodology for how we verify claims like these.

Previous: the law behind the job · Next: procurement language that actually works

---

Sources: 28 CFR 35.201 — Exceptions (opens in new window) · Web Content Accessibility Guidelines (WCAG) 2.1 (opens in new window) · DOJ guidance on web accessibility and the ADA (opens in new window) · U.S. Access Board — Revised 508 Standards (opens in new window)