The four doors and every capture flow are specified in Kept Things. This is the room’s instance.
What the two entry points in T open, which until now was nothing. A sheet, because keeping something is a flow through a room and the room is what it attaches to. The four kinds are tiles and there is no primary, because picking one settles nothing — it routes you to the camera, a browser, a color or the composer, and the step after it is where something gets kept. Selection plus a pinned button would make one act take two taps and put a label on the button that only repeats the tile you just touched. Tiles rather than rows because four verbs of three words need no caption under them, and in a grid they read as a set of equals instead of a list with a first item. A link is one of the four because a product page is the single most common thing anybody keeps while building a house, and until now there was nowhere to put one. Allowed in a room waiting on a date on purpose: a keep is not a decision, and the room being closed for answers is no reason it cannot hold what you noticed.
What it attaches to is stated rather than asked, because there is only one answer here. Once the eight decisions open, that line becomes a run and the question is worth asking — a photograph kept against a decision is evidence, and one kept against the room is a reference, and nothing can tell them apart later if the sheet never asked.
How the wall behaves, the full list and a keep’s own sheet are in Kept Things.
Where the hairline slot in T goes once there is anything in it. The wall is the one grid in the system that does not equalise its cells, because a photograph and a typed sentence are not the same height and cropping either to match is how a wall turns back into a list. The link at the top left is drawn on this page rather than by the system: KeepTile knows three kinds and a link is not one of them, so it is recorded as a component change rather than faked with a note tile. Each keep carries the day it was kept and nothing else: no count against it, nothing to do about it, and no suggestion that three of them means the room is further along. The room is still waiting on a date, so the head and the dated block are unchanged — keeping something did not start anything.
The position is a container decision, not a different component. One Sheet rises from the bottom edge below 840px because that is where the thumb is, and comes in from the right above it because a bottom sheet on a 1160px canvas is a letterbox. So there is no tablet gap to design for: the same drawing resolves to a bottom sheet on a phone, a bottom sheet on a portrait tablet, and the 480px rail in 4b on anything wider.
Drawn as screens, they have no such rule, and the capture shows what that costs: the question is the whole page, so the options stretch to the full window, the back control floats loose in the corner with nothing to sit in, and the room the question is about is gone from the screen rather than behind it. The three frames here are the one-question-at-a-time route drawn on the same component as 2b — the close bar goes, because the sheet carries its own close and a flow inside a room does not get navigation of its own.
One consequence worth naming, because it is the real difference between the two widths: at 88% of 844px the strip above the sheet is 101px, and the room's own app bar takes 94 of it. So on a phone the room is not visible behind the sheet at all — the scrim and the bar are the whole of the signal that you have not gone anywhere, and returning to the room is a memory rather than something you can check against. The desk keeps the room itself. That is an argument for the sheet on both, not against it: the phone cannot do better than the scrim, and a screen would not even offer that.
The fallback route, on the component it should have been on. The question is the sheet's own title, so the zone that held it on the screen version is gone — the sheet is the region, and the answers sit on its ground where selection is a fill. The count stays in the support line and there is still no denominator.
The scope rule, redrawn. The answer just given still sits above the follow-up, in the same card, because the follow-up is unreadable without it. Nothing else moves: an answer that could travel is still asked about rather than widened silently, and the note under the action still changes with the pick.
The same content as 4c, drawn on the phone. The room behind has already moved while you were talking, so closing this is not a transition to somewhere new. On the phone you cannot see that it has, which is why the footer names where closing it puts you rather than saying Done. On the desk in 4c the same sheet does not need the reassurance, because the moved room is on screen beside it.
2c on the desk, and it stays a whole screen because it is a destination: a room is a place you navigate to and come back to, so it takes the canvas the way 2d does. Left column is the room and what the conversation took out of it; right column is what is still yours to decide and what has already been settled. The two runs the phone has to scroll through in sequence are one eyeful here, which is the only thing the width buys.
These are never whole screens. 5a and 5b are the same flow on the phone, on the same component, and here it side-anchors at 480px with the room still on screen behind it. That is the one thing the width adds: the rule does not change, the position does. A flow through a room is not a destination, so it carries no navigation of its own — the sheet's own close is the way out.
