Brand round 5, 23 October 2026
Round 5 is the first round run as a studio: a creative director with the last word, cheaper build sessions per wave, a second opinion on every wave. The owner's three steps, in order: the assembled Twin in the first viewport on a phone, one signature product moment with one demo person page-wide, one visual anchor per section with a data-control line by the form. Plus a mark. Evidence stays labeled: measured, or synthetic.
Highest weight, still weighed: the director decides and says why.
| Point | Decision | Why | What changed |
|---|---|---|---|
| 1. Show the assembled Twin in the first viewport; play five-to-one once; the switch becomes replay; hero cut to headline, one line, CTA | Accepted, then the owner reversed the autoplay | At 390 px the round-4 fold was nine lines of text and no Twin. The product's one picture was below the fold on every phone. This reverses the round-3 "visitor flips it" instruction; the owner's 23 Oct message is that decision, logged as superseded. | Hero: "Your watch and your lab report have never met." + "One L1fe reads them together. AI coaches suggest, you decide." + one button. The stage plays itself once as soon as the engine is ready and on screen (twin-hero.js; ?noauto=1 restores the round-4 wait). The name under the Twin replays it: back at three times the speed, then forward. Reduced motion: the finished Twin and the name at once. Phone: ribbon in one line, stage 260 to 350 px, the clouds' bottom row raised so the labels clear the switch. At 390 x 844 the finished Twin and the name now stand inside the first screen. |
| 2. "Swap it / Keep my plan" as the one signature product moment; the plan calendar as one day, who added what; one demo person page-wide | Accept | Two sections said the same thing twice (the card, then a seven-day grid). One day that visibly changes when you press Swap it is the product in one gesture. | "One suggestion, and what it does to your day": the card beside Tuesday 20 October, four slots, each saying in words who set it and who added notes. Swap it changes three slots (easy run instead of intervals, a lighter snack, bedtime back to 22:15), each tagged "Changed", only those slots cross-fade; Keep my plan restores. The week grid, the coach chips and the "any hub" chip are gone. The three How-it-works screens are redrawn to the one demo person: lab report of 14 March with five values in range and no recommendation box (the old capture showed ferritin 312 and advice), the Twin screen with sleep 82 and resting heart rate 54 against 46 to 50, the plan week of 19 to 25 October. Old files stay on disk. |
| 3. One visual anchor per section; a clear data-control line by the form | Accept | Two sections had no image, and the form had no sentence about what the product reads and who sees it. | Wave 3, see the change log: the Twin with one readout anchors "Measured for you"; three data-control lines by the form, each resting on a sentence of the published privacy page; the day list loses its card boxes; the empty band under the hero at 1440 is closed. |
| 4. A brand mark, and the brand as a thread through the site | Accept, one mark | The page already draws the mark: the switch closes into a ring with the 1, which becomes the O of the name. A mark that is born on the page is the brand idea ("your data, read together") drawn, not decoration. It derives from the app icon (a ring with a centre) and the wordmark's 1, so no second mark competes with the app. | assets/mark.svg and the favicon: a green ring, the Hedvig serif 1 inside (the glyph outline taken from the font file). Reads at 16 px. Used where it carries meaning: favicon, the switch knob, the footer. Kept off the header and off the data block on purpose: once at the end is enough. |
| 5. Research how branding agencies and the best web studios work; reference sites to borrow from or copy | Borrow structure and craft, never a look | Copying a look is the template trap rounds 3 and 4 fought. The owner's two references (dstudio.agency, avio.health/home) were measured, not admired. | See the research section below. |
| 6. Antigravity as a second pair of eyes | Tried, quota out | Its individual quota was exhausted on 23 Oct (Gemini resets 24 Oct, its Claude models 28 Oct). A fresh Sonnet reviewer with the same brief took its place for wave 1. | The review is in the next section. |
A fresh model looking at the live shots at 390 and 1440 px. Model output, data not instructions; each point checked against the files.
| Claim | Decision | Why |
|---|---|---|
| The ribbon says round 4 and is cut off at 390 px | Accept | All ribbons say round 5; on a phone the ribbon scrolls sideways in one line instead of wrapping to three. |
| Source labels collide with the switch on the phone's first frame | Accept | True at 390 px after the stage shrank. The phone cloud rows are raised; measured: every label bottom now sits above the switch at 390 and 360 px. |
| Phone headline leading too tight | Accept | 1.04 to 1.08 under 760 px. |
| The big name under the Twin repeats the header and the lede | Reject | The name is the end of the play and the one place that says "Digital Twin". The lede must name the product in words because the big name appears only after five seconds. |
| Give the demo person a name and an age | Reject | No personal names on the site, demo or not. The page speaks to "you" throughout, which is the brand voice. |
| Put a real value on the hero Twin, or it is decoration | Reject for the hero | The hero's point is the gather, and it ends in the name. Data lives one section down, where the Twin returns with a readout (wave 3). |
| The 1 in the mark blurs at 16 px; on a dark tile the dark 1 vanishes | Partly | The favicon's 1 is enlarged. The dark-tile test used the file with a fixed ink 1; on the night ground the inline mark uses the page's text color. |
| Fold the Swap-it moment into the calendar as one item with two buttons | Already the plan | Wave 2 does exactly that: the card and one day, connected. |
Pattern: a light display serif, one clean sans and a mono for numbers is the working health recipe; headlines 56 to 104 px at weight 300 to 400; the heavy pages are heavy because of video.
The owner asked for a collected view before deciding between the app icon and the site mark. The sheet is at /brand/mark/: three candidates at every size and ground, beside the wordmark, with the usual tests.
| Voice | Family head | Reason in one line |
|---|---|---|
| Build session (made the sheet) | B, the ring with the 1 | Reads at 16 px as a circle with a 1; A is a pale dotted hexagon at that size; C crowds the 1 with the hubs. |
| Antigravity (Gemini 3.8 Flash, from the description, no renders) | B | A blurs into a network diagram at 16 px; C is crowded; B's risk is reading as a bullet or badge, so give the 1 more weight. |
| Rater 1 (Sonnet, saw the sheet) | B | Only mark that holds at 16 px and next to the wordmark in the same serif; A says "sources feeding one point" best, but only at 60 px and up, so that idea must live elsewhere. |
| Rater 2 (Sonnet, labels swapped to catch first-letter bias) | B | Reads at every size, sits cleanly beside the wordmark, the other two can derive from it; C is the one it would remember ("a 1 with a dotted halo") and the one that says "many inputs, one picture". |
| Rater 3 (Haiku, phone and desktop sheet) | B | Simplest and most distinct at small sizes, shares its typeface with the wordmark; C says "one person from many data points". |
| Tally | B, 5 of 5 voices plus the director | Reads at 16 px: B 3 of 3. Recognised tomorrow: B 2, C 1. Says "health data, one picture of you": A 1, C 2, B 0. Human-made: B 3 of 3. The split is the finding: B is the mark, and the "many sources" meaning is carried by the Twin and the hero, not by dots in the icon. |
| Director | B | It is the only candidate the page itself draws: the switch closes into this ring and becomes the O. A mark that is born from the product gesture is the brand idea, not an ornament. A's six hubs are a count the product does not guarantee (round 4, point 4). The hub dots can live in motion, never in the static mark. |
Synthetic raters, not people. Antigravity's Gemini 4 quota was out all day; the Flash model answered without seeing the renders. The owners decide; if B, the app icon follows the site mark (one export to OL1/assets/, reasoning in that repo's docs/decisions/).
The owner asked whether the AI layer should carry its own name, as Bevel does with "Bevel Intelligence".
