Concept redesign · Federal · UI/UX
AAD — Rethinking Access to Digital Archives
A self-initiated redesign of Access to Archival Databases, the National Archives service that provides record-level access to structured federal archival data. Its strength is the depth of the archive; its problem was that the depth was only reachable by people who already understood how the archive is organized. Rebuilt around researcher intent rather than archival structure, and designed to Section 508 / WCAG 2.1 AA.
Self-initiated concept. Not affiliated with, commissioned by, or endorsed by the National Archives and Records Administration. Created to demonstrate accessible design capability for federal digital services. Record counts shown in the mockups reflect the live service; all other content is illustrative.
01 / The challenge
A powerful archive built around its data structure, not the research journey.
AAD serves two audiences from one entry point. Someone looking for a family member, a military record, a place, or a historical event arrives with a name and a rough date. An experienced researcher arrives knowing the series, the record group, and the field they need. The legacy interface was built for the second person, and asked the first to become them before searching.
An audit across 7 page types identified 21 usability and accessibility issues, running from discovery and search through results, system feedback, and mobile access.
| Area | What we observed | Impact on researchers |
|---|---|---|
| Archive-first architecture | The experience is organized around series, subjects, file units, record groups, and database fields rather than common research intents. | You have to translate your question into archival terminology before you can begin. |
| Competing entry points | Search, Advanced Search, Browse by Category, Show All Series, Browse by Subjects, Highlights, and Most Popular all carry similar prominence — more than 20 browse links and subcategories at the first level. | No obvious starting point for anyone without archival expertise. |
| Uncertain system states | The interface doesn't consistently distinguish no results from invalid input, unsupported actions, or system limits. Empty queries can return “No records found”; sorting can fail only after activation; downloads can end without a recovery path. | Unclear feedback reduces confidence in the results themselves. |
| Unlabelled search controls | Fielded-search pages carry no programmatic labels at all — every sampled series page contains zero label elements, leaving 22 fielded controls unlabelled on the audited page. | A screen reader announces an edit field with no indication of what belongs in it. |
| No landmarks or skip navigation | Pages declare no semantic landmark regions and offer no route past the repeated header and category blocks. | Assistive technology has no structure to navigate by. |
| Keyboard-inaccessible interactions | Key interactions are JavaScript-driven without native semantics, and the fixed desktop layout renders against a 980px virtual viewport on a phone. | Some paths can't be completed by keyboard; mobile requires zooming and horizontal panning. |
The structural
findings above were reproduced against the live service in August 2026: no viewport
declaration, no landmark regions, and zero label elements across ten sampled
fielded-search pages.
The shift
Before & after.
The same entry point, in the same frame: drag the handle to wipe between the redesigned service on the left and the legacy one on the right. The legacy page opens on a category grid of roughly forty links, two search boxes, and a “What's New” feed. The redesign opens on one field, a plain statement of what the archive holds, and four ways in named for what people are looking for.
02 / Our approach
Move complexity from user decisions into system behavior.
Every question the interface asks is work it has handed to the researcher. The redesign is built on intent-first discovery, intelligent defaults, progressive disclosure, and clearer system feedback — four ways of moving that work back across the line.
Design decision 01 — Progressive precision
From expert-first forms to the right complexity at the right moment.
AAD's advanced capabilities are genuinely valuable to experienced researchers, so simplifying the product could not mean removing them. Instead we separated starting a search from refining one. A basic search begins with a single familiar input. Advanced Search stays immediately available for anyone who needs more control. Field-level filters and archival structures arrive as the research becomes specific enough to need them.
- 01Start with what you know
- 02Review what the system finds
- 03Narrow when necessary
- 04Use archival fields for precision
The result is not fewer capabilities. It is better timing for complexity.
The basic field accepts a name, place, organization, or keyword — what a researcher already has — instead of asking which archival field it belongs in.
Advanced sits inside the same control rather than behind a separate page, so expert users never pass through a simplified path they didn't need.
Series, record groups, and field-level queries surface once the research is specific enough for them to help, rather than as the price of entry.
Design decision 02 — Defaults over settings
If almost everyone needs it, it shouldn't be a preference.
An early concept introduced a control to ignore accents and diacritics while searching. We removed it. A researcher should not have to understand text normalization to find a historical name. Normalization became specified default behavior instead, so Müller and Muller are interpreted consistently without another decision appearing in the interface. The same principle governed the rest of the redesign: don't ask people to configure behavior the system can reasonably handle itself.
Design decision 03 — System states as part of the research
From dead ends to predictable feedback.
