What happened, when

Every day of the test sorted by profit after ad spend. Breakeven means within $50 of zero either way. A sale is dated to the day the person was first seen and ad spend to the day it ran, so single days are noisy; the point is the shape, not one day. The columns re-sort when you change the price basis.

Simple version. Send the click to the design's own page, run few ads with real money each, show the length in the picture, and let the ladder build the cart. Those four things made every dollar of profit in the test.

Simple version. Sending clicks to Shop All, spreading the launch over 58 ads, six of the seven pockets, and every sharp budget jump. On the site, a sticky bar nobody used and a page that loses most readers before the fold. In how we ran it, two winners in one budget, ranking on checkout traffic, and cutting a winner while it was winning.

Simple version. Two ads made most of the money. Same words, same picture idea, different nail. A011 got less money and started later and still did as well as A010. Next: give each its own budget and see if they stay cheap.

A010 Garnet"This is the whole length." · lands on the Garnet page

Spend$571

Days with real money12

Sales36

Cost per sale$16

ROAS3.44x

Profit after ads+$602

Average cart$54

A011 Cobalt Tips"Short like you mean it." · lands on the Cobalt Tips page

Spend$329

Days with real money8 of 11

Sales22

Cost per sale$15

ROAS3.25x

Profit after ads+$305

Average cart$49

What happened well

• One idea sold: a nail pinched against the fingertip so you can see the length, and words short-nail women already say.
• Two ads on it, Garnet and Cobalt Tips, made 58 of the test's 154 sales on 17% of the money.
• $16 and $15 a sale against a $22 bar; +$907 after ad spend between them.
• Their buyers shopped around: carts of three to four sets, mostly other designs.
• The same words carried a dark design and a light one. The idea is the winner, not one design.

What we did to improve it, and did it work

• 09-01, moved the click from Shop All to the design's own page and rewrote the copy. Worked: A010 went from 1 sale on $65 to 36 sales on $571.
• 09-03, doubled the Short set from $50 to $75 a day. Held: both ads stayed under the bar.
• 09-10, raised it to $100; Meta moved the money to A011. Mostly held: A011 went from $9 to $21 a sale, still under; A010 got cheaper as its share fell.
• 09-03, tried the picture on five other designs at $10 a day. Worked until pushed: three of five sat at the bar; the two raised five-fold went to $64 and $65 a sale; the one cut was winning.
• Never done: giving A011 its own budget, or stepping an ad back down after a raise.

What we can do next

• Run A010 and A011 in their own ad sets at $50 a day, budgets fixed for the week.
• Beside them, four $25 copies of A010 that change one thing each: Bare Minimum as before, "Short like you mean it" on Garnet, Garnet next to a penny, Garnet with the comment-reply line.
• Read at $100 spent, on sales and cost per sale. Keep the wall and the sheet running three days after the ads stop.
• If both stay under $22 a sale: about $100 a day at the bar; a scaling campaign with a minimum under each; plan on about three sales a day per ad.
• If they get more expensive: turn each back to where it held and grow the copy that holds. Kill only if over the bar at the low budget for six days.

The numbers, side by side

• Per dollar they are the same ad: $16 against $15 a sale, 3.44x against 3.25x. A010 has more sales only because Meta gave it more money.
• A011 sends slightly fewer visitors to the checkout, but more of them pay.
• Both are doors into the catalogue: most of their buyers bought other designs and built carts of three to four sets.

Where they ran and how the money was shared

• Both were born at the launch (08-27) pointing at Shop All, where they did almost nothing. On 09-01 fresh copies were made that point at the design's own page, and the old ones were paused.
• They shared one ad set and one budget. Meta chose the split: A010 got 61% of the money, A011 35%, A013 4%. A011 got $0 to $4 a day for its first four days.
• Neither was ever paused on its own. They stopped when the whole campaign was paused on 09-14 at 1:04 AM.

