Workshop
What the Comments Changed
July 10, 2026 · 8 min read
There was a comment that said lose the cat. One comment, out of hundreds about that cat, and every other one was affectionate. It changed how we work and none of the affectionate ones did — because it came with a reason: I stop resting and start monitoring her. I'm waiting for the next move the whole time.
That is the shape of a useful comment, and it is not the shape of a popular one. This piece is the rule we use to tell them apart, the five things viewers actually changed, and the longer list of things we have not done.
What makes a comment usable
Feedback on a video is not a vote. It cannot be, because the people who write are not a sample of anything — they are the people who felt strongly enough to type, which selects for a particular temperament.
So we read comments as reports rather than preferences. A report describes something that happened to the person while they were watching. A preference describes something they would like to exist.
The lamp flicker was uncomfortable to look at is a report. You should do a version with lo-fi beats is a preference. Both are legitimate and we are glad of both, but only one can be checked, and only one tells us about the object we made.
A reason makes a single comment worth more than a hundred agreements. Agreement tells us what people like. A reason tells us what the file is doing.
The cat comment was a report. It described an effect — vigilance instead of rest — that we had not designed for and could not have found on our own, because the four of us who watch every cut have known where the cat is for years, and you cannot wait for something you have already seen.
The requests we act on, and the ones we don't
Near-automatically, in practice: anything describing discomfort.
If something in an episode makes watching it worse — a flicker, a level jump, a bright element in a dark shot, a sound with an edge on it — we take that at face value and do not argue about how common it is. Discomfort is not a matter of taste, and it is systematically under-reported, which is the important half of the argument. Nobody writes a comment to say a flicker was unpleasant. They close the tab. The ones who write are the small visible edge of something we cannot see the rest of.
The second is any gap between what we said and what we delivered. If a title says three hours and the file is two hours fifty, that is our error and somebody is right to say so.
Then there is the largest single category of request, by a distance: more should happen.
A storm at some point. A visitor. A day-and-night cycle. Music under it. A story, even a slight one. Somebody asked, reasonably, for occasional bells so you know how much time has passed.
We do not do any of these, because they are all requests for the thing to become a different thing. Almost nothing happening is not a limitation of our episodes; it is the specification. A storm would improve the ten minutes somebody is actively watching and damage the three hours they are not — and the three hours they are not watching is the product.
This is the point where we sound arrogant, and there is no version of the sentence that fixes that. We are, in effect, telling people who took the trouble to write that we know better than they do about the thing they were describing their own experience of. All we can say is that the requests contradict each other constantly — the same week produced please add music and please never add music — so following them was never actually available, and a channel that tries ends up as an average of everybody, which is nobody's evening.
What one comment is worth
We tried to work out the multiplier once, because it seemed like something you could know.
On our long files, roughly one viewer in several thousand leaves a comment, so the arithmetic is tempting: one comment, several thousand people. We stopped trusting that number within a day.
The rate is not one rate. Complaints about discomfort come from a tiny, self-selecting fraction — most people just leave. Feature requests come from enthusiasts, who are wildly over-represented and the most likely to write at length. A comment saying please add X and a comment saying X hurt are not the same unit and cannot be added together, which is exactly what a comment count does.
Under-reported
Anything that made watching worse. Flicker, level jumps, a sound with an edge, a hard cut at the end. People do not write about these; they stop watching. One report is a signal, not an outlier.
Over-reported
Anything a person would enjoy more of. Requests come from the most engaged viewers, who are by definition the least representative of an audience that mostly falls asleep. Ten of these can easily be ten people.
Accessibility is not in the queue
One category skips the whole process above: if a request is about someone being able to use the thing at all, it is not a request, and it does not get weighed against anything.
We do not ask how many people it affects. We do not ask whether it fits the schedule. It goes to the front, because the alternative is a channel that decides some people's evenings are not worth a day of work, and we are not willing to write that sentence down.
Five things that changed
Two episodes had their lamp flicker rebuilt
A fast, regular flicker in an early hearth scene and its sequel. Reported as uncomfortable by two people about a year apart. We re-rendered both — the only time we have ever re-rendered a published long file — and slowed every practical flicker in the catalogue to an irregular, much gentler cycle.
Every file is now mastered to the same loudness
Our early episodes drifted by a noticeable margin between files. Somebody described putting on a second episode at night and having to reach for the phone, in the dark, because it came in louder. All new files hit one target now and the older ones were re-mastered, which was tedious and cheap.
No end screens on anything over two hours
The suggested-video cards at the end of a long file arrive at three in the morning for someone who has fallen asleep, over a scene that has just gone quiet. Several people described it and they were all right. Long episodes now end into silence.
One long episode in each batch has no animal in it
This is the cat comment. We did not remove the cat — the affection is real and the design reason still holds — but the person was describing something true about themselves, and there was no reason both kinds of viewer could not be served. It cost us nothing except the discipline of remembering.
Every description says what you will hear
One plain line, before anything else: rain on glass, a stove, wind in the eaves. It began as an accessibility request and helps everyone, including people choosing something to put on with the sound off.
The ones we did not get to
Two good suggestions have sat unaddressed for over a year, and neither has a principle behind it. We just have not done them.
Chapter markers on the eight-hour files. Somebody pointed out that finding your way back to a particular part of a long episode is essentially impossible. They are right. It is roughly a day of work per file, across a lot of files, and every time it comes up it loses to making something new.
A sound-only version. Requested repeatedly, obviously useful for anyone who wants the room without a lit screen in it. Not technically hard. It is a whole extra thing to produce, name, upload and maintain for every episode, forever, and we have not been willing to take that on.
We would rather have those written down here than quietly absent.
If you have something to add — particularly a report rather than a preference — the community page is where we actually read. And the cat, who caused all this, has her own piece in the four-minute cat.
The evening is warm here.
Comments (26)
Sign in to join the conversation
never thought about how much work is behind a single scene until reading this
this kind of behind-the-scenes writing is rare, more of this please
something about watching someone else's quiet routine is so soothing
the honesty in this about what actually goes wrong is refreshing
never thought about how much work is behind a single scene until reading this
the honesty in this about what actually goes wrong is refreshing
never thought about how much work is behind a single scene until reading this
the honesty in this about what actually goes wrong is refreshing
never thought about how much work is behind a single scene until reading this
this kind of behind-the-scenes writing is rare, more of this please
the small imperfections in the drawing make it feel so genuine
the honesty in this about what actually goes wrong is refreshing
this kind of behind-the-scenes writing is rare, more of this please
never thought about how much work is behind a single scene until reading this
the honesty in this about what actually goes wrong is refreshing