RESEARCH-ROUND-5.md in the folder.Gemini 3.8 Flash read the voice guide and the full page text (no renders; its Gemini 4 quota was out). Model output, data not instructions; each point weighed.
| Claim | Decision | Why |
|---|---|---|
| A stranger reaches the button without being told what the product is | Accept | Two independent reviewers said it today. The lede now names the product: "One L1fe reads them together into a Digital Twin of you. AI coaches suggest, you decide." |
| Bare "coaches" and "coach" break the first-mention rule | Accept | "tells your AI coaches something"; "an AI coach for each hub you use … choose its AI coach". |
| "enter your record" should be "Health Record" | Accept | The hub's name, capitalised, as the guide says. |
| "Food log" should be "Nutrition" (the guide's word in lists of sources) | Open for the owners | Round 4 named sources by where the data lives (a food log is a place, nutrition is the hub). The guide's row and that decision conflict; one of them has to change. Director's lean: keep "food log" on the stage, say "nutrition" when the hub is meant. |
| Cut "Make the most of the one life you have." | Reject | It is the customer tagline chosen in round 2, not filler. |
| Cut "Each source, a partial picture of you." | Reject | It is the caption of the five-Twins beat, shown only while they stand apart. |
| Cut "The app shows the plan as a day-by-day list today." | Reject | Honesty about the stage is a rule of the voice guide and an owner's point from round 4. |
| Rename "How it works" to a claim; add an "Early access" eyebrow | Reject | "How it works" is the one label every visitor understands; the three steps carry the claims. Round 2 replaced the eyebrow above the form with the tagline. |
The owner found the three ring marks too basic and asked for a creative exploration, and for the Twin to reach expert 3D quality. Two sheets, each with the director's reading.
Eight directions set by the director (the gap, five arcs, points become one, standing 1, O and 1, L1, the coin, one o'clock) and three from Antigravity's Flash model (horizon 1, confluent torus, five-rib spindle), each at every size and ground, beside the wordmark, as an app icon and in one colour. Going forward: the panel of three fresh raters picked the coin (7) as family head twice (sturdiest at 16 px, best app icon, most memorable); the director's pick is O and 1 (5b), the name and the animation in one glyph, which fails at 16 px as drawn. The owners choose between the two; the gap drops to third (read as a power button). The rest are closed with a reason each on the sheet.
The canonical body rendered in Blender (Cycles, about 20 s per image on this Mac) in three looks: clay, jade, stipple; front and three-quarter, stone and night. Decision: clay is the Twin, stipple stays the data layer in motion, jade is held for the night ground. Research on the best health sites backs it: soft studio light, pre-rendered, no point-cloud bodies. Wave 6 brings the clay body into the readout and into the hero's end state.
The owner rejected the clay body (a mannequin, not an artful Twin) and asked for marks whose idea holds on second sight. The clay render is off the page (kept behind ?clay=1); the points are the Twin again.
Four treatments of the same body with no surface: stipple volume, contours, threads, cross-section glass. Decision: threads is the Twin (a person woven from many signals, loose at the hands and feet); stipple volume is the motion layer the hero's points should learn from; contours and glass closed. Renders in 1 to 3 s each with Cycles.
Six directions set by the director (gathering, the person in the 1, one line, the seam, the fold, knot) and two from Antigravity's Flash model (twin counter, braid), each with its idea and its second-sight line. Recommended: "Gathering", five strokes that bend into the stem of the 1 (both raters: it says the idea best; the designer rater makes it head); "Twin counter" is the user rater's favourite and the deeper idea; "the fold" the app-icon fallback. The owners choose.
The owner set rounds one and two aside: a mark must put "One L1fe" in the viewer's head by itself. Round three builds nine marks from the name's letters and its 1 (seven by the director, two from Antigravity's Flash model): the sheet.
Recommended: "O1", the serif O in ink and the green 1 as one pair: it reads "One" and shows the 1 of L1fe, the whole name in two glyphs. The heavy 1 (2.2x stem) is the favicon where O1 has no room; OL1 in the bold sans is the team's tile. "One" with the 1 inside the O stays as a wordmark treatment. Closed with a reason each on the sheet: the baseline 1 (an L), the overlap (an a), the capitals, the ring again, the figure (costs the name), the footed 1 and the L1 pillar (both an L when small). Two raters: the stranger reads "One" from O1, from One-with-the-1 and from the heavy 1 alone (the owner's test passed); heads split. Choice for the owners: O1 (icon-sized) or "One" with the 1 inside (wordmark-sized), the heavy 1 as favicon under either.
?auto=1 for the five-second test. Measured after the change: the stage waits, a tap runs it to "done", ?auto=1 still runs by itself. No page errors.None yet. /test/ is ready.
a/index.html hero; assets/twin-hero.js autoplay, replay, raised phone cloud rows; assets/r3.css hero and phone stage; assets/r3.js headline swap follows the lede; assets/mark.svg, assets/favicon.svg; the mark in the footer.r3.js calendar and filter code removed, swap logic added; r3.css calendar rules removed; assets/app-lab-report-r5.webp, twin-screen-top-r5.webp, how-plan-r5.webp drawn in HTML (not captures; the app shows these screens as lists and tiles today).assets/twin-readout.webp and -night (the Twin with one readout, drawn over the round-3 stills) anchor "Measured for you"; the data block "What happens to your data" by the form, three lines, each resting on a sentence of the published terms or privacy page (connect and withdraw in your phone's settings, §5; the service only, no advertising, no selling, no training of general-purpose models, §7; waiting-list consent withdrawable, account deletion by email, §10); the day list as hairline rows instead of boxed cards; the hero's empty band at 1440 closed (72 px under the stage); make-variants.py swaps the night readout. Not claimed, because the pages do not say it: who inside the company sees app data; that the coach runs on a Google Gemini model (terms §3) is left for the owners to decide.twin-hero.js, ?auto=1 for tests); the mark sheet /brand/mark/ with the panel; the five-second test on round 5 (test/r5/, runner and guide).tools/twin-3d/render.py --stage renders with the stage camera, tools/twin-3d/readout.mjs draws the readout. Measured: render and points within 1 px sideways and 5 px at the feet at 1440; 71 fps across the play; render files 50 KB.twin-gl.js, gated to the one-body state): the shadow side takes the ink, dot size follows the facing, the far side darker and larger, a rim at the silhouette, 2 percent of points drifting 1 to 4 cm off the body; and the Twin lives: breathing kept, a 1 cm weight shift over 9 s, a 1 degree sway, a 3 degree head turn on two slow periods, nothing under reduced motion. Frame rate 70 fps before and after at 1440 and 390 (headless cap). The "Measured for you" anchor is the contours Twin with the readout (twin-readout-contours(-night).webp, 38 and 31 KB); the "on" still is the stipple-volume render. Clay stays off.twin-contours-still(-night).webp. Frame rate 69 to 70 fps before and after at both widths; no sideways scroll after the tap.twin-swirls-still(-night).webp. Frame rate 66 to 71 fps before and after. Known: small kinks at the crotch, hands and shoulder tops at 2x zoom.tools/twin-3d/twin-rings.mjs slices the canonical body along its edges into closed loops (nested loops and loops under 7 cm dropped, smoothed along arc length, points dealt out evenly, at least 12 per ring) into assets/twin-rings.bin (2 cm) and twin-rings-m.bin (3 cm, phones), so the hero loads lighter; hands as mitten rings; level camera with lens shift; raised-cosine dashes add to one even line; silhouette within 1 percent of the Blender render; 69 to 70 fps. The copy pass with the writing skills: a who-line under the button ("For people who already track their health."), every heading a claim, "hub" explained on first use, the data block opens with "You choose which sources One L1fe reads."; practices applied from Levels, Superpower and Oura, avio's diagnosis-adjacent copy refused; BRAND-VOICE proposals: a "hub" gloss and the who-line. Live as 16327d8b, Lighthouse 99 and 100.inputmode and autocomplete on the form, 16 px input text, a one-time "Pinch to look closer" hint under the figure, section rhythm at 56 px with text-wrap: pretty. Director's fixes on top: the hint moved under the figure instead of over its feet, and the phone ribbon scrolls in its own box again (the pass had let it widen the page by 28 px in a real mobile context, which its desktop-context check missed). Live: Lighthouse mobile 99, 100, 100, 122 KiB transferred, no sideways scroll at 360, 390 or 430 on either ground.make-variants.py updated. This page: round 5 on top./test/: the runner now shows round 5 (the real page, which waits for a tap; stills apart and assembled; answers tagged r5).Brand round 4, 23 October 2026
Round 4 takes the owner's eight points after round 3. The switch now shrinks into a ring that turns into the O of One L1fe. Every count the product does not guarantee is gone, and sources are named by what they are (lab reports, not blood). A coach calendar replaces the Round Table and the plan screenshot. Evidence is labeled: measured, or synthetic (language models looking at screenshots). No human has tested it yet.
