Now, as a week

Four of the five slots went to one order from one supplier, because cabinets is a single workstream at sort_order 350 carrying seven library decisions, and four of them are open in this kitchen at once. The ordering is correct — these genuinely are next — and the week still reads as though the house has stopped being a house and become a cabinet catalog. Nothing in the bathroom, nothing outside, nothing the reader can switch to when they stall on cabinet doors. The freedom the period was supposed to buy is gone, spent on adjacent variations of one question.

Island electrical is the one decision with a real needed-by inside the period, so it takes the top rung of the hero and the card shape 10 already earned for it, teardrop intact. Everything under it is the ramp, as rows of equals. The hierarchy is date or no date — not importance, not a score.

Settled by Thursday $0 against Settled after $1,200 is the most persuasive pair of figures in the product, and it is 102px on a card that now has a list under it to protect. Both figures move one tap in, to the decision itself, together: the pair rule says the two are meaningless apart, so this is a removal and never a trim. What survives here is the consequence in words — after that it's a wall to open — which is the same argument at a tenth of the height. The rule for when the pair is drawn is now positional rather than per-screen: the hero shows the pair when the hero is the only thing on the screen, which is true on frame 8 and on no other frame in this deck.

Settling something removes it and pulls nothing in. A steady-state five that always reads five is a queue, and a queue never feels finished; a set that empties as you work it is the only progress signal this screen is allowed, since it may not print a fraction. Carryover takes priority over new arrivals at roll-over, so a week nobody touched cannot hand the next week ten.

10 put it first among the bands, and it was right for the screen it was written on: one card, one band, and leading was the only way to be seen. Against a five-row week it cost the entire ramp its place above the fold to say one thing about one decision in a mudroom. It is unowned, which is a fact about who is on the hook — not a claim that it is next, and lane order on this screen means next. It keeps its own header, its Take it affordance and its guarantee of never being omitted. It just stops going first.

Three or more open decisions sharing a primary workstream collapse into one row, and the cap counts rows. The three slots that frees go to the next three items in ramp order, which is how the same five-item week grows from one room to two and from one trade to four. Nothing was hidden and nothing was reordered: the four cabinet decisions are still first, still next, still one tap away.

Two rows naming themselves are clearer than one row hiding two, and a reader who opens a cluster to find a pair has been made to work for nothing. Three is where the label stops being a lie about how much it saves you.

A row is now sometimes a decision and sometimes a container, which is a thing the reader has to learn once. It is paid for with the ring, the count and the chevron — three signals on one row, which is more than any other row on this screen carries, precisely because it is the one row that behaves differently when you touch it.

Rooms are the obvious grouping and the wrong one. decision_zones is many-to-many — the island electrical on the card above is Kitchen and Electrical — so a room grouping has to duplicate rows or pick a winner arbitrarily. A workstream is what a decision feeds, exactly one of them is primary, and it is already the axis the ramp is sorted on. Room stays where it reads best: the caption, and a filter.

Every date on frames 1 and 2 is aspirational. decisions.due_date exists as a column and is populated on nothing; lead_times is empty; builder_lead_time_days is 0 of 222; projects.current_phase_id has no writer anywhere in the new stack. A design that needs any of them to draw a first screen does not ship. This one needs none: workstream sort_order is authored content that is already in the repo.

This frame originally opened on a bare list, argued as a principle: the hero is what a date looks like, so with no dates the screen raises nothing. It is a coherent rule and it produced an unfinished screen for the only state that actually runs. A week is a thing you plan, and a plan opens by saying where you are. The hero is now unconditional and this is its floor: it names the rooms and it says nothing is late. It still does not promote a decision, still does not manufacture urgency out of ordering, and it is still true that a white teardrop card with an amber chip means a clock. What changed is that the absence of a clock now has a shape of its own instead of a hole where one goes.

This rung was a sage card for one draft, on the argument that the tint vocabulary already says settled and the reader could then tell the two hero states apart at arm's length: white and amber is a clock, sage is a plan. It was coherent and it is superseded. The hero is now one shape with two fills — the same white teardrop card on every frame, carrying a kitchen elevation with an amber chip and a CTA when something has a clock, and the house with a single line when nothing does. The distinction the tint was making is made by the drawing instead, which is louder, is legible at arm's length for the same reason a photograph is, and does not spend a color that the rest of the deck needs for settled decisions. No frame on this page draws card-sage now.

The single focus card said this one, now. It was a ranking presented as an instruction, and a reader who did not want to do that particular thing had nowhere to go but away. Ordering still happens — these five are genuinely the next five — but within the set the choice is handed back. That sentence used to sit under the list, where it read as a footnote to an instruction already given. It is in the hero now, above the list, where it is the frame the list is read through.