A research tool has to communicate not only results but what the system is doing and what the person can do next. The redesigned interaction model prevents empty searches before submission rather than reporting them as failed research, keeps search progress and asynchronous updates in context, states that an action is unavailable before it is activated, uses native patterns for expandable content, and ends errors with a next step rather than a full stop.
“Something didn't work.”
“I understand what happened, and what I can do next.”
An empty query is caught as an input state, not returned as “No records found” — which reads as a fact about the archive rather than about the form.
Search progress and asynchronous updates are announced in place rather than through intermediate pages or interrupting alerts.
Actions that can't be performed say so up front, instead of failing after the researcher has committed to them.
Recovery is part of the message, so a failed download or an unsupported sort doesn't end the research session.
Design decision 04 — Clarify where AAD ends
One archive ecosystem, clearer destinations.
Not every National Archives research task belongs in AAD. Someone searching structured database records and someone looking for a digitized photograph or a scanned original need different tools. Rather than expecting researchers to know that in advance, the redesign says so at the point of need — “Looking for digitized documents and images? Visit the National Archives Catalog.” — turning an organizational boundary between systems into wayfinding.
Design decision 05 — Accessibility includes emotional context
Some people searching casualty records are looking for a relative.
AAD holds sensitive historical material, casualty records among it. The visual system deliberately avoids combat imagery for casualty-related content. Someone arriving at that series may be researching a war; they may equally be looking for a family member who did not come home, and the interface should not decide which of those they are. This is not a compliance requirement. It is the same judgment applied one layer further out.
03 / Why it's compliant
Accessibility changed the interaction model, not just the compliance score.
The audit exposed technical barriers, but most of the resulting decisions improved the experience for everyone — labels that explain a field before anyone interacts with it help the sighted researcher too. Rather than patching each finding, we used them to reconsider the underlying components. Each decision below maps to the success criterion it satisfies.
Every search control carries a persistent visible label explaining what belongs in the field; placeholders are reserved for examples, against a legacy baseline of no labels at all.
Related search fields are grouped semantically rather than presented as a flat sequence of controls, and pages declare the landmark regions the legacy service omits.
Native buttons and links replace ambiguous JavaScript interactions, so every research path is operable from the keyboard by default rather than by added handlers.
Skip navigation and landmark structure give a way past the repeated header and browse blocks — neither of which the legacy pages provide.
Visible focus states are part of every component's definition rather than left to the browser default that a custom style can silently remove.
Disclosure components expose expanded and collapsed state programmatically, so assistive technology reports what a sighted user can see.
Search progress, validation, and asynchronous results are delivered as status messages rather than intermediate pages or alerts that move focus.
Failed searches, unsupported sorts, and interrupted downloads describe a recovery path rather than reporting that something went wrong.
The criterion's floor is 320px with no two-dimensional scrolling. The redesign reflows from 375px and above, replacing a desktop layout rendered against a 980px virtual viewport.
1.4.11 · Non-text Contrast — AA
Core text was designed against the 4.5:1 target and controls and interface states against 3:1, across the warm hero panel as well as the results tables.
Labels, icons, text, structure, and state indicators each preserve meaning independently of color.
Primary interactive controls are designed at 44–48px — beyond the 24px WCAG 2.2 AA floor, and sized for a phone rather than for the minimum that passes.
Base text size for the redesigned system, applied as a constraint rather than decided component by component.
Target size for primary interactive controls, chosen for one-handed use on a phone.
The contrast floor applied across the concept — text against its background, controls and interface states against theirs.
These are the rules the concept was designed against. Conformance is a property of a built system — validating it would come after implementation.
04 / Measured impact
What changed for the person doing the research.
The legacy experience and the redesigned prototype were evaluated on the same research tasks, so both sides of each pair below describe the same job attempted two ways.
The share of attempts that ended with the researcher holding the record they came for, rather than abandoning the search or settling for an adjacent result.
Reaching a useful result without reformulating the query — the clearest signal that the interface accepted the question the researcher actually arrived with.
A 42% reduction. Most of the difference is time that was previously spent working out where to start rather than searching.
How difficult the research felt to the person doing it — the measure that moves when complexity is handled by the system instead of the researcher.
Every search control in the redesigned system carries a persistent programmatic label. The live service still contains zero label elements across the fielded-search pages we sampled.
A fixed desktop layout rendered against a 980px virtual viewport, replaced by responsive reflow verified from 375px up as its own layout.
The two structural figures — unlabelled controls and mobile rendering — were reproduced against the live service in August 2026. The four task figures come from evaluating the prototype against the legacy experience on the same research tasks.
The archive stayed complex. The way into it became simpler.
The goal wasn't to simplify the archive. It was to make archival expertise optional at the moment someone begins searching.