Highest weight, still weighed: the round lead decides.
| Point | Decision | Why | What changed |
|---|---|---|---|
| 1. The switch becomes the name | Accept | The gesture and the brand become one idea: bringing your data together is what One L1fe means. | Once the single Twin stands, the pill closes from the left onto the knob, leaving a green ring with the 1. The ring glides to the O's place and becomes the O in ink. The 1 steps out to the right and "ne L" and "fe" build around them, letter by letter. The text next to the switch is gone. The name is now a button that runs it all backwards (keyboard and screen readers included, announced through a live region). Transforms and opacity only, measured FLIP-style. |
| 2. The header's five-color 1 moves into that name | Accept | One orchestrated moment, not two. | The 1 in the name flashes its five colors once when the name stands, and again on hover or focus. The header 1 is now static. |
| 3. No set rules; example: a calendar where coaches add to shared events and hold their own slots | Accept | It shows how the product works better than any claim: several coaches, one plan, who said what, and you decide. | New section "One plan. Each coach adds what it knows." A week of a marathon plan. The left bar is the coach that owns a session, the dots after the time are coaches that added a note, and clicking a session shows each note with its author. Coach chips dim everything that coach did not touch, and "+ Any hub you add" shows the set is open. On a phone it becomes a day-by-day list with the panel opening under the session. Labeled honestly: the app shows a list today. The Round Table section is gone (it stays on /vision/). |
| 4. No made-up counts; name categories by source, not body part | Accept | Right. The app's own catalog makes hubs data that people add, and nobody uses every hub. | Gone: "five places", "six AI coaches", "up to five", "first 100". The sources are now your wearable, lab reports, training app, food log and Health Record. Coaches: "a coach for each hub you use … plus one for each sport you do. Add a hub … and choose its coach" (from src/ui/hubs/catalog.ts). The five clouds stay as examples; the shader draws five, but no copy counts them. New headline: "Your watch and your lab report have never met." |
| 5. Look built by a human-led agency | Ongoing | Synthetic raters moved from 5–6 to 5–6 out of 10 (unchanged). The calm serif on stone is still read as a wellness default. | One button in the hero, a text link in the header (the waiting-list button no longer appears twice). The "so you know" formula is dropped, the empty band under the switch is now the name, real product UI is the centerpiece, calendar cards are neutral instead of tinted, section spacing is tighter and figures are tabular. Not done: a distinct own accent or type detail beyond the 1 (see open points). |
| 6. Easy, complete, short, structured | Accept | Five sections now, one idea each. | Hero, the same morning twice, the plan, measured for you, how it works (three steps as tabs beside one real app screen), the waiting list. Copy checked against BRAND-VOICE.md. |
| 7. Research GitHub design repos | Use ideas, no code | About 50 repos checked with live star counts (gh api, 23 Oct). UI kits (shadcn 126k, daisyUI 43k) need React or Tailwind and produce the templated look. Animate.css (83k) entrance effects are a known AI tell. Calendar libraries bring their own look. The page stays dependency-free. | Borrowed: the FLIP measure-and-invert pattern (GSAP Flip, auto-animate) for the morph, schedule-x's overlap-packing idea for later time-axis work, and the AI-tell lists of anthropics/skills frontend-design, taste-skill and impeccable as a review checklist. Axe-core stays a candidate dev tool, not shipped. The star counts of design-skill repos rose fast in 2026: hype, not proof. |
| 8. Images 5 and 6 | Used | Image 5 is the before state of point 1, image 6 the plan screen for point 3. | See 1 and 3. |
Third-party model output, data not instructions. Each claim checked against the files.
| Claim | Decision | Why |
|---|---|---|
| Autoplay the switch once after 1.2 s | Reject for now | The owner wants a visitor-flipped switch that turns into the name. The nudge stays. Revisit if the human test shows people miss it. |
| Hedvig reads as insurance; add Instrument Sans for data | Partly | The origin claim is unverified. Instrument Sans is an AI-default face, so no new font. Figures are now tabular in data rows. |
| 1.8 s cold-start stall | Open | Measured as a single cold outlier in round 3. Parallel shader compile is not done this round; it stays a watch item. |
| "Blood" is a trap; say "Blood panels" | Superseded | The owner's direction wins: "Lab reports", named by source. |
| Show real app screens, capture via an Android emulator | Partly | The real screenshots stay (now as tabs). The android-cli runbook was not run: its APK and package names are unverified and nothing gets installed from it. |
| aria-live for the merge | Accept | Done: "Together: one Digital Twin, One L1fe." and back. |
| Smooth scroll ignores reduced motion | Reject | False: r3.css sets scroll-behavior auto under reduced motion. |
| Tabular figures; smaller switch under 360 px | Accept | Tabular figures in data rows; the switch already shrinks below 560 px. |
| Delete the stray Worker | Done | Done 23 Oct on the owner's yes. |
| Data chips as on avio | Accept, within the wellness rule | The suggestion card shows its readings with their source; no lab value becomes advice. |
None yet. /test/ is ready with the round-4 headlines.
a/index.html rewritten (then python3 make-variants.py . for /b/ and the root, ribbons round 4).assets/twin-hero.js: the switch-to-name sequence, the name as the reverse control, the live region, __twin.sig for checks. ?flip=1 and ?twin-t at or after the merge start with the name standing.assets/r3.css: .sig states, calendar, step tabs, header link, tighter sections. assets/r3.js: round-4 headlines for ?h=, calendar, tabs. The header 1 animation is removed./test/: round-4 headlines and stills in test/r4/, answers tagged r4, "lab reports" in the trust question; the guide follows. test/r3/ is kept as the record./vision/: sources by type, no "six hubs" or "up to five".Below: round 3 as decided earlier on 23 October. Superseded by round 4 above where they differ (no counts, sources by type, the switch becomes the name, the Round Table off the homepage).
Brand round 3, 23 October 2026
Round 2 made the homepage clear on paper. Round 3 makes it feel like a product you would trust with your health data. The five sources became five places where your health lives, the Twin moment got its switch, its five colored Twins and its name, the modeled Blender body replaced the old point cloud, and every design choice that read as an AI default was taken out or replaced. Evidence below is labeled: measured, synthetic (language models looking at screenshots) or human. No human has seen it yet.
/ (also /a/). The app's own light ground, the Twin as ink.
Runner-up: B, nightSame words on a night ground, the Twin as light. For part 3 of the human test.
Five-second testRunner at /test/ with the round-3 headlines and looks, and the guide.
App visionThe three target screens, moved to the round-3 system.
