One Question at a Time

Where the patterns came from

Grounded per the standing directive, then repainted in LADS.

• Lovi: the reason for the question sits directly under the question, and every option carries a second line explaining it. This is the shape of frames A and B, and it is the one thing the shipped renderers cannot currently do: they draw the label as an 11px uppercase eyebrow and never render the help text at all.
• Alan: the same anatomy for a multi-select, with the explanatory line per option rather than per screen. Frame C is this.
• Hers: question as a real heading over full-width answer rows, not a form field with a label.
• Ulta Beauty: an explicit skip that sits beside the primary action instead of hiding in a corner.
• Monese: included as the pattern we are rejecting. Its step reads "Question 5 of 5". See the first note below.

Design calls worth arguing about

• A count, never a position. Every questionnaire pattern worth copying shows progress: a segmented bar, three dots, "5 of 5". All of them need a denominator, and this flow does not have one. Answering that two people cook adds an aisle question, so the total grows as you go. The strip says "Six answered so far" and nothing else. This is the same rule as the space screen, and it is why the .steps component declared in this stylesheet is deliberately not used here: it needs a denominator, and this flow has none. Onboarding and Vision are bounded, so they can carry a position, and they each draw their own dots rather than using .steps, which currently has no call site anywhere in the set.
• The help text is the product. "Drawers cost more and bring everything out to you. Doors cost less and ask you to kneel down and reach into the dark." Nineteen of the twenty-two authored questions carry a line like that, and today five of the seven field renderers drop it on the floor: only text and long_text read helpText at all, and they spend it as a hint inside the input. A bare pill list of four labels is a form. The same four labels with that sentence above them is a coach. The gap between those two is most of what the six-week proof is trying to demonstrate. Worth knowing before reading nineteen of twenty-two as nearly done (OPEN_QUESTIONS.md Q44): that count is the curated authored set. The live field table behind it is larger and thinner — 100 rows, 21 with help text — so the coverage gap this page is arguing for is wider than its own two numbers suggest. The rebuild's schema now takes this page's side (DB_DIAGRAM.md §2, 2026-08-11): help_text is NOT NULL on library_questions, so a question missing its reason fails at load instead of rendering as the bare pill list, and frame B's per-option consequence lines have their own table.
• Nothing is preselected. Frame A shows four options and no selection, matching the Dojo player's existing rule that auto-selecting on load reduces the flow to a tap-through. A default is an answer nobody gave.
• A boolean is two sentences, not Yes and No. Frame B never renders the words yes or no as the choice. "Do two people need to pass behind each other" answered "Yes" is not recoverable six months later; answered "Yes, that happens most evenings" is. The shipped boolean renderer hardcodes Yes and No, in English, and it is wrong for every question in this set.
• More than four options is still cards. "Who lives here?" has five, and the shipped single-select renderer silently falls back to a raw Material dropdown above four. A dropdown hides the options, and the options are where the thinking happens. Frame C carries five and stays cards.
• Blank is an answer, so it gets a sentence. Three authored questions say a blank answer means something real: no island, clear counters, or kids who only eat in there. Frame C gives that its own row rather than leaving the user to press Save with nothing checked and hope.
• The button appears when there is something to confirm. Frame A has no primary action, because nothing is chosen yet and a live Next button on an untouched question is a nag. Frame B has one, because an option is selected and the tap that advances should be a separate, undoable act rather than a hair trigger on the option itself. So a single-select question is two taps, and that is deliberate: the first tap is reversible and shows you what it costs.
• Skip is offered, and it is kept. "Not sure yet" is a first-class outcome, and frame E accounts for the two that were set aside instead of pretending the room is finished. The plan says a low answer rate means a question is misplaced, which is only measurable if declining is a recorded act rather than a bounce.
• Travel is asked once and named. Frame D names the four rooms this house actually has, not "everywhere" and not "this room only". "Everywhere" is a promise about rooms that do not exist yet, so it is worded as one, and the receipt in frame E repeats it. Rooms this project does not have are never mentioned, even though the question is authored for six.
• The travel answer is one fact, not four copies. Storing "drawers, house-wide" as four per-room rows means a room added next month gets asked again, and an edit in the kitchen silently disagrees with the laundry. Stored once at the house level, a per-room answer simply overrides it, which is what the existing answer-precedence rule already does.

Notes on the drawing

• Nothing was compressed to fit, and the count survived the retrofit. The page now opts into body.hc-air, so cards are 26 by 24 and the gaps between the question, its options and the action are 24 rather than 13 to 17. Frames B and D run past the bottom of the phone. What did not change is the strip: it still carries a room and a count of answers, and there is still no bar, no dot row and no denominator anywhere on this page.
• The count was dimmed and is not any more. .strip .n carried opacity:.8 over .c-mute's --on-surface-dense, which is an alpha on top of a tone that is already the quiet end of its pair. It measured 4.32:1 on --surface-lowest and failed AA at 11.5px; undimmed it is 7.01:1 and the strip reads no louder. .q-hint moved from --on-surface-variant to --on-surface-dense for the same reason: it is 13px, and that is where hc.css hands the muted tone over to the darker of the two.
• Frame D's room chips had to be renamed. They were .rooms, which under body.hc-air is the two-up grid of room picture cards from mockup 13. They are .trav-rooms now. hc.css names this collision at the declaration; opting in is what made it reachable.
• No frame carries the nav bar. The stepper is a takeover opened from a room, with its own back chevron and its own X, and frame E is still inside it. "Back to the kitchen" is the way out, and it is the screen behind this one that carries the chrome.
• Two pointers on this page are stale and are left as they are. The deck was renumbered — the pre-rename set is still on disk under docs/_archive/2026-08-03/adaptive-mockups/, where 13 was the binder and 01 was the space conversation — and cross-page pointers did not move with the files. The fast door above was corrected to 02, which is measurable: 01 is the front door and has no composer. The two mockup 13 references below cannot be corrected the same way, because no page in the current deck prints the binder's kitchen tab: 48 inches appears on this page and nowhere else. Where that content now lives is a decision, not a renumber.
• The kitchen's answers are the fiction's. Most nights from scratch, two cooks most evenings, and 48 inches of aisle, word for word as the binder's kitchen tab prints them in mockup 13. That is the point of the slow door: it writes the same rows the conversation writes, so a homeowner cannot tell afterward which one they used.

This flow is the slow door. It is not a fallback for a failed conversation and it is not a form to be completed: mockup 02 is the fast door, the composer is always there, and both write the same rows through the same upsert. Nothing in the middle of the system knows which one you used.