Checking the file, not the timeline

The timeline is not the deliverable

Everything you have verified so far, you verified on a timeline: the scopes, the loudness meter, the captions, the cut. None of that is what anybody receives. The deliverable is a file, and the file can differ from the timeline in colour, in loudness, in duration, in track layout and in what the first frame is.

So the last act of any job is a check performed on the rendered file, opened as a stranger would open it. It takes about twice the running time of the film and it is the highest-value half hour in the process.

Pass one: watch it, as an audience member

Open the file in an ordinary player. Full screen. Do not scrub, do not take notes on the timeline, do not touch anything. Watch the whole thing once, at speed, on the device it is meant for — the phone, if it is a phone.

This is the only opportunity you get to experience the film rather than inspect it, and you will lose it the moment you pause. What you are listening for is whether you understood it, whether you were bored anywhere, and whether the sound carried in the room you were in.

Write down what you felt afterwards, not during.

Pass two: the technical sweep

Now inspect. In order, because the order is roughly cheapest-first.

Duration. Does the file length match the timeline length, to the second? A mismatch means a render range, a missing tail, or a frame rate conversion.

First frame and last frame. Step to frame one. Is it black, or is it the last frame of the slate? Step to the end. Does the film end cleanly, or is there a frame of the next shot, or two seconds of silence you did not intend? Head and tail are the two places nobody looks and everybody notices.

Loudness, measured on the file. Not the timeline meter, which measured what the mix bus was doing, but the exported audio:

ffmpeg -i final.mp4 -af loudnorm=print_format=summary -f null -

Read Input Integrated and Input True Peak and compare them with the target for the destination. A render can be quieter than the timeline if a bus was excluded, or louder if a limiter was bypassed on export.

Scrub the joins. Jump to each cut you were unsure about, and to every point where picture and sound change together. Compression artefacts and sync problems both live at transitions.

Check the file's own properties. ffprobe or MediaInfo: codec, resolution, frame rate, pixel format, audio channels and sample rate. Compare against what you intended, line by line. This is where you catch the 29.97 preset.

Captions. If burnt in, check they sit inside the safe area on the actual platform, not in your editor's guides. If a sidecar file, open the SRT in a text editor and read the first and last cues, then load it in the player alongside the video and check the timing has not drifted.

Pass three: the other machine

Play it somewhere that is not your edit suite. A phone at half brightness. A laptop with the built-in speakers. A television, if the job is for one.

Three things surface here and almost nowhere else: shadows that were readable on a graded monitor and are now a black rectangle, dialogue that was clear in headphones and is not on a phone speaker, and text that was elegant at a metre and is illegible at arm's length.

The colour block's lesson on legal levels covers the measurement side of this. This is the human side, and they catch different faults.

The checklist to keep

Before anything leaves your machine:

1. Duration matches.
2. Watched start to finish, once, without stopping.
3. First frame and last frame are what you intended.
4. Integrated loudness and true peak measured on the file.
5. Codec, resolution and frame rate confirmed with ffprobe.
6. Captions checked in position or the sidecar opened and read.
7. Played on a phone.
8. Filename is the agreed convention, with version and date.
9. The master, textless version and stems are rendered and filed.
10. The music licence and any archive permissions are saved in the project folder.

Items 9 and 10 are not about this delivery. They are about the request that arrives in four months, and they are the two people skip.

When somebody else is checking

If a broadcaster or platform runs an automated quality check, it tests things a human never would: a single illegal luminance value, a frame of black longer than permitted, an audio channel below a silence threshold, a photosensitivity risk from rapid flashing. A rejection comes back as a timecode and a rule number, and it is usually one frame.

That is not an insult to your grade. It is a machine doing exactly what it was asked. Fix the frame, re-render, resubmit.

Honest limits

No checklist catches a film that does not work. The technical pass tells you the file is correct; it says nothing about whether the second act is long, and the two get confused precisely when you are tired and want to be finished. If the first pass told you that you were bored at four minutes, no amount of ffprobe addresses it, and the honest move is to go back to the timeline rather than to ship a technically flawless version of something you already know is slack.

Today

Take the last file you delivered. Run the loudness command on it and check the first and last frame. If either surprises you, that is the check earning its place.