Could we make this show without AI?

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.

Could we make The Grand National Show without AI? Yes. Could we make two of them a week, at this standard, with this team, and have the clips in every panelist's hands the same day? No.

I want to be exact about that, because the honest answer is more useful than the impressive one.

What we have not measured

Two weeks ago, in Making a Weekly Racing Show With AI, we said we had not run a timed comparison against producing the same package by hand, and that there was no defensible hours-saved figure. That is still true. We have not run that comparison and I am not going to invent a percentage for this post.

What we have measured is what the machine did, and what it cost when it was wrong. That is what the last thirteen posts are. So here is the ledger.

What it costs when we get it wrong

These are the measured ones. A day of graphics on Episode 2 ran nine hours and nineteen minutes, first file to last, and one card took fourteen proofs. Each graphics fix on Episode 5 cost about 45 minutes of rebuild. Running the delivery uploads at the same time as a 27 gigabyte archive copy cost about three hours this week, because the two fought for the same connection and the uploads fell from about 85 megabytes a minute to about 15. A one-pass conform put a master with a sync fault on the live page for an hour and fifty minutes.

Those are not AI costs. They are production costs, and the machine's part in each was to make the fix cheap once a person found the fault.

What the machine does

It hashes every recording on the way in and refuses to treat a file as a source until the hash is on record. It transcribes the recordings and matches them to the rundown. It flags clip candidates against the topics, 49 of them this week, from the transcripts and coverage we already hold.

It renders every graphic from a template and real data, in the right font from any folder, with official portraits cut out by face. It conforms picture and sound in separate passes. It measures every camera's length before the build and stops if one is off by more than a second.

It runs the checks. Two hundred and five on one master, 343 on another, 719 on the version of Episode 4 that fixed the drift. Sync within 30 milliseconds. Loudness on target. No muted panelist on screen. Every clip measured against the master for noise, tone and sync before it can be packaged.

It registers, hashes, versions and locks every file that leaves, builds each panelist's page, records each send and each download. It writes the archive to the shelf and verifies it: this week, 48.5 gigabytes, 5,569 files, zero mismatches. It mirrors everything off-site every night, with no list of what to include, because the list is what failed.

What the people do

Every judgment call in this series. Deciding what an episode is. Cutting an open ten times. Rejecting a rundown that was correct and unusable. Choosing nine clean clips over twelve. Watching the ninth cut and writing twelve notes that were all true. Noticing a five where a four should be. Watching your own clip and saying it is wrong when every check says it is right.

And the checking. Every name and number on every board is a person's responsibility. The machine renders Caruth in the wrong team beautifully. A person knows he drives the 88 for JR Motorsports.

The answer

Without the machine, one person could make one show a week and the clips would follow when they followed. With it, the same person makes two shows a week, every file is provable, every panelist posts the same night, and the mistakes are found by a check the next morning instead of by a viewer next month.

That is not automation replacing the work. It is automation making the work survivable. The show is still made by people who watch it. It is just no longer made by people who also have to hash 40 files at one in the morning.

What a person did: everything that required knowing what the show should be. What the machine did: everything that had to be true at every second of it, thousands of times a week, without getting tired.

This is No. 14, the last in How We Make the Show, the Queen City Garage build log on producing The Grand National Show. Both shows: Grand National Today.