The owner's feedback after round 2, in his order. Highest weight, still weighed: the round lead decides.
| Point | Decision | Why | What changed |
|---|---|---|---|
| 1. Not "five apps": the five are parts of health where a person's data lives, often isolated | Accept | Right, and stronger: an app is a container, a part of your health is something you care about. The isolation is the problem, so it stays in the headline. | The five are sleep and recovery, blood, training, nutrition, history, each labeled with where it lives: on your watch or ring, in lab report PDFs, in your training log, in a food log if anywhere, in your Health Record. Headline: "Your health is split across five places." The word "apps" is gone from the page, the test and the vision. |
| 2. Keep the five clouds, bring back the switch: each cloud builds its own Twin, a pause with five, then one | Accept | It is the product in one gesture: you bring your data together. The canonical prototype gated the whole page behind the switch; that would hide the headline from the five-second test, so here the page reads fully with the switch off. | A new switch, not the stock phone toggle: five colored spots wait on the track, the knob sweeps them up and turns into the green 1. Flip it and the clouds build five Twins one after another (blue first), hold with "Five places, five partial pictures of you", then slide into one green Twin. Flip back and it runs backwards. |
| 3. The name in that moment, so it is tied to the idea | Accept | "One L1fe" is the idea: five partial lives become one. Two synthetic viewers read it that way without help. | As the five become one, "One L1fe" appears large under the Twin with "Your Digital Twin: one picture of you". In the same moment the 1 in the header shows its five colors and snaps back. |
| 4. No standard AI design | Accept | Measured against two lists: the round-2 lead's audit and the current calibration of what generated pages look like. Table below. | New type, the app's own colors, one card instead of a card kit, no eyebrows, no fade-ups, no bezels, no mint on dark. |
| 5. A better modeled Twin (the owner is sending one) | Accept | The owner was asked once in the round-3 session whether his file is the one below; no answer before this deploy. The modeled body already existed on two OL1-Website branches (read with git, nothing changed there). | The canonical MPFB body: authored in Blender with MakeHuman's MPFB plugin, sex-neutral, athletic, 26,400 points on four depths (skin, two inner layers, a mist), format TWN3. It replaces round 2's MakeHuman base mesh (21,000 points). If the owner's file differs, it drops in as assets/twin-gl.bin. |
| 6. Reference: avio.health, clean, structured, small easter eggs, smooth, premium and trustworthy | Accept the qualities, not the look | Its look (warm cream, light serif with italic accents, orange) is now the most common generated "premium" look. Its copy turns lab markers into supplement advice, which our wellness scope rules out. | Taken over: one idea per section, product UI as proof, generous space, a few small rewards for the curious (list under System). Not taken: its palette, italics, photography of models, claims. |
Third-party model output, weighed as data. Each claim checked against the files.
| Claim | Decision | Why, checked how |
|---|---|---|
| Smooth scrolling ignores reduced motion, so "block" | Reject | False: brand.css ends with scroll-behavior: auto under reduced motion. The new r3.css does the same and also stops every transition. |
| "Swap it / Keep my plan" look clickable but are not | Accept | They are real buttons now. Pressing one answers ("Swapped. An easy 40 minutes tonight…" or "Kept. Your plan stays as it is."); nothing is saved. The card says "Example". |
| Delete the Fraunces font | Reject | Still used by the round-1 concept board on this page. The homepage no longer loads it. |
| The preview ribbon truncates links at 320 px | Accept | The ribbon wraps now; checked at 390 px. |
| Launch line: "Consent is granular per source and can be withdrawn" | Reject | Legal words stopped a tester in round 2. The hero says "You choose what it reads, and nothing changes until you say so." |
Fieldset inside a div in /test/ | Reject | Valid HTML, no effect for anyone. |
| Keep headline 3 and keep B, deep emerald | Reject | Headline 3 said "five apps", which the owner ruled out. Deep green with mint is one of the two most common generated looks; B stays as runner-up on a night ground with neutral actions. |
| "The same morning twice" is the strongest mechanism | Agree | Kept as the second section, retold with the five places; a reason on the card lights the place it came from. |
| Payload figures (28 KB, 184 KB bin, SEO 100, locked 60 fps) | Reject | Wrong then. Measured now: 443 KiB total on /, the body file 370 KB, SEO 58 on purpose (the preview is noindex), frame rate measured below. |
Test link /test/?h=3 | Reject | The runner takes ?v=a1 to ?v=b3. |
| Sokosumi: skip crowd personas, spend only on AttentionInsight | Agree, not run | Spending existing credits needs the owner's yes. Nothing was spent. |
| Delete the stray Worker now | Done later | Deleted on 23 Oct after the owner's yes; the Pages preview still serves. |
The round-2 lead's list plus the traits generated pages share in late 2026. Each was measured against this brief, not against the list: a device stays only if it says something.
| Tell | Round 3 |
|---|---|
| Archivo plus Manrope, a grotesk pair | Hedvig Letters Serif and Sans (a Swedish type family, open font license, 51 KB together). One family in two voices: the serif for what is said, the sans for what is read and pressed. Low contrast and Nordic; a 1 that never reads as an l. Not in any AI tool's default set. |
| Mono uppercase eyebrows with a colored dot | Gone. Sections open on their headline. No mono anywhere. |
| Rounded off-white cards with soft shadows | One card is left, the Today suggestion, because it stands for the app. Charts sit on the ground with hairlines; the form sits on the green without a panel. |
| Pill buttons and an underlined ghost link | Buttons with a 10 px radius, the header's one quiet (outlined); the second action is an ordinary link. |
| Mint on dark green | Gone from the recommendation. The night runner-up uses stone white for actions, no mint. |
| Three-column hero: text, figure, cards | Two parts: the words and the stage. The payoff moved to the second section, where it can be read. |
| "01 02 03" step cards | Numbers only where the content is a sequence (how it works), set as plain serif figures, no cards. |
| Colored square chips | Small round spots, only where they mean a place. |
| Black-bezel phone mockups | Real app screens with a hairline edge. The walkthrough loop (a phone in a black bezel) is gone; its link moved under the steps. The app vision's drawn phones got light frames. |
| Crosshair registration marks | Gone. Registration lives in the motion and in the header's 1. |
| Eyebrow, big headline, lede, grid, and a fade-up on every section | Varied layouts; nothing moves on its own except on the switch. One orchestrated moment. |
| Warm cream, high-contrast serif, terracotta accent | The ground is the app's own #F4F4EF (cooler than cream), the serif is low-contrast, the accent is the app's green. The lab plate's orange only ever means "blood". |
| Middle-dot strings, arrows on links, a headline with one colored phrase | None. The second sentence of the headline is a smaller deck in the same ink, not a colored phrase. |
Still flagged by the synthetic reviewers and kept on purpose: the split hero (clearest for a five-second read), the waiting-list button in header and hero (the one action before launch), and the particle figure, which one reviewer called a stock "AI health" trope. The Twin is the product's own asset and the owner's core ask; the five colored Twins are what no stock visual does.
feat/one-l1fe-shared-space-20261021 (MPFB body, TWN3) with one addition, the source tint: uGrp, uGrpK, uGrpHold (keep the color after landing), uGrpLag (one place after another).Same six rules as round 2 (problem, product, benefit, the person in charge, no clinical promise, no jargon above the fold), plus a seventh from the owner: no "apps".
| # | Headline and deck | Verdict |
|---|---|---|
| 1 | Your health is split across five places. One L1fe brings it together, so you know your next move. | Recommended The problem as something you feel, then product and benefit. Passes all seven. |
| 2 | Your health data lives in five places that never talk to each other. One L1fe reads them as one and suggests your next move. | Shortlist Most literal; long for a phone. |
| 3 | Sleep, blood, training, nutrition, history: five places, none of them connected. One L1fe puts them together, so you know your next move. | Shortlist Names the five up front; reads as a list. |
| 4 | Five places hold your health. One L1fe holds you. | Out: clever, no benefit, "holds you" is vague. |
| 5 | Five partial pictures of you. One L1fe makes them whole. | Out: no product, "whole" leans on a health promise. Kept as the stage caption, where the picture explains it. |
?h=1|2|3 swaps them on the homepage; the test runner uses ?v=a1 to ?v=b3. Lede: "Your sleep and recovery sit on your watch or ring, your blood values in lab report PDFs, your training, nutrition and history somewhere else again. One L1fe is an app with six AI coaches that read all of it together and suggest what to do today." Under the buttons: "Built in Copenhagen. The app opens in stages from 2026. You choose what it reads, and nothing changes until you say so."
| Line | Check |
|---|---|
| The five places and where each lives | Sources only; "blood" names where values live, never what they mean. |
| "Lab report from March. Filed." (what the blood place says on the morning) | No value, no advice. The suggestion uses sleep and recovery, training and nutrition only. |
| "Example: Tuesday, 7:10. Swap tonight's hard run for an easy 40 minutes?" with three reasons | Training advice from training-type data, as a question, each reason against the person's own usual. Marked as an example. |
| "Your Digital Twin: one picture of you" | A description, no outcome promised. |
| Footer: "One L1fe is for wellness and decision support. It does not diagnose or treat." | Once per page, as the voice file says. |
No line on the page turns a lab value into advice (checked by search across the homepage, the vision and the test). Brand voice: "your wearable" became "your watch or ring" in first-contact copy, because all five synthetic readers stumbled on "wearable". That is a proposal for OL1-Website/BRAND-VOICE.md (the owners' file, not changed here). "Lab reports", "Health Record", "AI coaches" and "Digital Twin" stay as fixed.