5c in the same rail, and it is the frame that proves the rail is right: the read-back is a claim about this room, and on the desk you can see the room it is a claim about while you check it. The room behind has already moved — three waiting, the rest answered — so closing the sheet is not a transition, it is the sheet going away from something the reader has been looking at the whole time. The one thing drawn from two answers is not in here; it is a question, so it belongs in the flow rather than in the summary of it.
2f and 2g are two phone screens because a phone cannot hold both, and they are the one pair the width genuinely repairs: the answer on the left, what it opens and what it closes on the right, so the consequence is visible while the choice is still live rather than reported after it. This stays a whole screen and not a rail — a decision is a destination you cite, arrive at from four places, and come back to. Pick a different foundation and the right column is what changes; that is the argument the product is making, and it is only an argument if you can watch it happen.
What has to happen before 2b can happen at all. Asked at the moment it is wanted rather than on first launch, because a permission asked before somebody wants the thing is a permission denied, and on most platforms denied permanently. Our two sentences go before the system’s, since the platform dialog cannot say the audio is thrown away.
A sheet, not a screen, and for the same reason 2b is one: this is a flow through the kitchen and the reader comes straight back to it. The way out is a door and not a dismissal — typing carries the same weight as accepting and the line says it arrives at the same place. Declining costs nothing here, because no system prompt has fired yet.
The typed twin of 2b, and now on the same component, which it should have been from the start: one of a pair drawn as a sheet and the other as a screen was the clearest thing wrong with this run.
It is one open description rather than a question at a time — the question-at-a-time path in 5a is the slower fallback, not this. The same prompt, no character count and no formatting, and it hands back to talking mid-thought. What you leave out becomes a question later, which is why nothing here is a form you have to finish.
The step between talking and 5c. Your own words come back verbatim once, at the moment they can still be corrected — a different job from 5c, where the quote would be a transcript to re-read rather than something to check. The elapsed time from 2b is not here and never will be.
A document rather than an answer, so this is the second sheet in the file whose action scrolls with the content instead of pinning: five named sections ending in a way out, and a pinned bar over them would be a permanent control above material it is not acting on. Each answer carries what it did, including the one that added a decision. An inference is confirmed rather than recorded, and scope is the question after it, never folded into it.
Where 3c's one open target goes, and the frame the whole run was missing: the coach leading rather than asking. It never asks how wide the island should be, because nobody can judge a dimension in the abstract and a guess looks exactly like an answer. It asks about things you already know — where you stand, how far you walk, whether you could hand a plate across — and the number falls out. Your turns are blue because they are your side of the conversation and not a state anybody reported, and they hug their words rather than stretching to the edge. The count is what has been answered and there is no denominator, because the flow ends when the field's own questions run out rather than at a number. Where two of your answers cannot both be true, it says so and hands you the sentence to say to your architect. The composer is always there, so going off script uses the same door as the opening description.
The other end of leaving 3b without finishing, and it is its own frame rather than a change to 2a: a room nobody has been asked about and a room holding an unfinished answer are different rooms, and 2a's premise is the first one. Leaving is not a failure state, so the draft is not a warning — it is the next thing to do, and it leads. Its own words are what identifies it, so the card shows them rather than a file name and a time. The first question is still here, demoted to the alternative it now is. Nothing has moved into the room, and the note says so, because a draft that looks recorded is worse than one that looks lost.
3e on the desk, in the same two columns as 2d. Left column is the room and what is not chosen in it; right column is the unfinished answer, leading, with the first question below it as the alternative.
The same content river as W, given more width and holding one question instead of eight decisions. Left column is the room and what is not chosen in it; right column is the question, what will be asked, and what the asking does to the room's decision list. The two runs W fills with decisions are absent rather than drawn empty, so nothing on the screen is a container apologising for having nothing in it.
2d with the same sheet 2b uses, side-anchored because the width allows it. The room stays on screen while it is answered, which is the whole reason this is a sheet and not a screen.
The case the build is missing. The room has no decisions in it because nobody has been interviewed about it yet, so the screen leads with the description that fills the room rather than with four zero rows. The plate carries the three faces of the room that are not chosen, which is what "nothing settled" actually looks like.
