We rendered the graphics in a web browser

Mamba Smith is the founder of Queen City Garage, which produces The Grand National Show: a weekly NASCAR panel show on Grand National Today where broadcasters and drivers argue about the last race and pick the winner of the next one.

Not one graphic on The Grand National Show is made by an image model. Every board you see, every driver card, every standings table, is a web page rendered by a browser with nobody looking at it.

That is a deliberate choice, and it is the reason a correction is an edit instead of a re-roll.

Why not generate them

Our graphics are data cards. A driver's name, his number, his team, his finish, his gap to the leader. Every one of those has to be exact, and image models garble text. They will put a plausible-looking number in a plausible-looking font and the number will be wrong. For a show that argues about whether Carson Kvapil can make up nineteen points, a graphic that says twenty-one is not a small mistake.

So each board is a template written in ordinary web markup. The template pulls real photographs, real logos and the live points data. A copy of Chrome running with no window, driven by a script, opens it, renders it at 1920 by 1080, and saves the frame. The same markup renders the same pixels every time. If a number changes, we change the number and render again. If a layout is wrong, we fix the template and every card that uses it is fixed together.

The proof is a screenshot. That is literally what the render is.

What that looked like in the first week

Six of the seven Episode 2 cards were rendered, uploaded and on the shelf the day before recording. The seventh, the sixteen-driver Cup Chase board, waited on that night's race because the data did not exist yet. You cannot render a board from a result that has not happened, which is the right constraint to have.

Then we looked at the rendered output, and looking is where the errors turned up. None of them would have been caught by the build.

One card said all three drivers fighting for the last Chase spot drove for Joe Gibbs Racing. Rajah Caruth drives for JR Motorsports. That came from a note written from memory and was fixed against the official entry list. The Darlington card carried claims about the track that nobody had sourced, so it was rebuilt from the track's own published history, which we hold as a document, and an invented superlative came out. A layout flaw across all six was fixed once, in the template.

The Ryan Sieg Daytona card took 14 proofs before it was approved: the driver crop corrected so neither Sieg nor the person beside him was clipped, the middle stat boxes rebalanced, the results block enlarged and centered and given its missing texture, a piece of internal production text removed from a public graphic. The Chase seeds card took 10. That day's whole visual build ran nine hours and nineteen minutes, first artifact to last, and produced 158 files.

Fourteen proofs is a lot. It is also fourteen renders, not fourteen regenerations, and each one changed exactly what was asked.

The grammar that came out of it

Once the boards were right we locked the show's visual grammar so it would stop being re-decided every week: the background frame at 30 percent, a half-second panel push, the ticker at 123.25 pixels per second, names-only boxes under the panel, feature clips muted under the conversation.

Locking that is what makes the layouts reusable. Episode 3 did not start from a blank page. It started from Episode 2's templates with new data in them.

The honest part

The rendering is deterministic. The judgment is not. The machine will render whatever the template says, including a wrong team and an invented superlative, and it will do so beautifully. The catches in that first week were made by a person looking at a rendered board and knowing that Caruth drives the 88 for JR Motorsports.

So the division of labor is clear. The template and the render are the machine's. Every name and number on the board is a person's responsibility, checked against a source, before the board goes on the shelf.

What a person did: wrote the boards, checked every name and number against a source, called the fourteen proofs. What the machine did: rendered every board from real data, identically every time, in seconds.

This is No. 04 in How We Make the Show, the Queen City Garage build log on producing The Grand National Show. Watch the show at Grand National Today.