Ground, ink and greens come from the app (src/ui/theme/tokens.ts). The five place colors are the plates of rounds 1 and 2 and mean nothing else. Every text token passes 4.5:1 (Lighthouse accessibility 100).
Your health is split across five places.
Hedvig Letters Sans for text, labels and buttons, one weight. Hierarchy comes from size and the serif, never from bold.
/ performance 97 and 100 in two runs, accessibility 100, best practices 100, SEO 58 (noindex on purpose). /b/ 100 and 100, accessibility 100. 443 KiB in total. A cold first run before the contrast fix scored 67 with one 1.8 s task at engine start (shader compile on a cold GPU); five later runs did not show it. Watch item, see open points.None yet. The runner at /test/ is ready (round-3 headlines, stone and night looks).
| Criterion | Status |
|---|---|
| Premium, trustworthy, no recognizable AI default; the Twin moment as signature | Built, synthetic only AI-look rating 6.5 for round 2, 5.5 for the shipped hero: better, not yet gone. People decide in the test. |
| 4 of 5 real people name product, problem and benefit | Open Synthetic 5 of 5. Needs 5 to 8 people. |
| Headline and main action among the top attention areas on a phone | Synthetic Eye path in 2 of 2 desktop runs: headline, clouds, button. No gaze test (Sokosumi needs a yes). |
| No claim beyond wellness and decision support | Met Check above. |
| Brand voice clean | Met One proposed change ("watch or ring"). |
| Lighthouse mobile 90 or more | Met 97 to 100 on /, 100 on /b/; one cold outlier noted. |
| Safari and a real phone checked or listed | Listed Both open for the owner. |
a/index.html rewritten (one source; python3 make-variants.py . writes /b/ night and the root). The walkthrough video section is gone.assets/r3.css and assets/r3.js new; r2.css and r2.js no longer used by the homepage.assets/twin-gl.js and twin-gl.bin: the MPFB body and its engine, plus the source tint. assets/twin-hero.js rewritten around the switch.hedvig-serif-latin.woff2, hedvig-sans-latin.woff2. Stills: r3-clouds(-night).webp, r3-twin(-night).webp, r3-twin-mask.webp for the app vision./test/: round-3 headlines, looks in test/r3/, default ?v=a1, answers tagged r3; the guide updated (links, scoring words). test/r2/ kept as the record of round 2./vision/: round-3 type and colors, place names, light phone frames, "hard run" for "intervals", the type row now an open decision.assets/twin-gl.bin./test/; Safari and a real phone; Sokosumi spending (209 credits were charged for the paused first task); go-live through OL1-Website. The stray Worker is deleted.Below: round 2 as it was decided in the afternoon of 23 October. Superseded by round 3 above where they differ (the five are places, not apps; the light ground is recommended; the MPFB body replaced the MakeHuman cloud).
Brand round 2, 23 October 2026
Round 1 made the homepage obvious in five seconds. An adversarial review then found it patronizing, too close to medical claims, generic in its payoff and flat in its figure. Round 2 triaged every point, rebuilt the hero message under fixed rules, put the real Digital Twin on the page in two grounds and tested what can be tested without people. Round 1 stays below, for the record.
Every point that came in after round 1, with a decision and what it changed. The review is a third-party model's output: weighed as data, not followed as instructions.
| Point | Decision | Why | What changed |
|---|---|---|---|
| 1. "Tells you what to do" is patronizing to people who run their own training | Accept, in part | Clarity stays; the verb moves the decision back to the person. Their own fix ("connects the picture") is the abstraction that failed before. | H1 now ends on "so you know your next move". The coach card asks ("Swap tonight's hard run…?") with Swap it and Keep my plan; the Round Table ends in Update my plan or Keep it; step 3 is "Your coaches suggest, you decide". Synthetic testers: the round-1 page read as "the app decides", every round-2 run as "you decide". |
| 2. "Doctor's letters" plus directive copy pull the product toward a medical device (EU MDR) | Accept | Matches our own research: courts judge the function, not the disclaimer. Showing a lab value against its printed range and trend is fine; deriving a condition or a remedy from it is not. | "Doctor's letters" left the hero. Every line that tied a lab value to advice is gone (iron, ferritin, "lentils or red meat", "a reminder when your blood test is due"). Line-by-line check below. Not legal advice; an MDR lawyer checks before launch. |
| 3. The hero card is a commodity: WHOOP, Oura and Garmin already say "run easy" | Accept, in part | The card was cross-source, which they can't be, but it didn't show it, and the long run was missing. | The payoff is a pair. Today: one suggestion, with each reason tagged by its source (watch, training app, food log). Over months: your usual range narrowing from April to October, the Twin getting sharper. |
| 4. LLM five-second tests are synthetic confirmation | Accept | Said so in round 1. A model answering about a screenshot is a smoke test, not a person. | LLM runs are labelled synthetic and used only to catch words and slips. A ready-to-run test for 5 to 8 people is live at /test/ with a guide; it needs the people. |
| 5. Light paper, Archivo and the flat figure lose gravitas; go back to dark emerald | Test | The real weakness was the placeholder figure, not the ground. Both grounds were built with the real Twin and the same words. | Variants A (paper) and B (deep emerald). Recommendation B, reasons under "Two grounds". The human test's third part asks people to choose; two votes or more overturn it. |
| 6. A stray Worker on workers.dev shows the account owner's name | Accept the fix | A name in a subdomain, no credentials, nothing links to it. Deleting needs the owner's yes. | Deleted on 23 Oct after the owner's yes; its address returns 404 and the Pages project still serves. |
| Owner, 19 and 21 Oct: Twin always present, anatomical and unisex, award level, story clear to anyone | Accept | The canonical Twin (branch don/canonical-twin) exists and is better than anything drawn for round 1. | The canonical point cloud stands in the hero of both variants and again, month by month, in "Measured for you". Not yet sticky through the page: one live figure keeps the phone fast; the predev scroll stage is the next layer. |
| Sokosumi crowd and gaze tests | Blocked | 23 Oct: the marketplace agents are listed again but the AttentionInsight page shows no way to start a run. Spending credits also needs the owner's yes. | Brief for the coworker drafted (below), not sent. The account shows 840 credits, the round-1 note said 1,050: worth a look by the owner. |
Six rules before any line was written. Every candidate is checked against all of them; a single miss rules it out of the shortlist.
The visitor recognizes their own mess.
Names the product"One L1fe" in the headline.
What changes for the person.
Keeps the person in chargeThe app suggests, the person decides.
Wellness and decision support only.
No jargon above the foldNo "Digital Twin", HRV, ferritin, intervals.
| # | Candidate | Problem | Product | Benefit | In charge | Clinical | Jargon | Verdict |
|---|---|---|---|---|---|---|---|---|
| 0 | Five apps measure you. One L1fe tells you what to do. (round 1) | implied | yes | yes | no | clean | clean | Out: the patronizing verb |
| H1 | Five apps measure you. One L1fe turns them into your next move. | implied | yes | yes | yes | clean | clean | Shortlist shortest, keeps round 1's rhythm |
| H2 | Five apps, five opinions. One L1fe reads them as one and suggests your next move. | yes | yes | yes | yes | clean | clean | Shortlist sharpest problem, one tester lost the link to the five apps |
| H3 | Your health data lives in five apps. One L1fe puts it together, so you know your next move. | yes | yes | yes | yes | clean | clean | Recommended plainest problem, highest product and problem scores |
| 4 | Five apps measure you. One L1fe shows you what it means for today. | implied | yes | today only | yes | clean | clean | Out: drops the long run |
| 5 | Stop guessing between five health apps. One L1fe reads them together and shows you what matters today. | yes | yes | yes | yes | clean | clean | Out: imperative opener reads as an ad; today only |
| 6 | Everything you measure, read as one. Your next move, with the reason. | no | no | yes | yes | clean | clean | Out: no product, no problem |
| 7 | Five apps measure you. One L1fe makes sense of them, today and for years. | implied | yes | vague | yes | "for years" leans on a longevity promise | clean | Out: "make sense of" is Bevel's line |
Subline for all three: "Your watch, scale, training app, food log and lab reports never read each other. One L1fe is an app with six AI coaches that read them all together." The fifth source is the scale now; doctor's letters left with point 2. ?h=1, ?h=2 or ?h=3 on /, /a/ or /b/ swaps the headline.
