I led the design and execution of a multi-phase, mixed-methods research program to validate and operationalize the Social Security Administration's design system after years of untested component growth during COVID.
The SSA Design System is the shared foundation for 60+ public-facing and internal applications supporting millions of Americans navigating Retirement, Disability, SSI, and Survivors benefits. When COVID-19 halted in-person usability testing, components continued to ship without formal validation — and the organization had no infrastructure for remote testing at scale.
By 2024, this had created more than a backlog. It was a systemic governance failure: teams were building high-stakes federal services on patterns that had never been validated with real users. Accessibility compliance was at risk. Stakeholder confidence had eroded. And the agency's FY25 modernization milestones depended on a design system no one could fully trust.
The challenge wasn't simply to test components. It was to rebuild evidence-based governance from the ground up — in a way that would outlast any single research initiative and expand organizational capacity in the process.
Unvalidated patterns weren't just a UX concern — they represented compliance risk, modernization blockers, and trust failures at national scale. For users navigating disability applications under financial or health stress, even minor friction is a barrier to completion.
The mandate required a research leader who could operate at two levels simultaneously: rigorous enough to produce defensible findings, and strategic enough to translate those findings into governance decisions that would scale across dozens of products and outlast the initiative itself.
No process defined who owned pattern decisions, how findings were escalated, or when patterns required re-validation. Knowledge was tribal. Drift was inevitable.
Patterns in production across 60+ applications had never been validated for WCAG compliance, cognitive load, or assistive technology compatibility.
Sensitive identity fields — gender, pronouns, sex at birth — were deployed without testing with affected communities, risking harm at millions-of-interactions scale.
FY25 mobile-first milestones required validated mobile patterns. Without a cross-platform program, modernization efforts were built on an unstable foundation.
I designed and led SSA's first post-COVID, cross-platform design-system validation program — a structured, multi-phase initiative that paired mixed-methods usability research with durable ResearchOps infrastructure and hands-on practitioner training.
The solution wasn't a study. It was a governance model — one that produced validated patterns, prioritization frameworks, and institutional processes that would continue operating long after the research team moved on.
The result: a COVID-era backlog cleared, 25+ design-system updates implemented, 3 new governance processes launched, and 5 non-UX practitioners enabled to carry validation forward — all in service of 200M+ annual digital interactions.
Structured desktop and mobile studies with 23 participants across 4 device platforms — generating defensible, cross-platform evidence for governance decisions.
Behavioral metrics, qualitative observation, and attitudinal measures triangulated to surface findings that task-success rates alone would have missed.
Standardized protocols, centralized repositories, severity scoring, and biennial re-testing cycles — transforming research from episodic activity to organizational standard.
Every finding translated to component updates, token-level changes, and process enhancements — packaged to drive immediate action, not future consideration.
This question shaped every decision: method selection, participant design, synthesis approach, and how findings were packaged for governance. The answer had to be reusable — not a one-off study.
Findings needed to hold across desktop, iOS, and Android — not just validate a single context.
With 15+ patterns to evaluate, severity × confidence scoring was essential to focus governance decisions.
Task success rates alone would miss the trust and comfort signals critical to high-stakes federal services.
The program had to produce reusable infrastructure — protocols, templates, and trained practitioners — not just a report.
As UX Research Lead and Research Governance owner, I was accountable for both research rigor and organizational sustainability. My scope extended well beyond facilitation — it encompassed program strategy, cross-functional alignment, ResearchOps infrastructure, and team enablement.
I independently designed the end-to-end research strategy, led a team of 5 researchers, and defined how evidence translated into governance decisions affecting 60+ applications. The dual mandate — produce rigorous findings and build lasting capability — shaped every choice.
This problem could not be solved with a single method. Each method was chosen to answer a specific class of governance question — not to produce data for its own sake.
Task success rates, error frequency, path deviation, and recovery behavior — measured per pattern and per platform.
Think-aloud protocols and post-task structured interviews capturing confusion, hesitation, emotional friction, and trust signals.
5-point Likert surveys measuring perceived helpfulness, clarity, and comfort — critical for inclusive identity fields in high-stakes contexts.
