Clean Health Lab homepage options

Judge scores, in full

The in-stock test, Oct 23. You asked to lie to the judges: two test pages show all seven formulas as in stock and change nothing else (same layout, photos, no prices on the six, free bottle on L-Methylfolate only). They are for the blind judges only and are not on this board. A56 (A54 in stock): overall 8.99, the highest of any version, looks 8.98, feels like a homepage 8.47, would buy 8.92, first in both rounds. Against A54 on the same seeds, would buy rose from 8.00 to 8.92 and nothing else moved; GPT Astra went from 7.5 to 8.2 and 9.0, Claude Opus from 7 and 9 to 9 and 9, and no judge mentioned stock. A57 (A53 in stock): would buy 8.98, but homepage stayed 7.65, because its hero still pushes the free bottle. So the stock decides would buy and the hero decides the homepage score: a brand-first hero with the line in stock scores high on both. Every judge, every score

Both high, Oct 23 (A51 to A55). You asked for a high homepage score and a high would buy together. A53 (the judges' both-high plan: Shop all 7 as the hero button, the free bottle second with its terms, the line right under the hero with plain Backorder links) is the new Board A: overall 8.92, looks 8.85, homepage 7.67, would buy 8.74, a little above A46 on all four. A54 (A49 with the free bottle named in the hero link and the terms on its card) has the highest homepage of the day, 8.45, looks 9.0, overall 8.93, but would buy 8.00. A55 put the terms in the hero too: homepage 7.75, would buy 8.42. A52 (plain Backorder links on A51): 7.35 and 8.54. The pattern holds in every round: the four Sonnet judges give would buy a 9 on almost every version, while GPT Astra (7.0 to 7.6 on brand-first pages) and Claude Opus (7 to 9) hold it down, and both name the same cause each time: six of seven formulas are on backorder, so "Shop all 7" ends in dead ends. Their pages do not exist on the live store yet (checked Oct 23), so no card can honestly take an order. Every judge, every score

Would buy, Oct 23: eight builds from the blind teardown. A43 sold only L-Methylfolate (one-bottle hero, offer and terms, a buy box): would buy 8.90, the highest yet, but every judge called it "a one-product funnel, not a homepage" (4.87) and it fell to second behind Seed. A44 to A47 brought the brand back step by step. A46 (family hero, a pain headline, one button for the free bottle, a buy box, fuller reviews, need cards to the in-stock formula, an FAQ): would buy 8.68 (A42 8.05), overall 8.91 (A42 8.89), first in both rounds, but homepage 7.33. A49 (A42 with only the need cards and the fuller reviews): every score over 8, overall 8.90, homepage 8.09, looks 8.90 against Seed's 8.84 in the same rounds, would buy 8.35. A48 (the offer said fewer times) and A50 (A49 with an On backorder label) both lowered would buy (8.28 and 7.90): the more the six backorders show, the lower it goes. Every judge on every build gives the same reason for the gap: six of seven formulas are on backorder (Shopify, Oct 22: only L-Methylfolate has stock), so pushing the one in-stock formula lifts would buy and costs the homepage score. One A49 judge (Sonnet) mentioned "AI-looking bottle renders"; no other blind read did. The teardown · Every judge, every score

Why overall slipped while the looks gap closed. Overall: A42 8.89, A40 8.93, A41 8.96. The whole gap is round 1, where GPT Astra (8.4 to 8.2) and Sonnet 3 (9.2 to 9.0) each scored A42 0.2 lower; round 2 is 8.94 for both. That is inside the judges' noise: Seed's page never changed, and its overall ran from 7.84 to 8.24 across these six rounds. Our looks did not go up. Every judge gave every version a 9 in every round (one 9.1 on A40). The gap closed because Seed came down: Claude Opus gave Seed a 10 in both A40 rounds and a 9 in both A42 rounds. The one score that really fell is would buy, 8.71 to 8.05: Claude Opus went from 9 and 9 to 8 and 7, Sonnet 2 from 9 and 9 to 8 and 8. Their notes blame the six backorders, which they also named on A40, so the notes don't say what changed. The visible difference is that the new need cards picture the bottles, and three of the four lead with a backordered formula (our read, not a judge's). Every judge, every score

A42: your notes on A40 (Oct 22). "What's bothering you?" is rebuilt: one bottle per need, each on its own colour of paper and lit like that time of day (gold 3 PM sun, blue 7 AM window light, a grey rainy day, one light on deep green), with the time set on the photo. The shop grid gives every formula its own prop instead of seven identical photos, the lab section is a real steel lab bench instead of a fourth travertine scene, the folate chart has solid bars, and the review videos open on matching frames. Two rounds on the same seeds as A40: looks 9.0 against Seed 9.08 (A40: 9.0 against 9.27), the closest yet. Claude Opus no longer gives Seed a 10: 9 against 9 in both rounds (A40: 9 against 10). GPT Astra still has Seed ahead on looks (9.3 and 9.5 against 9.0). Its notes on us are about the six backorders and prices, never the pictures, and in a looks-only blind read it ranked us above Seed (9.0 against 8.7). Feels like a homepage 8.66 (A40 8.38), overall 8.89 (8.93), first in both rounds. Would buy fell to 8.05 (8.71), explained above.

