Why Eight Hours

Duration is not a metric here. It is a material, like a colour or a sound, and we choose between one hour, three and eight the way you would choose between a mug and a pot.

What a length promises

Every duration makes a promise before a single frame plays, and the promises differ.

A one-hour file says: this has a shape, and you will see the end of it. A three-hour file says: this will outlast the thing you are doing. An eight-hour file says something that is not about content at all: this will not end before you do.

That is the whole of it. The value of hour seven is not that anyone watches hour seven — it is that at hour one, nobody has to wonder about it. No arithmetic about whether it will run out, no small decision waiting somewhere ahead. The length removes a question rather than adding content.

Nobody needs the eighth hour. What they need is to not be doing sums about the second one.

Seen that way, a lot of strange behaviour makes sense — including why the same person starts an eight-hour file at eleven at night and a twenty-minute one at lunch, and would not swap them.

Three lengths, three occasions

We build to three rough durations, aimed at three moments rather than three audiences. The same person uses all three in a normal week.

1. 1

   Around an hour — the break

   A pause with edges. You are between two things and you want the second one to still happen. An hour has a beginning, a middle and a visible end, so it can be structured: weather that develops, light that changes, an ending that arrives. The only episodes where we allow anything like an arc, because the only ones anybody finishes.

2. 2

   Around three hours — the evening

   Runs while you do something else: reading, cooking, work that needs calm but not silence. Long enough to stop being an event and become the room's weather. Nothing may develop here — a change at minute ninety pulls your head up, and pulling your head up is the failure mode.

3. 3

   Eight hours — the night

   Not made to be watched. Made to be left on. Its only design requirements are that it never surprises you, never gets brighter and never runs out. The least interesting file to make and the one people write to us about most, which took a long time to accept gracefully.

Our first-year mistake was treating these as one episode at three lengths. An hour-long episode stretched to eight is unbearable: everything that makes an hour good — development, structure, a shape you can feel — becomes an intrusion once the file is furniture.

The end screen, and the damage it does

Here is the thing that pushed us toward long files. It is small, slightly petty and completely real.

A video ends, and the room you were in ends with it. What arrives next is a bright grid of suggested thumbnails, an autoplayed video louder than the one you chose, or an advert. At three in the afternoon that is nothing. At two in the morning it is a bright screen and a stranger's voice in a dark room.

We have made this as un-bad as we can. Our episodes end in near-silence and stay dark: no music sting, no outro, no brand card. The last two minutes of every file are quieter than the rest, so that if it does end while somebody is asleep, the change is a fade rather than a cliff.

Length is the one lever we hold. An eight-hour file outlasts most nights, so the end screen never happens — the ending never arrives while anyone is there to be interrupted by it. That is the honest engineering reason for the long files, and it weighs more than any aesthetic argument.

What the numbers say, and what they don't

We look at the viewing data. It is less informative than people assume, and it is worth being precise about where it goes silent.

It tells us that average watch time on an eight-hour file is a small fraction of its length, that most sessions start in the evening, and that the drop-off in the first two minutes predicts everything else. Long files get left on; short ones get finished.

It does not tell us whether anyone was in the room. It cannot distinguish a person watching from a person asleep from a laptop left open in an empty kitchen. A session that ended after fourteen minutes might be someone who disliked it or someone who got what they needed and went to bed — opposite outcomes, identical in every chart we have.

The rule we work to: a number can tell us something is broken; it cannot tell us what to build. We have never added an event to a scene because a retention curve wanted one.

What long files cost, and who they suit

The boring accounting.

An eight-hour master is around a hundred and thirty gigabytes before compression. The final encode takes most of a day, during which nothing else can be worked on, and a failed encode found the next morning costs the whole day again — which has happened twice, both times a full disk.

Storage is the part nobody warns you about. We keep every master, and a handful of eight-hour files take more space than everything else combined. A re-render is never a small edit either: fixing two seconds at hour six means paying the entire encode again, which is why the flaws we have written about elsewhere mostly stay where they are.

Against all that, a long file keeps being useful for years. Per hour of eventual use it is the most efficient thing we make, and that is not an accident of taste.

And now the part it would be easy to leave out.

Long watch time is good for us. A file that runs all night accumulates playback hours whether or not anyone is awake, and a platform that measures attention in hours does not ask who was in the room. The incentives point where our design choices point, and anyone claiming their eight-hour ambient video exists purely for the viewer is telling you half of it.

We do not think we are doing anything dishonest — a night file that never brightens and never surprises is a good faith object, and every alternative makes the experience worse. But when virtue and self-interest agree this neatly, everyone should look twice, us included. It is the easiest place on this channel to fool ourselves, and we would rather write it down than leave it sitting quietly in the incentives.

Honest numbers on the label

One rule, no exceptions: the number in the title is the length of the file. Eight hours means eight hours with weather in all of them — no silent tail, no scene that quietly becomes another at hour four, no cross-fade from older work to pad the total.

A low bar, broken constantly in this corner of the platform, which is why it is worth saying. Duration is the main thing a person agrees to when they press play at midnight. It should be the one number nobody has to check.

None of this is a claim about sleep. We do not know what an episode does for anybody's night and are not in a position to find out; a long file is a design decision about interruption, not a health product. The first quiet hour of an evening is a different subject, and it is over in the hour that isn't sleep.

The evening is warm here.