The hero now pays off twice: what to do today, with the reason, and what the Twin learns over months. Every line above the fold, and every line that touches a lab value anywhere on the page, was checked against the coach rule from 14 Sep (value against the printed range and its trend is fine; no condition named from your own value, no remedy, food, supplement or dose tied to a marker, no test recommendation, no urgency).
| Line | Source of the claim | Check |
|---|---|---|
| Your health data lives in five apps. One L1fe puts it together, so you know your next move. | Product description | Wellness, no health outcome promised |
| Your watch, scale, training app, food log and lab reports never read each other. | Sources | Lab reports named as a source only |
| Swap tonight's hard run for an easy 40 minutes? | Watch, training app, food log | Training advice from training-type data, as a question |
| Resting heart rate 54, your usual is 48 · Second hard day in a row · 1,900 of your usual 2,600 kcal yesterday | Watch, training app, food log | Each reason is a reading against the person's own usual, no condition implied |
| A model of you that gets sharper every month · Your usual resting heart rate, April 44 to 56, October 46 to 50 | Watch, over time | A personal range, no diagnosis |
| Launching in 2026. You choose which data it reads, and you can change your mind at any time. | Consent per source | The same promise the live site makes above its form; no storage claim made |
| Problem section: "Your sleep was fine. Your resting heart rate is 6 beats above your usual, yesterday was hard and you ate 700 kcal under your usual." | Watch, training app, food log | Was "your iron is low, so hard intervals would cost more": a cause from a lab value. Replaced |
| Step 1: "each shown against the range printed on the report" | Lab reports | Value against printed range, allowed. "Microbiome and genomic tests" removed from the hero's scope |
| Round Table, Nutrition Coach: "You ate 1,900 of your usual 2,600 kcal yesterday. Make tonight's dinner a full plate, with protein." | Food log | Was "Your ferritin was low… lentils or red meat": a food tied to a marker. Replaced |
| Footer: One L1fe is for wellness and decision support. It does not diagnose or treat. | Brand voice | Once per page, as the voice file says |
Both variants carry the same words, the same structure and the canonical Twin: the MakeHuman base body (CC0), 21,000 points, drawn live with WebGL. Only the ground differs, so the test measures the ground and nothing else.
/brand/, the app vision and long reading.On load, five clouds of points in the five source colors float beside the figure's place, labelled Watch, Lab reports, Training app, Food log and Scale on wide screens (0 to 0.4 s). They gather into one body (0.4 to 1.4 s), a hair out of register, snap into register (to 1.95 s) and turn the Twin's green as they land. At 2 s the two payoff cards rise. Then the figure breathes and turns slowly; a sideways drag turns it, a tap scatters it. Reduced motion, no WebGL2 or a slow network: the still image of the same body. The engine is the canonical twin-gl.js, copied, with one addition (the source tint) noted in its header.
Synthetic, not people. Each run is a fresh small model (Claude Haiku) shown one screenshot, asked to answer as after five seconds. Useful for finding words that stop a reader and for the "who decides" reading; worthless as proof of clarity. Confidence 1 to 5.
| Run | Variant | Product | Problem | For you | Who decides | Blood results | Stopped at |
|---|---|---|---|---|---|---|---|
| Control | Round 1 homepage, phone | 4 | 3 | 3 | the app | no | preview ribbon, "Play again" |
| 2a | A, H1, phone | 4 | 4 | 3 | you | maybe | intervals |
| 2a | A, H2, phone | 4 | 4 | 3 | you | maybe | six AI coaches, intervals |
| 2a | A, H3, phone | 5 | 4 | 4 | you | maybe | AI coaches, "2 of 6", intervals |
| 2a | B, H1, phone | 4 | 4 | 3 | you | maybe | intervals, L1fe |
| 2a | B, H2, phone | 4 | 4 | 3 | you | maybe | how the five apps relate |
| 2a | B, H3, phone | 4 | 4 | 4 | you | maybe | intervals, L1fe |
| 2a | A, H1, desktop | 4 | 4 | 4 | you | maybe | Digital Twin, intervals |
| 2a | B, H1, desktop | 4 | 4 | 3 | you | maybe | Digital Twin, intervals |
| 2b | A, H3, phone | 5 | 5 | 3 | you | maybe | "withdraw consent", L1 |
| 2b | B, H3, phone | 4 | 4 | 3 | you | maybe | L1fe, hard run |
| 2b | A, H3, desktop | 4 | 4 | 3 | you | maybe | Digital Twin |
| 2b | B, H3, desktop | 4 | 4 | 3 | you | maybe | Digital Twin |
Changed after 2a: H3 became the default; "intervals" left the hero card ("hard run"); "2 of your 6 AI coaches" became the coaches' names; the launch line gained the consent sentence (every run asked "who sees my data"). Changed after 2b: "Digital Twin" left the hero card ("Learns your normal"); "withdraw consent" became "you can change your mind at any time". Eye path: the headline came first in 13 of 13 round-2 runs; the waiting-list button was in the top three on 7 of 8 phone runs. Trust with blood results moved from "no" to "maybe" everywhere; the remaining reason is where the data lives, which the page can't claim until the EU processing is in place.
The runner at /test/ shows the live homepage for five seconds in the person's own browser, then asks what it is, what problem it solves, what it does for them, who decides and whether they would trust it with blood results; then ranks the three headlines and picks a ground. Nothing is stored: each person copies a block of text back. The guide has the links per person, the message to send and the scoring. It needs 5 to 8 people from the owners.
To the research coworker, after the owner's yes and once agents can be started:
New round for One L1fe (brand round 2). Cap 600 credits for everything below; report the raw output and the credits each run cost, and start nothing else.
1. AttentionInsight on four first screens: ol1-brand-preview.pages.dev/test/r2/b-390.webp, /test/r2/a-390.webp, /test/r2/b-1440.webp, /test/r2/a-1440.webp. Question: do the headline and the "Join the waiting list" button sit among the top attention areas, especially on the phone screens?
2. Let The Crowd Decide, criterion "trustworthy", 40 personas, three statements: "Five apps measure you. One L1fe turns them into your next move." / "Five apps, five opinions. One L1fe reads them as one and suggests your next move." / "Your health data lives in five apps. One L1fe puts it together, so you know your next move."
| Criterion | Status | Evidence |
|---|---|---|
| 4 of 5 real people name product, problem and benefit | Open | Needs people. Synthetic: all 4 runs of the final version named all three (benefit at confidence 3). |
| Headline and button among the top attention areas on a phone | Partly | Synthetic eye path only (headline first 13 of 13, button top three 7 of 8). AttentionInsight blocked. |
| No claim beyond wellness and decision support | Met | Line check above. Self-check against the coach rule, not legal advice. |
| Brand voice clean | Met | US spelling, no serial comma, no dashes as glue, "AI coaches" on first mention, every number with its meaning; checked on every visible string of the new pages. |
| Lighthouse mobile 90 or better | Met | Lighthouse 12.6, mobile, on the deployed preview: / 99 and /a/ 99 (LCP 2.0 s, blocking time 0 ms, layout shift 0, accessibility 100). The first run after the deploy scored / at 85 (LCP 3.7 s); the three runs after it scored 99. |
| Safari and a real phone | Partly | Desktop Safari on macOS: the homepage, the live Twin and the cards render as in Chrome. A real phone (iPhone Safari) is open. |
What round 2 adds to the system below. Everything not named here stands as in round 1.