2a with the answer open on top of it. Talking is not a screen: you cannot afford to lose the room you are answering about, which is the one job the system's overlay exists for — bottom-anchored on the phone, side-anchored on the desk in 2e, the same component both times. Closing it keeps what was said; the way out sits beside the way on, and typing is the same pipeline rather than a second screen, so it can take over mid-thought. The dock goes while the sheet is open — a dock would claim this is a destination, and it is a flow.
The point of the interview, reported. Two minutes of talking decided which of the fourteen kitchen decisions apply, answered the ones the answers imply, and left three that genuinely need a choice. Every answer carries the sentence that settled it, so an inferred answer is citable rather than assumed. What the conversation ruled out is not drawn as a run of struck-through rows: a decision that does not apply is absent rather than hidden, and one line says what went and why, because an absence is invisible unless something says it out loud.
The scoping mechanic at single-decision grain, and the reason the interview can rule things in and out. The options are a run inside a card, so the marker carries selection and each one states what it locks — the consequence is authored content the screen reads, not a sentence composed about somebody's foundation. The head says five decisions depend on this answer, never that five unlock: a full basement opens five and closes none, a crawl space opens one and closes four, and a slab closes all five, so a number in the headline would be a lie on two of the three. This belongs on the decision screen rather than the room; it sits here so it can be read against 2c.
2f settled. Five decisions that did not exist a moment ago are now real, and the run says which answer made them real. The rows are chevron rows like every other decision run in the app — a blue "Open" on each line is a bare word attached to nothing, and the primary already says which one to start with and why. Five arrivals with five deadlines is a to-do list, so only the one with a real clock on it is marked, and what the answer closed is reported beside what it opened.
The room on the desk canvas. The house name sits in the rail, where it already was, so the room's own name is the only title on the screen. Left column is the room and the next answer in it; right column is the decision runs, marked by who holds each one. What has been kept in the room runs full width underneath, where a wall of tiles has the space to be a wall.
The accordion is gone. The room leads with its next answer on the room's own plate, then the eight decisions in named runs of four or fewer, each row carrying the mark that says who holds it — which is the thing the old screen could not tell you without two taps.
No badge. Where a space sits is a sentence in the head's meta line, and what it is linked to is drawn the way the house board already draws spaces — the two sibling rooms as room cards, with this one marked "You are here". Under them, the answers that are given once and land in all three rooms, which is the part a bedroom screen currently cannot say at all.
Bathroom 1, which in the current build reads "0 photos, 0 colors" and "0 answered". The pair to 2a, and the two are told apart by who acts next: this room is waiting on a date, so it carries no action and says when its decisions open; 2a is waiting on you, so it carries exactly one. A run that holds nothing is absent either way — the zero rows are gone.
The way into Y, and it is the room's existing actions menu rather than anything new on the screen. It is a sheet like everything else that sits over a room, because this is a menu and the system has no dropdown. Only the account owner sees these rows at all — a non-owner gets the menu without them rather than a row they cannot press.
Linking is yours to set. Anyone else on the build sees this room and its answers, not these rows.
Linking is an answer, so it happens in the system's one overlay — you do not lose the room you were in. The candidates are the house's other spaces as a row run, the marker carries selection, and the action states its own scope rather than saying "Done". The line above it says what linking does to answers that already exist, because two of these rooms have already answered one of the shared decisions.
Linked spaces are asked once for the answers that only make sense once — tile, paint, trim. Everything else stays per room.
Between Y and the link actually being made, because linking two rooms that each answered the same question has to pick one and must never pick it quietly. Both existing answers are shown with the room that recorded them, and choosing is the whole screen rather than a warning to get past — nothing here is red, nothing says overwrite, and the answer not picked stays on its room. The action cannot state its scope until both are settled, so until then it says what is still missing and does not draw itself as the primary. Try the rows.
The spaces you are linking answered the same question different ways. Pick the one the suite keeps.
The same control reads and undoes the link: turn a space off to take it out, turn all but one off and the suite is gone — the action says which it is. Unlinking is not destructive and is not drawn as if it were: nothing recorded changes, the two shared answers stay on all three rooms, and from then on each room is asked on its own. Try the rows — the button follows.
Three spaces, linked since 4 March. Turn one off to take it out of the suite. The space you are in stays.
The screen is a filing cabinet for a room. Everything real about the room is one tap behind a category name.