Debrief UX Audit
A first-pass review of the interface I built, against the goal it was actually meant to serve: letting an ordinary person understand what happened in their session without learning any neuroscience.
UX audit
A first-pass review of the interface I built, against the goal it was actually meant to serve: letting an ordinary person understand what happened in their session without learning any neuroscience.
Note: this audits the five-tab developer panel as it stood at the time. Its conclusions were acted on, and the shipped product is the three screens described in the redesign.
01 The diagnosis
The app is organised around the data model instead of around the question the user came with.
Five tabs, each showing a different slice of the data: the live stream, the file list, the analysis, the chat, the internals. That’s a developer’s map of the system. It’s also, structurally, the same thing the Neurosity console does, which is the comparison that started this audit, and it’s correct.
A person putting on a headset has one question: how did that go, and what should I do differently? Everything in the interface should be an answer to that, with the data available underneath for anyone who wants it. Right now the answer exists, the written debrief and the suggestion are genuinely good, but they’re the fourth thing you see, on the third tab, below a chart and six unexplained numbers.
The one-sentence version
The app currently shows you your data and expects you to interpret it. It should tell you what happened and let you check the data if you want to.
02 What works: keep these
Keep
The suggestion box›
“Protect 09:44-11:59, that was your longest good stretch, 2 hr 15 min of it.” That’s the best thing on the page: specific, actionable, in plain words, and it needs no EEG knowledge. It should be near the top, not two thirds of the way down.
Keep
The written narrative, as a concept
Prose beats a dashboard for this. The writing is fine; the format is wrong, five unbroken paragraphs is a wall to a tired person. Same words, broken into a two-sentence headline and a “more detail” section.
Keep
Honesty about signal quality
“94% usable signal” is exactly right for a consumer product, and most tools in this space hide it. It just needs saying in words, “most of this session recorded cleanly”, with the number available on hover.
Keep
Sources on guide answers, and refusing to guess
The principle stays. The presentation doesn’t, see finding 3.6.
Keep
The synthetic-data label
Visible, persistent, unambiguous. Keep it exactly as it is.
Keep
Copy summary
Quietly one of the strongest features, it lets someone take their session to any assistant they already trust. It just needs a label that says what it’s for.
03 What’s broken
Critical
3.1 · It opens on the least meaningful screen›
First launch lands on Live: two decimal numbers, five unlabelled bars, and eight electrode codes. Nothing on that screen answers a question a person has. It is the most technical view in the app and it is the first impression.
Fix The app opens on your most recent session’s debrief. If there are no sessions, it opens on a single instruction telling you how to record one.
Critical
3.2 · Numbers are shown without saying whether they’re good›
“Focus median 0.385.” “Your usual range 0.26-0.58.” “Calm median 0.479.” A person cannot act on any of these, and the entire justification for building this app rather than using the Neurosity console was that raw numbers need interpreting.
“Your usual range” is the worst offender, it’s a statistical concept wearing a friendly label.
Fix Every number gets a plain sentence attached: “0.385, a fairly typical day for you” or “higher than your last four sessions”. The number stays; it stops being the headline.
Critical
3.3 · Electrode codes and band bars are on the consumer screen›
CP3 C3 F5 PO3 PO4 F6 C4 CP4 means nothing to anyone who hasn’t read a 10-20 montage diagram. The five band bars are an instrument readout, a person cannot do anything differently because delta is taller than gamma.
Fix Both move to a developer view. The consumer sees one line: “All sensors have good contact” or “One sensor isn’t reading, nudge the headset.”
Critical
3.4 · Five equal tabs, one of them called Diagnostics›
Tabs suggest five equally valid destinations. Four of them are for you, the builder. Putting “Diagnostics” in a consumer’s navigation tells them the tool isn’t for them.
Fix One scrolling page for the person. A quiet “Developer view” link that opens Live, Diagnostics, bands, and electrodes.
Major
3.5 · The chart’s shaded regions are never explained›
Teal blocks and pink blocks appear behind the lines with no key. Those are the peaks and slumps, the two most important things in the session, and they’re rendered as anonymous background colour. You cannot tell which is which, when they were, or how long they lasted without cross-referencing two tables further down.
Fix Every event gets a label drawn on it: “Best stretch · 2h 15m” and “Dip · 47m”, with times. See section 4.
Major
3.6 · The guide’s answers and sources are unreadable›
A real answer from the screenshot: “Sources: A plain-English EEG primer § Brainwave bands, delta, theta, alpha, beta, gamma · A plain-English EEG primer § The eyes-closed test, checking the data is real · A plain-English EEG primer § Focus and calm scores”. That citation is longer than some answers and uses a typographic symbol most people have never seen.
Worse, the answer to “what is alpha” opens with “The wobbles have rhythms…”, a mid-paragraph excerpt from a document, not a reply to the question.
Fix One source, shown as “From: the EEG primer”, expandable. And each knowledge section gets a one-sentence direct answer at the top, so retrieval returns a reply rather than a passage.
