A digital film has a running time to the frame, the server logs the start and the end of playback, and the projector logs the same from its own clock. In screen 1 both clocks agree that every feature ends four minutes and ten seconds after it should, and the film has not been altered.
I am the projection technician for a nine screen site.
How long a film is.
A digital cinema package lists its reels and each reel has a frame count. Twenty four frames a second, so a feature of a hundred and forty thousand frames runs ninety seven minutes and thirteen seconds, and it does every time, because there is no film to stretch and no motor to drift.
The server logs playback start and end. The projector logs the same from its own clock. They are separate machines with separate clocks, and they agree to the second on every screen I have, and I check because a drift between them is the first sign of a server fault.
In screen 1 every feature ends four minutes and ten seconds late.
Both clocks. Start on time, end four minutes ten late, on a package whose frame count says it should have finished four minutes ten earlier.
The audience does not notice. That is the first thing anybody asks and the answer is no. Nobody has ever complained that a film in screen 1 was long, and the ushers who stand at the back for the credits say it ends when it ends.
The next show does notice, because the schedule is built on the frame count and screen 1 runs four minutes into its own turnaround every time, and the cleaners have learned to wait.
What has been checked.
The clocks. Both synchronised to network time, both checked against my watch and a radio clock. They keep time on every other screen and they keep time in screen 1 during the day when nothing is playing.
The package. I have played the same package in screen 1 and screen 2 on the same night, ingested from the same drive. Screen 2 runs it to the second. Screen 1 runs it four minutes ten over.
The playback. This is where it stopped being a clock problem. The server reports the frame it is on, and I have sat in the booth with the log open and watched the frame counter.
It does not slow. Twenty four a second, steadily, from the first frame. And it reaches the last frame four minutes and ten seconds after it should, having played, by its own count, every frame at the right speed.
The only way to reconcile that.
I have put it to the manufacturer and to two colleagues at other sites, and the only arithmetic that fits is that the server plays about six thousand frames that are not in the package.
Six thousand frames is four minutes ten at twenty four a second. The package does not contain them. The server's log does not list them, because it lists reels and the reels are the reels. But the counter that ticks once a frame ticks six thousand more times than there are frames, and the projector's lamp is lit for all of them, and the clocks are right.
Where in the film.
I have tried to find it by watching. A feature I know well, from the booth, with the script beside me. It runs the way it runs, and it ends, and my watch says four minutes ten over, and I could not tell you where the time went, and I have done this eleven times.
Except that on the ninth time I noted the time of every scene change against the script, and the scene changes drift. Not steadily: they are on time for the first reel and a half, and then somewhere in the second reel they are all four minutes late, and I cannot find the join.
Between two scenes of a film I have seen thirty times, in screen 1, there is four minutes and ten seconds of something, and I have watched that stretch of that film with my eyes on the screen and I cannot tell you what was in it.
What is done about it.
Screen 1's turnaround has four extra minutes built into the schedule, which is a line in the booking system that says technical, and the owner has never asked what it is for.
The manufacturer's engineer downloaded the logs, confirmed that the frame counter and both clocks agree with each other and disagree with the package length, and recommended that the projector be replaced. It was, in 2023. The new one does it.
What I do.
I build the playlists and I check the logs.
The one change is the log itself. The server can be set to log at the reel level or at the frame level, and I had screen 1 on frame level for a month to see where the six thousand frames sat.
They sit in the second reel, in one block, at a point that is nineteen minutes and forty seconds into that reel on every feature regardless of the feature, and the log for those frames has a field for the reel's identifier, which on every real frame is the package's identifier, and on those six thousand it is a string that is not in any package I have ever ingested, and it is the same string every night, and I have set the log back to reel level because I do not want to know what it decodes to.