The rows below already carry a zone marker each, so the strip is only worth its height if it says something the rows do not. It does: three icons is the answer to "is this week one room or my whole house", available before a single row has been read. Frame 1 would draw one icon here and frame 2 draws three, which is the argument those two frames exist to make — and the only place on this page it is visible from inside the phone rather than from an annotation beside it. The title above it deliberately does not name the rooms as well; it did in the first draft, and a card that says the same thing twice in two shapes is 56px of nothing.

10 frame 2 carries a contingency bar under a Your money header, ruled 2026-08-10, and this page quietly dropped it while replacing frames 1 through 4. Not a decision, an omission — found by reading the frames it claims to replace rather than by anybody noticing it missing. It is back, verbatim, on every frame that draws a full week.

It also belongs here more than it did on the single card, because the contingency pool is where the hero's second figure comes from. Settled after: $1,200 and Committed: $5,250 are the same money seen at two distances, and a screen that shows what a late decision costs and not what is left to absorb it has told half a fact. The loan bar stays on 13: a loan is a draw schedule, and this screen is about decisions.

The first prohibition bans fractions and fills, and .segbar is a fill. It is not a violation and 10 already worked out why: the ban is on a fraction against a moving total — a contract sum, 26 of 47 — because that answers how far through am I, which nothing can be done about. Contingency is a fixed pool set aside once, where spending against it does not grow the total, and how much is left is a question the reader can act on today. Two of the product's figures earn a fill, this and 13's loan, and both for that reason.

Re-measured 2026-08-16, after the permanent On site now block and the hero drawings landed. The first ramp row now lands at 853px against a 748px fold, so none of the five clear it. Three things moved it: the block costs 70px at two categories and pushes everything under it down, the hero took a 16/9 drawing, and the rooms strip left the hero for a card of its own. The same measurement across the page — frame 1's first row at 798, frame 2's at 821, frame 9's at 798, frame 12's at 773. Every full-week frame now opens on the hero alone.

Ruled 2026-08-16: the screen stands as drawn, and it is the fold rule that is retired rather than the screen that failed it. The draft this replaced put the first row at 396 and cleared four of five, and this paragraph argued the numbers were the whole point: a week view that has to be scrolled before it looks like a week has already failed. That was written when the hero was the only thing above the list and the list was the only thing on the screen that said week. Neither is true now. What the 457px bought is what a week view opens with: where you are, what is on site, and a picture of it. A plan says where you are before it says what to do, so the rows are the second thing read on purpose, and a thumb reaches them in one motion.

The measurements stay because the next person to add forty pixels to this head should have to see what the head already costs. Two hundred are recoverable by dropping the drawing on the undated rungs, and that is the first thing to trade if anything ever has to be. The block is the last, because it is the only writer this screen has.

A decision's phase says when it can be decided; its primary workstream's sort_order says when the work it feeds happens. Kitchen cabinet style is a design_precon decision pointing at cabinets at 350. The gap between the two is the runway, and sorting on the second means the queue is always drained in the order the site will consume it. No dates, no schedule, no estimate — and still nothing arrives late.

Every label here is the authored category value verbatim — Interior Finishes, Fixtures, Electrical, Exterior, Inspections, Closeout — sentence-cased and otherwise untouched, because a chip set that quietly rewords its own source is a mapping nobody wrote down. If any of them needs homeowner-facing language later, that is a rename in the data or a labels table, not a rewrite at the frame. workstreams.json carries twenty category values across 157 streams, which is a menu rather than a question. Filtered to the phase the project is in and the next one, it lands at five or six every time. The next phase is included on purpose: builders start the following phase early far more often than they finish the current one cleanly, and a chip set that cannot express that teaches the reader the question is not worth answering.

A homeowner has no use for slab_prepour_inspection and should never be shown it. The rollup is the product; the streams underneath are how the app does its arithmetic. This is the same argument the Systems axis already won — seven curated labels over an engineering vocabulary, and the mapping is the point.

No dates asked, no percent complete, no "are you on track". The failure mode this feature has to avoid is becoming a status report the homeowner feels responsible for filing — which is micromanagement wearing a friendly chip. Same as last week is a first-class answer and sits beside Skip, because the two most likely truths on a Monday are "the same thing" and "I have not been out there".

The first version of this frame was a screen with no route into it, drawn and reachable by nothing. The second gave it the hero's slot on the first open of each period, which was a real answer and a narrow one: it made the question an event, so correcting a wrong answer on a Thursday meant waiting until Monday or hunting for a control that did not exist. Ruled 2026-08-15: the block is permanent and the answer is editable at any time. There is no route into this frame because there is no frame to route to — it is the same block frames 1 through 12 all carry, with the off categories brought in. What used to be a weekly interrogation is now a fact sitting under the project name that happens to be tappable.

