Buying a ticket to a game is one of the few purchases where you pay first and find out what you bought later. A seating chart tells you a section number and a price. It does not tell you what the court looks like from row I.
Cinemas solved this years ago: you see the room, you tap the seat, you know what you are getting. I wanted the fan app to do the same for an arena, with one more step the cinema never needed, which is to sit in the seat and look at the court before you pay.
That sounds like a visual feature. It turned out to be mostly an engineering problem with a design problem wrapped around it. This is how it works and what went wrong on the way.
The frame time was measured on a Mac desktop at 1200 × 750 and sits at the display's refresh cap, so it shows headroom rather than a limit. It has not been measured on a phone.
How the seats are drawn
The building comes from one drawing
The arena is not modelled by hand. It is generated from a single seating plan, an SVG file. The code reads the sections from it and builds the rows, the seats, the aisles, the rails and the fascia. That is why every seat stays individually addressable: a seat is not a texture painted on a slope, it is an entry with a section, a row, a number, a position and the direction it faces.
One model, drawn thousands of times
Eighteen thousand chairs cannot each be sent to the graphics card as a separate thing. Instead, each section hands the card one seat model and a list of where to put it, and the card stamps it out in a single pass. The colour of each seat (open, sold, yours) is one value in that list, so marking a pair as yours changes two numbers and nothing is rebuilt.
Detail only where you are looking
The seat itself was built in Blender twice. The close version has a body and a hinged pan and is 2,288 triangles; it is used only in the section you are in. Every other section uses a 352-triangle version of the same seat. From across the bowl you cannot tell, and the far seats cost about a sixth as much to draw.
| View | Frame time | Draw calls | Triangles |
|---|---|---|---|
| Arena overview | 8.3 ms | 198 | 1.5 M |
| From a seat, with the roof and detailed seats | 9.4 ms | 710 | 4.6 M |
The ticketing experience
The flow is four steps, and none of them asks a question before showing something.
Pick the night. Tonight's game is the first thing on Home, with tickets one tap away. The app does not ask how many seats or what you want to spend before you have seen the building.
Pick the section. Tap a section and the camera flies into it. Open seats, sold seats and your seats are three colours. Tap a sold seat and it says so, instead of doing nothing.
Pick the pair. Tap a seat and the app takes the pair. The row and seat can also be chosen from two menus, so the picker does not depend on hitting a small target with a thumb.
See the view, then pay. The camera drops into the seat at eye height and looks at the court. One total, fees included, is shown the whole way through.
What went wrong, and the fixes
The seats you picked were hidden by the panel that describes them
The seat panel and the mode chips sit on top of the 3D view. The camera centred a section on the whole canvas, so half the section, and the pair just chosen, ended up under the panel. When a pair was picked the panel grew and covered more.
The app now tells the 3D view how many pixels its controls cover at the top and bottom. In the section view the picture is shifted so the point the camera aims at lands in the middle of what is still visible, and the shift eases as the panel grows.
Not every fan carries this year's phone
A full 3D arena is a lot to ask of an older phone or a slow connection. A seat picker that only works on a flagship is not a ticketing feature.
There is a second mode. Light mode draws the same building as line work only: no surfaces, no lighting, no roof, no seat models. Sections, seats, availability and prices are identical, and it behaves like a map that pans and zooms. The app chooses it for the fan when data saver is on, the connection reports 2G or 3G, the phone reports little memory, or it cannot draw 3D at all, and a line under the map says why. A switch in the header overrides it either way.
The camera whipped round on the way into a seat
Flying from above a section down into a chair by aiming the camera at a moving point made the view spin as the two points met.
Moves to and from a seat now turn the camera itself from its old heading to the new one, and time is counted in capped steps, so a slow frame pauses the move instead of making it jump.
A 3D scene that runs when nobody is looking at it
The arena sits behind the Tickets screen and inside a long case-study page. Left alone it kept drawing at full speed off screen.
The scene stops drawing when its canvas is scrolled out of view, when the tab is hidden, or when the page around it asks it to pause. A full scroll through the case study went from an average of 61.7 ms per frame to 11.2 ms after this and a handful of related changes, measured on one Mac.
Scrolling the page got caught by the map
A map that takes the pointer traps anyone who was only trying to scroll past it.
On the case-study page the picker stays behind an "Explore the arena" button. Clicking takes control; scrolling away or pressing "Done exploring" gives it back.
What is still approximate
This is a prototype, and I would rather say where it stops than let it look more finished than it is.
Section numbers, row counts and seats per row are estimated from the seating plan, not taken from the venue. Which seats are sold, and what they cost, is sample data in the same shape a real feed would have; nothing is read from a ticketing provider. Checkout is a demo: no card is asked for and nothing is charged. And none of this has been tested on real phones yet, only in resized browsers.
A refresh, not a replay. This is a refreshed prototype, not the exact work I did. A lot has changed since 2018, so I rebuilt the idea for today.
The opportunity I see is bigger than one team. Nothing in the seat picker is specific to one building: give it another arena's seating plan and it generates that arena. The app around it can be white-labelled for any team, with any additional feature added on top.
See it running in the case study, where you can pick a seat yourself.
