- Peer review
- CHI 2027
- The methodology is being formalized for submission to ACM CHI — institutional credibility a wrapper can’t replicate.
PreflightUX
Calibrated Synthetic UX Testing
PreflightUXEmpirical Signal Before the Build
Run calibrated synthetic expert users against your early designs and prototypes to surface deep workflow friction — graded against the same foundational UX frameworks human researchers trust — before you write a line of production code.
The cost asymmetry
A flaw caught in a design file costs a revision.
The same flaw caught after launch costs a re-build.
Yet the specialized users who would surface it — growth scientists, SREs, clinicians, associate lawyers — are scarce, expensive, and slow to recruit before there is anything built to test. So the riskiest decisions get made on the thinnest evidence. PreflightUX closes that gap.
What you get
Each engagement delivers a structured usability audit: your early artifacts driven by synthetic expert personas, evaluated against the foundational UX frameworks human researchers trust, and returned as an auditable, severity-rated critique — with a calibration score that says how closely the synthetic findings track real users.
Synthetic expert-user audit
A cohort of synthetic expert personas drives your actual interface — or reads your PRD — thinking aloud and hitting friction the way your real power users would.
Foundational-framework scoring
Every interaction is graded against Nielsen's 10 usability heuristics, heuristic evaluation, and cognitive walkthrough — with severity ratings, not vibes.
Auditable Actor / Critic critique
An independent Judge Agent reviews each session against domain-specific heuristics, producing a structured, traceable critique you can defend in a design review.
Published calibration score
We measure how closely the synthetic findings track real users and publish the result — so you know exactly how much to trust each finding.
Grounded in foundational UX science
Rigor isn’t the new part. The users are.
Synthetic expert users are the bleeding edge. The standards we judge them by are not. PreflightUX evaluates every finding against the same battle-tested frameworks expert UX researchers have relied on for decades — so the signal is both novel in how it’s produced and orthodox in how it’s assessed.
Heuristic evaluation
Each persona inspects the interface against the full heuristic set, flagging violations the way a trained evaluator panel would.
Cognitive walkthrough
Personas attempt real tasks step by step, exposing the exact points where the product breaks the user’s mental model.
Severity ratings
Findings are scored 0–4 on frequency, impact, and persistence — so you fix the right things in the right order.
Nielsen’s 10 usability heuristics
The Judge Agent scores every session against all ten — alongside the domain-specific heuristics authored for your product.
How it works
One pipeline, two agents, and a discipline of measuring itself against reality.
Artifact in
Hand us a prototype, a Figma flow, or a written PRD. No production code required.
Double-grounded personas
Personas grounded in real, anonymized interview transcripts and skill-graded on the exact axes your product demands.
The User Agent drives the UI
A browser-driving agent navigates as each persona — thinking aloud, hitting friction, and generating an interaction trace.
The Judge Agent scores it
An independent critic grades each trace against Nielsen's heuristics and your domain-specific rules — separating simulation from critique.
Signal out
A severity-ranked critique plus a calibration score, visualized as an interactive cohort constellation.
Why it’s different
Every commercial platform synthesizes consumer personas. None grade for domain expertise. None publish calibration reports against real users. The expert-user, calibration-disciplined lane is wide open — and that’s the one PreflightUX runs in.
| Dimension | PreflightUX | Aaru | Synthetic Users | Maze AI |
|---|---|---|---|---|
| Primary focus | Expert-grade workflow usability & structural friction | Macro-market sentiment & demographic polling | Qualitative concept & copy validation | Unmoderated human usability testing |
| Mechanics | Browser-driving agents on live interfaces | Statistical respondents to surveys | Chat-based generative interviews | Humans executing click-paths |
| Moat | Published calibration + academic peer review | Proprietary scale & data ingestion | First-mover brand in the AI niche | Access to paid human panels |
- Calibration report
- In progress
- The first public benchmark — synthetic findings vs. real users on the same scenarios — is underway.
- Substrate
- Open source
- Built on the OpenCosmos UI system. The foundations are auditable; the methodology is the moat.
Run a pre-flight on your next build
PreflightUX is taking a small number of design-partner engagements. If you’re shipping a complex B2B product and want empirical signal before you commit engineering capital, let’s talk.