Major
3.7 · Notes are buried and disconnected from what they describe›
The notes box sits at the bottom of a long page, behind a dropdown reading “peak at 09:44 AM”. You have to scroll past everything, understand the dropdown, and mentally connect it back to a shaded region on a chart you’ve stopped looking at.
Fix The note lives on the event. Click the dip on the timeline, or the dip’s card, and type into it there.
Major
3.8 · No hierarchy in the six stat cards›
Six identical boxes at identical weight means nothing tells you where to look, so you read all six or none.
Fix Two numbers at the top that matter to a person, how long you recorded, how much of it was good, and the rest folded into a “the numbers behind this” section.
Major
3.9 · Time is under-labelled and can’t be narrowed
Five tick marks, no date on the chart, no way to look at just the afternoon, no hover readout. For a seven-and-a-half-hour session that’s very little to navigate with.
Fix Section 4.2.
Minor
3.10 · Smaller things›
The state badge says “calm” with no explanation of what that means or where it came from. The chat was scrolled into the middle of a previous message on load. “mock source” is developer wording on a consumer chip. And the narrative repeats figures already shown in the cards directly above it.
04 The redesign
One page, answer first, data underneath. Here’s the shape.
Crown Debrief, one session
A good morning and a heavy afternoon.
7 hr 30 min recorded · most of it recorded cleanly · synthetic sample data
Whole day Morning Afternoon Drag to select
Best stretch · 9:44-11:59
Dip · 13:58-14:45
09:00 11:00 13:00 15:00 17:30
Your best stretch 09:44-11:59 · 2h 15m
Your longest sustained focus of the day, well above your usual level.
What were you working on?, tap to add
Afternoon dip 13:58-14:45 · 47m
Focus dropped below your usual level and stayed there.
meetings after lunch tired + add your own
4.1 · What changes structurally
Now
Opens on a live stream of numbers
Five tabs, one called Diagnostics
Chart, then six stats, then prose, then the suggestion
Peaks and slumps as unlabelled colour
Notes at the bottom behind a dropdown
Next
Opens on your last session’s verdict
One page, with a quiet developer link
Verdict, then timeline, then events, then detail on request
Every event named, timed, and clickable
Notes typed directly onto the event
4.2 · Your four asks, specified
Build
Labels on times›
Each peak and slump is drawn with its own label, a name, a clock range, and a duration, directly on the timeline. The date sits in the header. Hovering anywhere on the chart shows the exact time and the reading at that moment. Axis ticks go from five to one per hour.
Build
Filtering by time›
Three preset buttons, whole day, morning, afternoon, plus dragging across the chart to select any window. Selecting recomputes everything: the stats, the events, the written summary. “What happened between two and four” becomes a two-second action rather than a question you can’t ask.
Build
Notes on specific peaks and slumps›
Every event card carries an open prompt, “What were you working on?”, that you type into in place. No dropdown, no scrolling, no separate form. The storage format already exists and already works; only the way you reach it changes.
Build
Descriptions of what you were doing›
This is a different thing from a note, and worth separating: a note explains a moment; an activity covers a stretch of time. “Writing the parser, 9:30 to 12” is an activity. “The build broke” is a note.
Activities are what eventually make the debrief say something genuinely useful, “your focus runs about 40% lower in meeting blocks than in solo work”, which needs the same label applied across many sessions. So activities get a small reusable set of tags you build up over time, not free text every session.
05 Order I’d build it in
-
Restructure to one answer-first page~90 min Verdict at the top, timeline, event cards, detail folded away, developer view moved behind a link. This alone fixes findings 3.1, 3.4 and 3.8, and it’s the difference between “confusing” and “obvious”.
-
Labelled, clickable timeline events~60 min Names, clock ranges, durations drawn on the chart; hover readout; hourly ticks. Fixes 3.5 and half of 3.9.
-
Notes and activities on the events~60 min Type in place. Activity tags stored separately from moment notes so they can accumulate across sessions.
-
Plain language on every number~45 min Each figure gets a sentence saying whether it’s normal for you. Fixes 3.2, which is the deepest problem after the structure.
-
Time filtering~45 min Presets plus drag-to-select, recomputing everything for the selection.
-
Guide answers rewritten as answers~45 min A one-sentence direct reply at the top of every knowledge section, and citations reduced to one readable line. Fixes 3.6.
The first three are what a first-time user would actually notice. Four through six are what makes it hold up on the second and third session.
06 One thing I’d push back on gently
Everything above is worth doing. But it’s worth being clear that a lot of what makes the current version confusing isn’t layout, it’s that the underlying numbers genuinely are hard to interpret, and no amount of design fully fixes that.
A focus score of 0.385 is not meaningful in isolation even to someone who understands EEG. It only becomes meaningful compared to your own other sessions. Which means the single biggest readability improvement isn’t a UI change at all: it’s having five or ten real sessions to compare against, so the app can say “quieter than your usual Friday” instead of “0.385”.
That argues for getting the Crown back online and recording sooner rather than later, in parallel with the redesign. The interface work is real and I’d do it, but comparison is what makes the numbers speak.
Audit of the build committed as 5b31610, reviewed from the three screens as delivered. No code changed.
Explore more
This site covers what the documentation doesn't: the things I wish someone had handed me first.