Closed it states what is on. Open it toggles all six. That is the whole component — no sheet, no modal, no expansion panel, no second chooser. The deck has no sheet to borrow (.scrim exists only behind 20's nav dock) and this needed none, which is the strongest argument for the shape: it is the only version of this that adds no new component to the system. The chips open in sort_order, the build sequence, so nothing moves between the two states except by its neighbors arriving.

--surface-container is exactly what .asof.act puts behind a sentence the moment that sentence stops being read-only. Same meaning borrowed here: there is now something to do in this block. And the reassurance travels with the toggles rather than with the label — nothing goes to your builder is only load-bearing at the instant somebody is about to tap, and printing it permanently would make a calm fact look like a disclaimer.

Open costs a flat 177px whatever is selected, because open always draws all six — so the reflow is the same every time rather than depending on how busy the site is. The hero and a ramp row stay on screen throughout, which is what says this is a question about your week, and your week is right there rather than a gate to clear. Measured, and it is tight: the dated hero's CTA lands at 724px against a 740px viewport once the dock is allowed for. Sixteen pixels. Anything added to the helper line or the chip padding puts Settle it now under the fold.

Where the builder is actively using their side, the same signal is inferable from what they are already doing — sends, draws, capture ticks, trade visits. The chip screen is the fallback for the homeowner who has no connected builder, which is most of them, rather than the primary path. Asking somebody a question the product can already answer is its own small insult.

A project-level pulse row, which refines projects.current_phase_id (no writer today) and seeds project_phase_status.started_at (no column today). Those two, plus restoring phases.typical_min_days and typical_max_days, are what let the period tighten from bi-weekly in design to weekly once the site is moving. Until they exist the period is a constant, and the screen still works — it is just never bi-weekly.

now-model.mmd makes dontknow a first-class outcome and states the rule as "a gap must not render as reassurance". VIEW_MODELS.md claims the state exists. 10 frame 4 draws it. The shipped now_screen.dart renders Nothing needs you this week whenever the focus is null, with nothing distinguishing a computed quiet week from a project nobody has ever walked. Grep of app_en.arb and lib/ui/: no such string, no such widget.

A capped ramp is rarely empty once decisions exist, which is exactly why this frame has to land with the redesign rather than after it. The emptier a state gets, the rarer it gets, and the rarer it gets, the more confidently the wrong copy sits on top of it. A new homeowner told "nothing needs you" by a screen that has never looked is the same bug the single card had, wearing a friendlier face — and this product exists to reduce that person's anxiety, not to earn their trust and then spend it.

Never walked is this frame. Everything blocked says what it is waiting on, because "nothing needs you" is false when four things do and all four are stuck behind one prerequisite. Genuinely quiet is frame 6. One condition per screen, and the shipped code currently collapses all three into the third.

A screen built to hold five things has an obvious way to fill an empty week, and the ramp can always produce another row — there are twenty-one open decisions on this house. Every one of them would be honest and none of them would be needed. The cap is a ceiling, not a quota, and the ramp only offers what the work is genuinely about to consume. When it has nothing, the screen says nothing, exactly as it did before.

Three weeks, and the trusses, every single time. Unchanged and unchallenged by anything on this page.

.oc.none, styled in this page's own sheet rather than the shared hc.css, has carried its rule from the first draft that split the two nothings apart, and no frame ever exercised it. The implementation shipped it from the rule alone: a plain sentence, no ground, no dashes, sitting on the same baseline the chips would have. This frame is the source that rule was always missing.

Frame 6 pairs a quiet site with a quiet week, which reads as one fact wearing two labels — a reader could reasonably conclude the sentence follows from the ramp being empty. It does not: site_activity and the period's own selection share nothing, not a table, not a query, not a write. This frame keeps frame 1's busy ramp under a quiet site on purpose, so the two facts are visibly not the same fact.

Frame 5's control asks, because there is nothing yet to amend. This one offers the ordinary Edit, the same link frame 1 and frame 6 carry, because there is something to amend: a real answer that happens to be "nothing". The link is how the reader tells the two states apart at a glance, before reading a word of the sentence beside it.

Every other full-week frame on this page carries a measured first-ramp-row position — §7.1's ruling rests on those numbers. This one is frame 1's layout with a one-line block in place of frame 1's two-chip one, which should land the row a little higher than frame 1's 798px, and that number was not taken. Recorded rather than assumed, on the same page whose own rule is that a number a doc states about itself has to be re-measured.

Cap of one, and every other rule holds: the same two lanes fill it, the same ordering picks the winner, and it still does not refill until the week ends. Nothing here is a special case in the view model, which is the only reason the option is affordable at all — two screens that answer the same question is how a product ends up with two answers.

The shipped Now shows one decision and never admits to a second. That silence is what a reader fills in themselves, usually with something worse than four. Even at a cap of one, naming the number and offering the way out costs a sentence and removes the whole failure mode.