Simple version. Two more ads cleared the bar, on smaller numbers. A095 shows a damaged nail growing back under a set over 14 days and sold 13 times at $23 each. A085 shows a nail next to a penny and sold 7 times at $23 each, with the biggest carts of any ad. Both are promising, not proven; both go into the next test with their own budget.

A095 Gold Leaf, day 0 / day 1 / day 14"It grows back under the set." · Conditions story · lands on the Gold Leaf page

Spend$299

Days with real money7

Sales13

Cost per sale$23

ROAS2.14x

Profit after ads+$88

Average cart$49

A085 Champagne Dots beside a penny"Short enough for short nail beds." · Short · lands on the Champagne Dots page

Spend$162

Days with real money7

Sales7

Cost per sale$23

ROAS2.48x

Profit after ads+$87

Average cart$57

What happened well

• A095: the picture is proof over time. One nail on day 0 (hidden), day 1 (a set going on) and day 14 (grown back). The first line names the person: "the nail you've been hiding". 13 sales at $23 each, 2.14x, +$88 after ads; 29% of the people it sent to the checkout paid.
• A085: a single nail next to a penny proves the size, and "Short enough for short nail beds" names the person. 7 sales at $23 each, 2.48x, +$87 after ads; 39% of checkout arrivals paid.
• A085 has the biggest carts of any ad: $57 and 3.6 sets on average, and 12 of its 25 sets were Almond, the shape the rest of the test barely bought.
• Both use A010's trick: the picture proves the claim, the first line says who it is for, and the click lands on that design's own page.
• A095 is the only Conditions ad that cleared the bar. The pocket can sell when the picture shows the outcome instead of a chart.

What we did to improve it, and did it work

• 09-06, built in iteration 2; 09-07 at 1:20 AM, switched on at $30 a day. A095 shared its ad set with one other ad (A093, which got $8); A085 shared with three penny and coffee-bean ads (which got $13 to $22). Meta gave A095 97% of its set's money and A085 76%.
• 09-09 at 9:35 PM, A095's set raised from $30 to $50. Same fade as every raised ad: $14 a sale before (7 sales on $101), $33 after (6 sales on $198). Still near the bar.
• 09-10 at 3:54 PM, Arjun paused A085's set; 09-11 at 5:28 PM it was switched back on. Before the pause $71 for 2 sales; after it $91 for 5 sales at $18. Whether the restart helped or the sales were simply late cannot be told apart.
• Never done: its own ad set and budget for either. A085's three set-mates never got a read.
• Two of A095's 13 sales are people first seen at the launch who came back and bought through A095's tag. Late sales are real sales, and they are why the last days of any ad read low.

What we can do next

• Both in their own ad sets at $50 a day, budgets fixed for the week (line-up sets 3 and 4).
• A095: the question is whether the Conditions story holds at $50 on its own money. It sits a dollar over the bar on 13 sales.
• A085: the same question on 7 sales. If it holds, re-run its three unread set-mates (penny with white dots, penny with red dots, the coffee bean) at $25 each, one per set.
• Read at $100 spent, on sales and cost per sale. Both stay "promising, not proven" until they clear $22 on five or more sales in their own set.

The numbers, side by side

• Per dollar they match: $23 a sale each, a dollar over the $22 bar, both well over 2x.
• A085 pays almost twice as much per visit as A010 ($0.48 against $0.26) and makes it back on bigger carts; A095 buys visits as cheaply as A010 and converts fewer of them.
• Both are on small numbers: 13 and 7 sales. Promising, not proven.

Where they ran and how the money was shared

• Both were built on 09-06 in iteration 2 and switched on by MJ on 09-07 at 1:20 AM, each at $30 a day in a set shared with other ads.
• *A095's total includes 2 sales from people first seen at the launch who bought later through its tag, so the daily column adds to 11.
• Neither ad was ever paused on its own. A085's set was paused by Arjun on 09-10 for a day. Both stopped when the whole campaign was paused on 09-14 at 1:04 AM.