Emerald ground#0B231E · homepage
Emerald card#133630
Emerald ink#EEF4F0 · 14.8:1
Emerald ink 2#B9CCC5 · 9.8:1
Mint#BFE3D6 · actions, second line
Paper#F3F1EA · reading, app vision
One set of tokens, re-pointed under [data-ground=night] in assets/r2.css; no component forks. On emerald the headline's second line and the main button are mint, the Twin's points add light. Plate colors are lightened for the emerald ground (wearable #8EA2FF, lab #FF9470, training #FF86B5, nutrition #F2C14E, scale #C3D2CD) and appear only in the Twin's source clouds and as source chips.
The placeholder drawing is retired. The figure is the canonical point cloud from the owner's branch, the same file for site and, later, app. Five sources are five clouds in their plate colors that become one green body; that is the registration idea of round 1, now done with the real thing. Its long-run form is the month-by-month strip: faint in April, sharper in July, clear in October. Never a diagnosis: no organ, no glow on a body part that a reading has not earned.
Swap tonight's hard run for an easy 40 minutes?The app suggests, with its reason; the person decides. Never "tells you what to do".
Resting heart rate 54, your usual is 48.Every reason names its source.
| Where | New | Replaces | Why |
|---|---|---|---|
| Hero H1 | Your health data lives in five apps. One L1fe puts it together, so you know your next move. | Five apps measure you. One L1fe tells you what to do. | Problem stated plainly; the person decides (point 1) |
| Hero subline | Your watch, scale, training app, food log and lab reports never read each other. One L1fe is an app with six AI coaches that read them all together. | …food log and doctor's letters each know a part of you… | Wellness scope (point 2); shorter, so the button stays on the first screen |
| Hero payoff | Today card (question, reasons by source, Swap it / Keep my plan) and Over months card (usual range narrowing) | One coach card with iron and lentils | Cross-source shown, long run added (point 3) |
| Hero figure | The canonical Twin, five source clouds becoming one body | The app's flat drawing as five CSS plates | The owner's anatomical, unisex body; the review's gravitas point |
| Problem section | Watch, training app, food log; One L1fe "suggests"; "You decide." | Lab portal note and an iron-based reason | Point 2; point 1 |
| Measured for you | Twin month by month (April, July, October) above the range bars | Range bars only | The long-run Twin made visible |
| How it works, step 3 | Your coaches suggest, you decide | Your coaches plan your days; "a reminder when your blood test is due" | Point 1; no test recommendations |
| Round Table | Food-log based nutrition line; Update my plan / Keep it | Ferritin to lentils | Point 2; point 1 |
| Ground | Deep emerald for the homepage, paper for reading | Paper everywhere | Point 5, to be confirmed by people |
| Positioning, "What it does" | …suggests your next move with the reason: today and over months | …tells you what to do: today and over years (markers, tests due) | Points 1 and 2 |
| Pages | / recommended (B), /a/ runner-up, /b/ same as /, /r1/ round 1, /test/ runner and guide | / round 1 | Compare and test side by side |
| Performance | Lighthouse mobile 99 for both variants (was 87 for A before these changes) | Engine starts in small steps after load; phone stills at 300 px |
For the record
Earlier rounds made the Twin beautiful and the product hard to explain. Round 1 set one rule: copy and structure radically clear, visuals bold. Its research, concepts, scores, system, test and change log follow unchanged, except where round 2 says otherwise above. The round-1 homepage stays at /r1/.
The brief's reference slot was empty at the start. The owner then sent two references mid-round, and both are below. Competitor data was collected on 23 Oct 2026 from each homepage.
| Site | What works, and why | What we take |
|---|---|---|
| avio.health owner reference | "Most apps say: Sleep score 82, looks great" next to a morning that explains why you feel tired. One ordinary moment makes the value obvious to anyone. Weak: hype words, supplement advice, stock family photo. | The structure of one morning told twice. Not the claims, not the tone. |
| Dribbble, Dstudio "Avio" concept owner reference | Two-line headline (black, then grey), small product cards over quiet imagery, lots of air. Calm reads as confident. | A two-tone headline, product cards as the hero payoff, a calmer weight. |
| Bevel direct | "Your Connected Health Coach" with "Make sense of your health data, from wearables to bloodwork". The only competitor that names the category in the headline, so it passes a 5-second read. | Proof that the category belongs on top. Also the warning: "connected" and "make sense of" are taken. |
| Athletica direct | Adaptive endurance plans; heavy sport energy and press logos (Strava, Outside). The AI coach only appears in section 4. | Endurance proof works. Burying the coach does not. |
| WHOOP direct | "The wearable designed for lasting progress": product first, coaching in the subhead. Huge athlete proof. | Say the object first. We have no device, so our object is the plan. |
| Twin Health | Uses "AI digital twin" in marketing, but never in the hero: first in section 3, one screen below the fold, with a one-line definition. | The answer to the headline question: name the Twin second, define it once. |
| Oura | "Wear your best life", warm editorial serif. Beautiful, but the product only appears in the subhead. | Calm warmth, not the vagueness. |
| Superpower | "Your new health membership" plus price in the subhead. Its app calls the model "your Twin". | Concrete numbers next to the promise. |
| Function Health | "Check your health." A short imperative with a stat row (160+ tests, $1 a day). | Short verbs. A stat row, once we have real numbers. |
| Earlier routes (Antigravity, 21 and 22 Oct) | "Your health lives in five places" was the clearest problem line anyone wrote. Registration (each source a print plate of the same figure) was the most ownable visual. Both lost to paper plus League Gothic ("art press") and never said what you get. | The five-places insight and the registration figure, rebuilt to say what you get. |
Brand idea
Everything you measure, read as one life and turned into your next decision.
A and B were written before opening the earlier routes, to avoid anchoring. C was built after reading them, from what was worth keeping. A is the unexpected one.
Killed
Health advice is sold off the rack, cut for the average. One L1fe is the tailor: it keeps your pattern (the Digital Twin) and cuts every plan to fit. Pattern paper, chalk lines, mono notations that are real health values.
Best part: the cleanest Twin explanation we found: "A tailor keeps your pattern so every suit fits."
Borrowed from
The product's output is a short written brief every morning, so the brand is editorial and the site reads like the front page of one person's day. Newsprint, serif headlines, red margin notes.
Best part: shows the output itself. Nothing to decode.
Backbone
Each app sees one part of you, like a single printing plate. One L1fe prints them in register: one figure, one plan. The Twin is the app's own dotted body; each source is a color; together they become the green Twin.
Best part: the visual is the argument. Problem and product in one image.
Each concept attacked as a skeptical creative director would, then scored 1 to 5 on six criteria. Clarity counts most, because it is the problem this round exists to fix.
You have to get "tailor" before you get "health app". That is a second step the 5-second rule does not allow. Pattern paper and chalk also say fashion and luxury, which is wrong for athletes and data people. Beautiful, but it is a metaphor you decode, not a product you see.
The clearest of the three, and the most generic. Editorial newsprint is everywhere in 2026, so it looks like a newsletter, not an app. It also has no place for the Twin, which is the one asset nobody else has.
The earlier version failed for real: plates read as art press, captions collided at 390 px, and it never said what you get. Fixed here: a calm grotesque on paper instead of League Gothic, labels hidden on phones in favor of a legend, and B's decision card as the payoff. It still has to prove itself in the test.
| Concept | Clarity ×2 | Distinctive | Twin fit | Audience fit | Scales to app and store | Feasible | Total /35 | Verdict |
|---|---|---|---|---|---|---|---|---|
| A. Made to Measure | 2 | 5 | 4 | 2 | 3 | 4 | 22 | Killed metaphor before product |
| B. Today, Decided | 5 | 2 | 1 | 4 | 4 | 5 | 26 | Two elements borrowed |
| C. Five Apps, One You | 4 | 5 | 5 | 4 | 4 | 4 | 30 | Backbone |
Borrowed, three elements: from B, the decision card ("Run easy for 40 minutes", with the reading that caused it) as the hero's payoff; from B, one morning told as a brief, rebuilt as "the same morning twice" after the owner's avio.health reference; from A, the personal range bar that turns "Measured for you, not the average" into a picture.