The pair comes off the hero on the phone's week, because there it has a list under it and 142px of figures and working costs that list its place above the fold. At a cap of one there is no list: the hero is the screen, and the height is free. Stated positionally so it needs no per-frame exception list — the hero draws the pair when the hero has a column to itself. Here that column is the whole screen; on frame 11 it is the desk's left column; on a phone at cap five the hero shares its column with the week, and the figures move one tap in. One rule, three frames, and no mention of a device in it.

Not the default, and not recommended in the settings copy. It exists because some people genuinely do work best with one thing in front of them, and because a product that took the single card away without leaving a door back is telling those people they were holding it wrong.

The hero is not "the dated ones", it is the soonest one. Two heroes is two answers to what needs me first, and a screen that gives two has given none. Everything else in the dated lane drops to a row and keeps its date on a chip-time, which is the same amber the hero's own chip uses — so the band reads as the same kind of thing at a lower volume rather than as a different feature.

A dated decision and a ramp decision are answers to different questions: this has to happen by Friday against this is next in the order the work needs. Sorting them into one list would make the reader read every date to find out which rows have one. The band costs a section header and removes that scan entirely.

Five is five: one hero, one dated row, one cluster and two ramp rows here. If six things carried dates, the ramp would be empty that week and the screen would be right to show it — a week with six real deadlines in it does not also want to be told what to get ahead on.

This slot used to hold a version with no hero at all: the date became a chip on a tinted row and all five items sat as equals. It got five rows above the fold, which was the honest advantage. Ruled 2026-08-15: the hero stays, and it is now unconditional — the opposite call, and the argument that settled it was that removing the hero fixed one frame and left the four without dates looking unfinished. The fold cost it was answering is real and was paid down three ways instead: the card lost its room drawing, then its late pair, and Nobody has this one moved to the bottom. Together those are worth more than the hero was.

The fact lived in a 230px band over the drawing for three passes, and each pass invented a way to survive six categories: cap at two, a +N counter, a ladder that stopped listing at five and said the whole site is active. All of them were measured and all of them fit. All of them were also solving a width problem that only existed because the fact had been put somewhere narrow. Under the project name the head is full scroll width and can grow downward for the price of a little scroll, so the set wraps and there is nothing to rule on. Four categories is two lines. Six would be three, and would name every one.

Closed, this block is 58px at one category and 98px at four — the difference between this frame and frame 1 is 40px and one line of chips. Open it is a flat 177px whatever is selected. The band cost nothing above the hero, which was its one real advantage, and it is the advantage being traded away here: the hero starts 116px lower on a busy week. Everything that matters still clears the fold, and on the dated hero it clears it by sixteen pixels.

Chips with an overflow chip is what the field does — FotMob and BitePal both draw it. Inside the hero it does not transfer, because the card already spends its chip vocabulary on chip-time and chip-count and you get two rows of pills in two languages. In the head it survives, because the drawing sits between the two zones. A six-segment strip lit where the work is fixes the width by construction and shows what is not happening, which nothing else did — but it names nothing at any count. This block gets the strip's one advantage for free: open it, and the unlit chips are the negative space, with their names on them.

Whether the 3–4 rung should lead with the two longest-running categories rather than the first two in sort_order. Sort order is the build sequence, so it currently names whatever is earliest in the phase — which on a busy week may be the thing that has been true for a month and is the least worth saying. There is no started_at on a pulse answer today, so it is not buildable, and it is recorded here rather than solved.

A real screen with its own title and its own room for context — the same "room screen's context for free" the page's closing note already names. It is also the only option here that costs a new route: /now/cluster/:workstreamId, a new query (open decisions by primary workstream) and a new empty state for the case where the last of the four just settled while this screen was open.

A tap to get here and a tap to leave, for four rows that were already legible as a count on the week itself. The week is not visible while this is open, which is the thing frame 14 avoids and frame 15 does not need to avoid.

The reader keeps their place. They can settle Cabinet doors, come back to the same scroll position, and settle Drawer fronts next, without the week ever leaving the screen — the "without losing their place" case the page's closing note already names. No new route, no new query: the four rows are the same now_period_items the closed card already held, just re-drawn.

The card grows by four rows in place, which pushes Nobody has this one and Your money further down a screen that is already the one this whole page fought the fold over. And it teaches a second interaction on a screen that otherwise has exactly one: tap a row to settle it. This row now also expands.

Zero new screens. The row keeps its badge and its chevron — it still reads as "four, tap to act" — but the tap reuses /now/decision/:id, the exact push every plain row already has. The only change from what ships today is the caption, which now says which one you will land on instead of who is doing the work.

The badge promises four and the tap delivers one. A reader who wanted Toe-kick trim and got Cabinet doors instead has to settle or back out to find it — there is no browse step to route around it in. This is the option that treats the cluster purely as a queue-jump, not as a room worth entering.