Working Slowly on a Fast Platform

This is not a piece about beating an algorithm and we have no advice to offer on that subject. It is an account of a scheduling problem that has no solution, only positions, and of what our particular position costs.

Where the six weeks go

The honest breakdown first, because the number sounds like a boast until you see it spent.

Week one is deciding. Which room, which season, what time of evening, what the weather is doing. Perhaps four days of real work, and the rest is the part nobody puts in a schedule: not deciding yet. An episode chosen on the first afternoon is nearly always somebody's habit rather than an idea, and we have shipped enough of those to recognise them now.

Weeks two and three are drawing. Layers, mostly. A single room is a dozen elements that have to move independently, and the ones that take the longest — fire, water, anything with a leaf on it — are the ones that will be looked at hardest.

Week four is sound. Longer than people expect. A three-hour bed is not a three-minute bed repeated; the layers have to be different lengths, nothing distinctive can survive at the front, and a single recognisable creak can undo the entire file.

Week five is assembly. Offsets, loop lengths, the first full playback, and the list of things that only become visible when you sit through it.

Week six is the machine's. Rendering, then fixing what the render exposes, then the thumbnail and the title and the description and the upload, which take a full day between them and are the least interesting of the six weeks.

Total hands-on time is maybe a hundred and fifty hours. The other four hundred are waiting, deciding, and the slack that makes deciding possible. That slack is the real luxury here, and it is the first thing frequency takes away.

Six weeks is not six weeks of work. It is a hundred and fifty hours of work with enough room around it to change our minds.

Filling the gap, and what each one cost

Between two long episodes there is a hole, and we have tried four ways of filling it.

1. 1

   Re-cuts of older episodes

   Take a three-hour file, cut a forty-minute version, publish it. Costs a day. We did it twice and both are among the weakest things on the channel — not because a shorter version is worse, but because a piece designed to last three hours has its structure removed when you take an arbitrary slice out of the middle. It also has the smell of a shelf being restocked. We will not do it again.

2. 2

   Single-room short episodes

   Twenty to forty minutes, one room, one weather, no ambition beyond being properly made. These started as filler and turned into a real format. They are the only gap-filling method we would defend on its own terms, and they work precisely because we stopped treating them as a smaller version of something else.

3. 3

   Compilations and playlists

   Free. Also, honestly, close to empty — nothing new is made and the viewer knows it. We keep playlists because they genuinely help people find an evening, but we stopped presenting them as anything more than shelving.

4. 4

   Writing instead

   This blog, partly. It fills a gap for us more than for anyone else: it is a way of being present without pretending to have finished an episode. It also quietly eats production time, which took us a year to admit.

The fake part of the dilemma

The choice is usually framed as quality or frequency, and that framing is wrong in a specific way that took us a long time to see.

Care is not a dial. When a deadline arrives early, nobody decides to be less careful; that is not a thing a person can do on purpose. What actually happens is that the last week disappears — the week of full playbacks, the week that finds the seam and the level jump and the object that tells too much. The work does not get worse made. It gets less checked, and the difference shows up as small errors that survive into the file rather than as a visibly hasty episode.

So the real variable is not quality. It is scope, and the person who says yes to frequency is really saying yes to a smaller episode, whether or not they know it.

A fixed day is worth more than a fast one

The one thing viewers reliably notice is not how often we publish. It is whether we publish when we said we would.

We settled on a fixed weekday. It sounds like a triviality and it is the single most useful scheduling decision we have made, because a fixed day converts a channel from a stream of surprises into something a person can arrange an evening around. People write to say they save an episode for Friday. Nobody has ever written to say they wished there were more of them, which is not the same as nobody wishing it, but it is telling.

We miss it. In the second year we missed four slots out of twenty-six, and the missing weeks did measurably more damage to the habit than a slower fixed rhythm would have. That is the actual argument for slowness, and it is a modest one: not that slow work is better, but that a rate you can hit beats a rate you admire.

Shrinking the thing itself

What has actually made this sustainable is not discipline or better tools. It is that the episodes got smaller and we stopped counting that as a defeat.

One room instead of three. One weather state instead of a progression. A room we have already built rather than a new one — which is most of what fifty-two episodes, one building is about, and the reason that piece exists is that geography turned out to be a schedule tool as much as a design one. Reusing a kitchen you have already drawn is not laziness; it is the only reason there is a kitchen worth returning to.

The episodes people name as favourites are, with irritating consistency, the small ones.

Where we gave in

Three times we let the calendar decide, and all three are weaker. We are naming them because a piece like this is worthless without the list.

The second spring rain episode. Published eleven days after the previous one, during a stretch where we had convinced ourselves momentum mattered. The checking week never happened. There is a level shift about forty minutes in where a sound layer comes back at the wrong gain, and it is exactly the sort of thing a full playback catches.

The December hearth re-cut. One of the two re-cuts described above. Made in a day, published because a slot was empty, and it does not have a shape.

The autumn short that became a long. We had a twenty-minute piece that was finished and good. Someone suggested extending it to two hours to fill the slot properly. The extension is padding, and you can hear where the original ends.

The cost of working this way is real and we should state it plainly rather than let the tone imply otherwise: a channel publishing twice as often would be substantially larger than this one, and we know that, and some months it stings. There is no claim here that slow work is better work. There is only the observation that this is the rate at which we can make things we would put on ourselves — and that a rate you cannot hold is not a schedule, it is a wish with a date on it.

The rest of the reasoning behind how this place operates is on the about page.

The evening is warm here.