Primary Study Aim
Evaluate the usability, accessibility, and emotional trust of 15 core UEF 3.0 design-system patterns across desktop and mobile — surfacing behavioral breakdowns, cognitive friction, and trust signals for users navigating high-stakes federal benefit services. Each pattern is scored using a severity × confidence framework to determine development readiness, with passing patterns advancing to FY25 production. A secondary objective assessed design preference across three Expand/Collapse Accordion variants.
Shared Protocol · Applied to Both Phases
Format
Moderated · One-on-one · 60-minute sessions · Think-aloud protocol throughout
Prototype
6-step "Apply for Disability" application hosted on Axshare Cloud
Session Team
1 facilitator · 1 moderator · 1+ notetaker per session
Metrics
Task success/fail · Error rate · Post-task Likert scale · Qualitative sentiment
Participants
Inclusive of male, female, and nonbinary · Ages 18–65 · Prior interaction with SSA digital services required
Post-Task Assessment
Likert-scale questionnaire via screen share · Semi-structured interview administered as informal Q&A
Task Success Definition
Scenario completion without facilitator intervention · Failures logged by severity for escalation scoring
Analysis & Synthesis
Post-session debriefs consolidated pass/fail scores, Likert results, and qualitative observations into a pattern-level analysis spreadsheet; findings were then mapped against severity × confidence criteria before escalation.
Recruitment & Screening Criteria · Both Phases
Recruitment
Eligibility Criteria
Consent & Privacy
Why this was sufficient: The goal was decision reliability, not statistical generalization. Patterns were escalated only when risks appeared consistently across multiple data sources, or were severe enough that scaling would harm real users. Requiring nonbinary representation in both phases was non-negotiable — the only way to get valid signal on sensitive identity patterns.
Rather than testing patterns in isolation, we embedded all 15 into a realistic 6-step "Apply for Disability" prototype — so we could evaluate how patterns performed under authentic task pressure, not abstract UI review. Participants experienced patterns the way real users would: in sequence, under cognitive load, with real emotional stakes.
Design System Test Prototype
Applying for Disability Journey
The Axure prototype was structured as an end-to-end Disability benefits application journey.
Participants were given a specific scenario: you are a dual citizen living abroad who needs to apply for disability benefits. You go to the SSA website, find the link to apply, and begin.
I chose this scenario intentionally — one of the patterns under test was Address (International), and we needed participants to have a realistic reason to interact with it. A dual citizen living overseas provided that naturally: international address entry wasn't a forced task, it was just part of applying.
The most important finding wasn't a failed pattern. It was the gap between what metrics showed and what users experienced.
High completion rates masked emotional friction, comprehension failures, and trust erosion — the signals that matter most in high-stakes federal services. Without mixed methods, these patterns would have been marked "good enough."I support inclusivity, but this question caught me off guard. I wasn't expecting it here, and it made the experience feel less comfortable.
Study Participant · Gender Identity Field · Desktop Study
This quote captures the core tension of the entire program: a user who values inclusive design still experienced discomfort — not because of the intent, but because of the absence of a graceful exit. Emotional usability isn't about content; it's about control.
Testing surfaced two distinct failure modes across the full pattern set: emotional harm from patterns that were functionally correct but socially uncareful, and behavioral breakdowns where comprehension and error recovery diverged sharply by platform. Both required different governance responses.
Failure Mode — Emotional Trust
Patterns that passed every functional metric but caused emotional friction, coercion, or trust erosion — consistently across both desktop and mobile. High Likert scores masked low emotional safety.
All participants completed the field — but success masked polarized reactions. Some valued flexibility; others were deeply offended by the question's presence. The opt-out ("prefer not to share") felt coercive rather than a genuine choice. Likert helpfulness avg: 4.38/5 desktop, 4.26/5 mobile — strong on clarity, weak on emotional safety.
Despite 90% task completion, 60% of desktop participants reacted negatively. Nonbinary participants were specifically offended by lowercase pronouns in helper text — a grammar error that signaled carelessness. Mobile Likert avg 4.26/5 for helper text clarity suggests the example text itself works; the pattern design and opt-out path are the failure points.
Technically flawless across every device. Likert: 4.78/5 helpfulness, 4.5/5 ease of understanding. The design itself is not the issue — but two desktop participants called it "very controversial" and "ridiculous." The problem is forcing engagement with no graceful exit. A targeted fix — not a redesign.