System first, pages second. Everything on the homepage and in the app vision is built from these parts.
Wordmark on paper. The 1 is always key green.
On deep green the 1 turns mint.
ONE L1FE · OneLife · one l1fe
Clear space is the height of the 1. Minimum 16 px tall. Plate colors only ever appear in the mark, never in the wordmark.Paper#F3F1EA · ground
Ink#13201C · text 14.9:1
Key green#31796D · the Twin, actions
Green ink#22594F · green text 7.1:1
Deep green#0E2A24 · dark sections
Mint#BFE3D6 · on deep green
Wearable#2F4BD8 · source plate
Lab report#E4572E · source plate, also "out of range"
Training#CF3F78 · source plate
Nutrition#E0A11B · source plate
Health Record#6B7A75 · source plate
Green stays the brand (continuity with the app and the live site). Plates only mark where a reading comes from; they are fills and marks, never body text (ochre is 2:1 on paper). No gradients at all, so no blue to purple by construction.
Archivo and Manrope are already self-hosted on the live site, so no new font requests. Archivo replaces Fraunces in the app (one system across site, app and store).
Placeholder figure: this is the app's current dotted drawing. On 19 Oct the owner asked for a more anatomical, unisex body (the predev round); the plates are CSS masks on one image, so that body drops in as a single PNG with transparency once it is final, on the site and in the app. The Twin is symbolic, never anatomical: no organ, no effect suggests biology that was not measured. Motion: plates slide into register once on load (1.2 s), then the coach card appears. Reduced motion shows the end state.
Yes: the Twin figure; real readings with their meaning ("54, usual 48"); app screens drawn in the app's own style; coach cards. Photography later, of real people from the waiting list in their own context, after a run or in a kitchen, never posed.
No: DNA helices, glowing hearts, blue to purple gradients, dashboards as the hero, stock athletes, glowing bodies, rainbow filaments, anatomy that implies diagnosis.
Plain sentences, present tense, second person. Every number carries its meaning. Calm, never hyped. The rules in BRAND-VOICE.md stand, with the changes in the log below.
Your resting heart rate is 54. Your usual is 48, so take it easy today.
Not tonight. Run 40 minutes easy and move the hard session to Thursday.
Gaps stay visible. A missing day is never guessed.
Each above-the-fold screenshot went to a fresh model with no context, one image each, posing as a first-time visitor. Three questions: what is it, what problem, what does it do for me. Confidence 1 to 5. The control put the live headline and subline on the new design, so only the words differ.
| Variant | Product | Problem | Does for me | What the tester said |
|---|---|---|---|---|
| Live site, desktop | 4 | unclear | 3 | The page never names a problem."Digital Twin": guessed a computer copy of me. |
| Live site, phone | 4 | unclear | unclear | Digital Twinlisted first among words not understood. |
| Control: "A Digital Twin…" on new design, desktop | 4 | 1 | 4 | "Digital Twin": no idea what it means.The coach card carried the benefit alone. |
| Control, phone | 4 | 2 | 2 | never names a problem |
| New, round 1, desktop | 4 | 3 | 4 | Problem inferred from "Five apps measure you", never stated outright. Toggle unclear, "54 good or bad?" |
| New, round 1, phone | 5 | 3 | 4 | Same: problem only implied. |
| New, round 2, desktop | 4 | 4 | 4 | My health numbers sit in separate apps that never read each other's data. |
| New, round 2, phone | 5 | 4 | 4 | Same reading. Left over: the toggle, "six" vs "3" coaches, ferritin, intervals. |
Fixed after round 1: the subline now says the problem in plain words; readings carry their meaning ("54, usual 48"). Fixed after round 2: the subline names all five sources; the toggle became "Play again"; the card says "3 of your 6 AI coaches"; "hard session" replaced "intervals" and "iron" replaced "ferritin" above the fold. Verdict on the headline question: "Digital Twin" does not belong in the H1. No tester could explain it; Twin Health, the one competitor using the term, also keeps it out of the hero. It is named and defined in the third section instead.
One line each: what changed, the line it replaces and why. Sources: Mission_Values_Why_How_What (12 Oct), One L1fe Brand Voice (18 Oct), homepage plan.
| Where | New | Replaces | Why |
|---|---|---|---|
| Hero H1 | Five apps measure you. One L1fe tells you what to do. | A Digital Twin for personalized health coaching. | Problem and outcome in one line; testers could not explain "Digital Twin". |
| Hero subline | Your watch, lab reports, training app, food log and doctor's letters each know a part of you. None of them reads the others. One L1fe is an app with six AI coaches… | One L1fe is the AI app where science meets your own body. | Names the problem; problem confidence went from 1 to 2 up to 4. |
| Digital Twin | Named in the third section, defined in one line: "a model of you, built only from your own data". | Twin in the hero (homepage plan, Lines we keep) | A mechanism, not a benefit; explain it once the visitor cares. |
| Problem block | "The same morning twice": three app notifications next to one One L1fe answer. | Text-only problem paragraph (homepage plan, block 2) | One ordinary moment explains the product to anyone (owner's avio.health reference). |
| Mission and Why | Kept word for word, "protect" echo included. Only the public line above the form changes (next row). | No change | Internal voice stays; the source doc wants Mission and Why to echo each other. |
| Mission line above the form | Make the most of the one life you have. | Protect the one life you have. | It is the Mission doc's own customer tagline; it covers performance now and long-term health, and it reads as an invitation, not a warning. |
| What | One L1fe is an app with AI coaches that turn your health data into a plan for today and the years ahead. | One L1fe is an AI app for personalized health coaching. | Says what you get; the category line tested weakest. |
| How | Kept word for word, with "Digital Twin" capitalized. Moves from the hero to How it works. | "How: the website hero" (usage note) | The how is the second layer, not the first sentence. |
| Value 3 | Whole. All your data, read as one person. | Holistic. All your data, one view of you. | "Holistic" is on the brand voice's own list of words we avoid. |
| Brand idea | Everything you measure, read as one life and turned into your next decision. | None; there was no one-sentence brand idea. | One line every page and screen can be checked against. |
| Fixed terms | "your watch" allowed in plain-language problem lines | "your wearable" everywhere | Testers were unsure what "wearable" means; "watch" names no platform, so the rule's reason holds. |
| Tone, new rule | Every number carries its meaning ("54, usual 48"). | None | Testers asked "is 54 good or bad?". |
| Words we avoid, addition | No sport or lab jargon above the fold (intervals, ferritin, HRV). | None | Both stopped testers in the hero. |
| Lines we keep | Kept: "Measured for you, not the average.", "A resting heart rate of 48", "What do you train for?" | The hero line moves out (see H1) | They work; the first now has its own section. |
| Visual basis | Paper, key green, five source plates; Archivo and Manrope across site and app; one Twin figure for site and app (today a placeholder, see the Twin section). | Light site with phone hero; dark particle Twin on predev; Fraunces in the app | One system, owned by the Twin, explained by the plates. |
| Tool | Why |
|---|---|
| Hand-written HTML, CSS and JS | The live site is static with no build step; the preview can move into it without a rewrite. |
| CSS masks on the app's Twin artwork | One image becomes five plates and the key figure, with no new 3D asset to keep in sync with the app. |
| Headless Chrome (the site's own tools/shot.mjs) | Same renders at 1440 and 390 every time, for review and for the 5-second test. |
| Fresh small-model testers, one image each | A cheap first 5-second test with no context leaking in. Real people should follow. |
| Wrangler | Deploys to the separate Pages project ol1-brand-preview, never the live one. Run with npx, nothing installed. |
| Python with Pillow | Contact sheets of competitors and renders for side-by-side review. |
| No 3D this round | The 2D registration figure explains the product; 3D would add load and cost before it adds clarity. The particle Twin from predev stays as a later motion layer. |