Simple version. Run the four proven ads again, each with its own money, and see whether they stay cheap. Beside them, a handful of small copies of the winning picture with one thing changed. Touch nothing for three days, kill the dead ones on day four, and count the winners at the end of the week. If the winners stay under $22 a sale, we have a business; if they get expensive, we turn them back down and grow the copy that holds instead.

The line-up, about $325 a day

Rules for the week

• One ad per ad set, always.
• Every copy of A010 keeps A010's words and lands on the design's own product page with no shape preselected. Only the picture changes.
• Budgets are set on day one and never raised during the week.
• Nothing on the collection page.
• Tracking checked before the first dollar.
• Rank on sales and cost per sale, never on checkout arrivals.
• Keep the wall, the sheet and the dashboard running three days after the ads stop, so late sales land before the read.

What each result means

• If A010 and A011 stay under $22 a sale at $50 a day for six days: the two ads can carry about $100 a day at the 20% bar. Move them to a scaling campaign with a minimum under each. Plan inventory on about three sales a day per ad (arithmetic on the first test's rate, not a forecast).
• If they get more expensive: do not kill them. Turn each back to the budget where it held and let it keep selling there. Grow whichever $25 copy holds instead.
• Kill a winner only if it stays over the bar at the low budget for six days.
• Either answer is worth the money. The second one is worth more.

Simple version. Count money, not days. $100 spent decides a kill, $150 decides a winner, and the bar an ad has to hold is $22 a sale. Double a proven winner once and step back if it drifts. Never kill a winner to pay for its variations. The written rulebook is Ads/_AD-TESTING-RULES.md.

How to call each ad

• Count money, not days. No ad is judged before it has spent $100.
• $100 and 0 sales: kill.
• $100 and 5 or more sales at $22 or less each: winner. Hold six days, then double.
• $100 and 1 to 4 sales: halve the budget, count again two days later. 5 or more, restore it; 0 or 1, kill; 2 to 4, keep that budget and never raise it.
• $150 and over $44 a sale: kill.
• Between $22 and $44 a sale it is a keeper: it earns its own money and nothing more. Keep the budget it held at, never raise it. Raise it and you turn a keeper into a loser.
• At $50 a day an ad reaches $100 on day 2 and is decided on day 4. At $25 a day: day 4 and day 6.

The kill or keep table

The same thing as a picture

The CBO rulewhen a winner is big enough to leave the ABO

Left running too longby the rules above, ad by ad

Every ad replayed against the rules above. Total excess spend is the money spent past the point a rule said stop: past the $100 read for a loser, past $150 where the $44-a-sale rule applied, past the $300 keeper line for A045 on Tortoise, and the whole raise above the old daily level for the three ads raised on 09-10. Money is New Price.