Failure Mode — Behavioral Breakdown
Patterns where task completion rates were high but platform-specific interaction behaviors — hover vs. tap, layout width, OS conventions — caused comprehension failures and error recovery breakdowns that only surfaced on mobile.
All participants successfully selected a date — but the circle marking "Today" was widely misread as a selection indicator rather than a reference point. On desktop, comprehension was higher with more deliberate cursor behavior allowing users to read the calendar at rest. On mobile — particularly iPad — the larger touch targets and tap-first interaction removed the hover pause that helped desktop users orient. Zero iPad participants understood the indicator's meaning without prompting.
Desktop participants identified file upload errors at 100% — but mobile recovery dropped sharply. The root cause: the Adobe PDF file-type icon used the same red as the error state indicator, causing users to confuse the icon with the error message itself. On desktop, participants could read error text at a glance alongside the filename. On mobile, the compressed layout and icon-color collision made it impossible for nearly half of participants to match the error to the correct file — the most stressful moment in a disability application.
I could see something was wrong — there was a red icon — but I couldn't tell which file it was. I didn't want to delete the wrong one and have to start over.
Study Participant · File Upload Error Recovery · Mobile Study
This moment — paralysis at the point of error — is the exact failure the research was designed to surface. The participant wasn't confused by the concept of uploading a file. They were stopped by a color collision and a layout that worked on desktop but broke under real conditions on mobile. Task completion metrics would have called this a pass.
Divergent Findings · Where Platforms Split
Convergent Finding · What Both Platforms Confirmed
Date Picker Comprehension
Desktop users hovered before clicking — creating a natural reading pause. Mobile users tapped first. This behavioral difference, not design intent, drove the comprehension gap. iPad scored lowest: largest screen, tap-only, zero comprehension without prompting.
File Upload Error Recovery
Desktop's wider layout let error messages and filenames coexist visually. Mobile's compressed column collapsed that relationship. The icon-color collision compounded a layout problem that only appeared at smaller viewport widths.
Button Order & OS Convention
iOS users expected Cancel before Save; Android and desktop users expected Save first. A single convention regardless of platform created friction that only surfaced on mobile. Fix: Save → Cancel standardized per OS context.
Emotional Friction Was Platform-Agnostic
The most important convergent finding: emotional discomfort with inclusive identity fields was not a desktop problem or a mobile problem — it was a design problem. Likert helpfulness scores were nearly identical across both studies (Gender Identity: 4.38 desktop / 4.26 mobile; Pronouns: consistent negative sentiment on both). Platform did not mediate the experience. The absence of a graceful opt-out caused the same harm regardless of device — confirming that the fix had to be at the pattern level, not the platform level.
We validated 15 design patterns across desktop and mobile experiences. While task completion rates were high, several patterns revealed a critical gap: users could successfully complete tasks but did not feel in control or trust why sensitive information was being requested. The following patterns are highlighted because they most clearly illustrate this tension between usability success and emotional risk.
From the outset, this program was designed to outlast itself. The goal wasn't a report — it was a repeatable validation capability embedded into the organization's governance model and carried forward by practitioners who didn't start as researchers.
Research findings are only as valuable as the systems that act on them. Every insight from this program was translated into component updates, token-level design changes, and governance processes — not recommendations in a slide deck.
The governance model defines how patterns enter the system, who reviews them, when they require re-validation, and how decisions are documented — creating institutional memory that persists well beyond any individual researcher or project cycle.
The most important outcome wasn't validated components — it was a culture of evidence. Research at the design-system level has disproportionate impact: one validated pattern reaches millions. An unvalidated one carries millions of small harms.
This project reinforced that effective research leadership means operating at two levels simultaneously: rigorous enough to surface findings that hold under scrutiny, and strategic enough to translate those findings into governance decisions that survive team rotations, project changes, and institutional pressure.
Balancing rigor with speed required deliberate tradeoffs. Under resource constraints, I prioritized cross-platform consistency and decision reliability over larger sample sizes per pattern — accepting smaller subgroups in exchange for defensible governance decisions. Patterns were escalated only when risks appeared consistently across multiple data sources or when the severity of scaling failure was high enough to justify action.
If I were starting again, I would integrate emotional-trust metrics even earlier in the synthesis process. Their influence on inclusive design decisions — particularly for sensitive identity fields — proved stronger than anticipated. Functional success rates alone would have led to very different, and far less human, governance conclusions.