One camera ran two percent slow

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.

One panelist's camera on Episode 4 of The Grand National Show ran 2.06 percent short. That one number explained every note I sent on the ninth cut.

I sent twelve. "They're talking over each other." "Audio and video are off here." "The picks are in the wrong order." Every one of them traced back to a single recording file that was 1,981 seconds long when it should have been 2,022.

What 2.06 percent does

A remote panel show is four separate recordings, one per person, lined up on a shared clock. If one file is short, it is not short at the end. It is short everywhere, evenly, so that person drifts earlier by about 1.2 seconds every minute. Five minutes in, he is six seconds early. Twenty minutes in, he is twenty-four seconds early and answering questions before they are asked.

On the timeline that showed up as nine places where two people overlapped. The edit had treated those overlaps as crosstalk and cut around them, which also cut about 80 seconds of that panelist as dead air, because the "dead air" was the gap his own drift had opened.

The fix, and the number that proved it

The fix was to stretch the short track back to its true length. Overlaps went from nine to zero. The picks fell into the order they were spoken. The 80 seconds came back.

Version ten rendered at 30 minutes 52 seconds and passed 719 automated checks with no failures, including a new one written that day: no muted panelist is ever on screen. That check exists because the drift had put a silent person in the frame while someone else talked.

The gate that came out of it

Every recording now has its duration compared against the others before the build starts. If any camera differs from the rest by more than one second, the build stops and says which one.

That is a two-line check. It would have caught Episode 4 on the first cut instead of the tenth. It is exactly the kind of check that does not get written until the failure has already cost a day, which is why every check in this series got written down the day it earned its place.

I should say plainly that this was not the panelist's fault. Recording software drops frames. Networks stutter. The file that comes down is the file that comes down. Our job is to notice, and for nine cuts we did not.

Why the notes were right and the diagnosis was wrong

Every note I sent was accurate. People were talking over each other. Audio and video were off. The picks were out of order. What none of the notes could see was that one cause produced all of them, because from the viewer's chair the symptoms look like three different editing problems.

That is the general lesson, and it is why the check measures the source files and not the finished cut. By the time a fault is visible in the edit it has already been spread across thirty minutes of decisions. Measure it at the door.

What a person did: watched the ninth cut and wrote twelve notes that were all true. What the machine did: found the one number behind all twelve, and now refuses to build a show where it is wrong.

This is No. 07 in How We Make the Show, the Queen City Garage build log on producing The Grand National Show. Watch the episode: Jesse Love Wins at Gateway as the Chase Tightens.