Every agency archive contains a file called final_FINAL_v3(2).pdf. Usually several. The cost of that is not embarrassment; it is that nobody can tell what the document is, whether it duplicates another one, when it can be destroyed, or what a screen reader should announce when it opens.
Naming and metadata are the cheapest infrastructure in a records program. They cost nothing but agreement, and they are almost impossible to retrofit at scale.
What a name has to do
- Identify the document without opening it.
- Sort correctly in any list, which in practice means dates first and in ISO order.
- Survive every system it passes through — web server, records system, email attachment, someone's laptop.
- Support machine processing, so retention and publication rules can be applied in bulk.
A convention that works
A pattern that holds up across departments:
YYYY-MM-DD_dept_series_subject_v01.pdf
For example:
2026-08-12_council_packet_regular-meeting.pdf2024-03-01_pw_capital-project_elm-street-resurfacing_v02.pdf2019-06-30_fin_budget_adopted-fy2020.pdf
The rules behind it:
- ISO dates.
2026-08-12, never8-12-26. This is the whole reason files sort chronologically. - Date first when the date identifies the record — meetings, budgets, notices. Department first when the series identifies it and dates are incidental.
- No spaces. Use hyphens within a word group and underscores between groups. Spaces become
%20in URLs and break scripts. - Lowercase only. Some systems are case-sensitive and some are not; picking one avoids a class of broken links.
- No special characters at all — no
&,#,%, commas, parentheses, or apostrophes. - Controlled abbreviations. Publish the list of department and series codes.
pw,fin,cd,hr— agreed once, used everywhere. - Keep it under about 80 characters. Long paths still break things.
- Version explicitly with
v01,v02. Neverfinal.
The file name is not the metadata. It is a human-readable handle. Anything you need to query — retention series, legal hold status, disposition date — belongs in a field somewhere, not encoded into a string. The name should be enough to recognize the file, not enough to run the records program.
Metadata inside the PDF
Every PDF carries its own metadata, and most government PDFs carry it wrong — a title reading Microsoft Word - staffreport.doc, an author field naming an employee who left in 2015, and no declared language.
Four fields matter:
| Field | What to put in it | Why |
|---|---|---|
| Title | The real document title, in plain language | Announced by screen readers, shown in the browser tab, used by search engines |
| Author | The agency or department, not a person | Staff leave; personal names in metadata are a small privacy leak |
| Subject / Keywords | Series and topic terms | Improves internal and external search |
| Language | The document language, e.g. en-US | Determines how assistive technology pronounces the text |
The title is not cosmetic. Accessibility standards require a PDF to have a document title and to be configured to display it rather than the file name. A file called 2026-08-12_council_packet.pdf whose embedded title says Microsoft Word - Document1 fails on that point regardless of how good the rest of the tagging is. See the WCAG checklist for government PDFs.
Records metadata
Alongside the descriptive fields, a records program needs five values it can query:
- Record series code — links the file to a row in the retention schedule
- Retention trigger date — the event the clock runs from, which is often not the creation date
- Computed disposition date — trigger plus retention, stored so it can be reported on
- Legal hold flag — suspends disposition regardless of date
- Content hash — SHA-256, which is how duplicates are identified across systems
With those five present, "produce a list of everything eligible for destruction this quarter, excluding holds" becomes a query rather than a project. Without them, it stays a project forever. See retention schedules.
Renaming what already exists
The convention is easy for new documents. The archive is the hard part, and the risk is breaking links that appear in staff reports, other agencies' records, and search results.
- Inventory first, capturing current URL, current name, and content hash for every file.
- Deduplicate by hash. Renaming five copies of one document is wasted work; consolidate to one and redirect the rest.
- Rename in place, one department at a time. Do not attempt a site-wide rename in a single change.
- Add a permanent redirect from every old URL. Non-negotiable. A 301 from the old path costs nothing and preserves every citation and every accumulated search ranking.
- Fix embedded titles in the same pass. You are already touching every file; setting the PDF title at the same time turns two projects into one.
- Update the sitemap and re-submit it so search engines re-crawl the new paths quickly.
Do it during the remediation pass. If you are already reprocessing every document for accessibility, that is the moment to set titles, languages, and names correctly. Handling naming, metadata, and tagging as three separate projects means opening every file in the archive three times.