A40 = A37 in plainer words. You asked to get "sounds human" up. In both A37 rounds the voice judges flagged the same lines: the hero "Stop taking your vitamins on faith.", "Nothing else in the capsule.", "Read this before you buy any vitamin.", the chart caption, the folic acid line and the footer. A39 rewrote them and sounded more human (8.50), but its hero ran into a bottle on desktop and read as vague. A40 has a plainer hero, "Most vitamins come in the wrong form.", that clears the bottles. Same two seeds against today's A37: sounds human 8.50 (7.88), would buy 8.71 (8.32), feels like a homepage 8.38 (8.13), overall 8.93 (8.93), looks 9.01 (9.0), first in both rounds. Seed sounds human 4.75. Today's A37 was re-judged because its timeline changed after the Oct 13 rounds.

A37 = A36 and A38 = A34, each with a new bottom photo. The old pour shot had a bottle wider than her palm with her fingers over the open top. The new one is fingertips holding one small pill with the open bottle on the ledge behind, the setup of your approved scale photo. Same two seeds: A37 overall 8.99 (A36 8.91), would buy 8.60 (8.33), feels like a homepage 8.59 (8.34), first in both rounds. A38 overall 8.92 (A34 8.88), would buy 8.12 (8.30). No judge mentioned the pill or the photos. Round 2 of A37 lost its GPT Astra design judge to a capacity error; that one judge was re-run alone.

A36 = A35 with Backorder back on the six unstocked formulas. Same two seeds: overall 8.91 (A34 8.88), would buy 8.33 (A34 8.30, A35 7.80), feels like a homepage 8.34 (A34 8.38), looks 9.0 (Seed 9.09). First in both rounds. So the hero's Get L-Methylfolate link, the proof on the first phone screen and the new photo surfaces are a wash to the judges: A36 and A34 are level, and the choice between them is yours by eye.

A35 tried the judges' own fixes and did not beat A34. Same two seeds: overall 8.88 for both, would buy 7.80 (A34 8.30), feels like a homepage 8.15 (A34 8.38), looks 9.0 for both (Seed 9.15). The reason is in every judge's note: "Restocking / Notify me" on six of seven formulas reads as sold out ("the grid reads as mostly sold out", "kills shopping momentum"), while "Backorder" read as something you can still buy. The hero's Get L-Methylfolate link, the proof on the first phone screen and the new surfaces did not move the scores either way.

A34 = A33 with the dark lab section running straight into the ingredients photo (a cream strip sat between them). Same two seeds: first in both rounds, overall 8.88 against Seed's 8.18, feels like a homepage 8.38 (A33 8.06) against Seed's 8.44. Looks 9.0, the same as every version since A28; Seed scored 9.26 this time.

A33 = A32 with one normal gap before the footer (two section gaps were stacking there) and the review videos leading with customers nearer our buyer's age. Two rounds on the same seeds: overall 8.94 (A32 8.91), looks 9.01 against Seed's 8.99, first in both rounds. Feels like a homepage came in at 8.06 (A32 8.45); neither change touches the page's structure, and the judges' fixes for it are a product picker near the top and the backorders.

Kevin's notes are in, and every pill photo now reads small. A32 is A30 with no prices on the six backorder formulas, need cards that line up, the comparison chart reading Clean Health Lab on the left like the ingredients chart, an unbroken bottle in the pour photo, and every capsule photo remade. Two blind readers (Claude Sonnet and Opus) estimated each pill's length in millimetres from the things around it. The new photos read 14 to 17 mm. Your approved C27 and C31 statics read 16 and 16.5, the avocado photo Kevin flagged read 20.5, and the real fingertip photo read 18.75 (a real capsule is 18 mm).

Same two rounds as A30: overall 8.91 (A30 8.90), feels like a homepage 8.45 (8.30), would buy 8.18 (7.95), looks 9.0 against Seed's 9.03. A32 placed first in both rounds. No judge mentioned AI photos or pill size.

Prices: 4 of 10 judges asked for prices on the product cards once they were gone from the backorder formulas. Would-buy still rose. The other fixes the judges asked for: stock (5 of 10, six of seven on backorder), a shorter page with a clearer place to start (3), and the free-bottle terms (1).

The rounds are blind. Random IDs, and the prompt never says which page is ours or that anything is generated.

Each version was judged alone against the same 8 competitor homepages, from phone and desktop screenshots (5 design judges and 4 voice judges a round). Scores out of 10. Bold green = best in the row. Average rank = the average place among the 9 homepages, where 1 is best. Animations and taps are not scored.

Side by side (average of each version's rounds)

Every judge, every score

One row per judge per round, the same two seeds for every version. Green = that judge scored A53 above A42, red = below. Grey = the score that judge gave Seed in the same round. Tap a score to open it.

Every judge on A53 vs Seed

Green = we beat Seed with that judge, red = Seed beat us.

Full tables, all 9 homepages