1. Launch budgets, 08-28 → 08-30. phases.json: "Budgets cut $50 to $30 per set on 08-31". Shopify Apps/_STATE.md (08-30) and the build manifest: "7 × $30". Meta's activity feed shows the sets created at $30 and no budget event until 09-01. Spend is the referee: ≈$47 per set per day for three days, then $30 from 08-31, the sets were effectively at $50; the raise and the cut are simply missing from Meta's feed.
2. Gap + A013 force test pause time. phases.json / RUNBOOK say 09-05 02:20 AM PT; Meta's log and Ads/_STATE.md say 09-04 23:17 PT. Arjun's laptop was on the wrong timezone until 09-10; trust Meta.
3. Who switched iteration 2 on. Page text: "Arjun at midnight PT 09-07". Meta and Ads/_STATE.md: Mj Patio, 09-07 01:20 PT.
4. Design test "still PAUSED". Ads/_BUILD-MANIFEST-DESIGNTEST.md says built 23:39 PT and still paused; the sets were created 08:38 PT (23:39 was Manila) and Arjun activated them 10:56 PT the same morning.
5. Iteration 2 is 19 ads, not 17, and phases.json was never updated after 09-06: AS-14 and AS-15 (video) have no phase of their own, the exact failure mode the RUNBOOK warns about.
6. The consolidation caption is stale ("two ad sets at $75 each"): Conditions went to $50 on 09-09 21:35, off on 09-10 15:54; Short ran at $100 from 09-10 23:46.
7. Live theme. Workspace memory and theme-repo notes: Production_v1.2. Shopify: Production_v1.3 (price +2, ship 6) since 09-09 17:10, made in the admin, the theme repo has no commits after 09-02 and a dirty tree, so the live theme has drifted from git again.
8. The pricing canon is stale everywhere but Arjun's repo. _MASTER-CHECKLIST.md, Shipping Cost Model/_STATE.md, _PRODUCT-TRUTH.md still say $14 / $12 / $10 + $5 flat; the RUNBOOK still says $5 in two places; the discount titles in Shopify still read "($14/set)" and "($12/set)". Live since 09-09 18:00: $16 / $14 / $12, $6 flat under $60, free at $60.
9. Early spend totals. "$1,571 Aug 28 → Sep 3" (Ads/_STATE.md) vs "$1,657" (Shopify Apps/_STATE.md) were both mid-day reads; settled Meta spend 08-28 → 09-03 inclusive is $1,867 (through 09-02: $1,593).
10. Meta's activity feed is incomplete: no budget change before 09-01, no ad-level pauses after the build, nothing from 09-07 08:20 until 09-09 21:35, hourly snapshots of data/meta.json (Arjun's repo) fill the gaps and agree with the feed where both exist.
11. Pay-people totals by sensor: dashboard 154 people / $7,187 as seen / $7,829 repriced; sheet Pivots $7,151 (formula spill, lags by a few rows); Shopify abandoned checkouts 176 records (anyone who typed an email, a different thing). Use the dashboard / sheet YES rows for people, never the abandoned-checkout list.
12. The meta-ads connector currently holds the WRITE token (ads_management, business_management, expires 2026-10-25), memory and CLAUDE.md say it holds the read-only one. Only read calls were made today. Swap it back before anything automated touches it.
13. Dashboard chips still say LIVE / "Live now 0": the campaign has been paused since 09-14 01:04 and the hourly job has not pushed since 09-14 17:30.

Answered by MJ on 2026-09-15

• Why the price rose ($14 → $16, $5 → $6 shipping): Arjun's decision. Two aims: test whether a higher price cuts the number of people who press Pay (it did not seem to), and lift the margin. Site copy was changed to match; ad copy was left untouched so the ad data stays comparable.
• The end: planned. The test was budgeted at about $5,000; the campaign was paused at $5,214.
• The video ads: A102 (Dice) and A103 (Raspberry Velvet) are the same concept and copy, only the design differs.
• Direct-traffic spikes on 09-05 and 09-07: probably a mix of team QA and sharing; not known for sure.
• Team traffic: MJ made no test purchases during the test and believes team sessions were filtered.
• The yardstick for every phase: profit margin of 20% or more after ad spend, and intent ROAS of 2x or more. (Whole test: −3.8% and 1.51x. The consolidation wave alone: 21.4% and 2.44x.)
• Price basis for the readout: both, side by side, with a toggle.
• Headline count: the dashboard's 154 people are exactly the sheet's 154 YES rows (one row per person; the 7 rows marked duplicate are excluded by both), so there is no difference to choose between.
• The 09-10/11 re-allocation: Arjun's call. Switching off Conditions (A045/A054) was intended, to free budget for Short, Espresso Velvet, Ballet Tips and the two video ads.
• Launch budgets 08-28 → 08-30: Meta's change history has no budget entry for those days; the spend (≈$47 a set a day, then $30) says the sets ran at $50. Treated as $50.

Still open

• AS-9-Short went off on 09-10 (15:54) and came back on 09-11 (17:28); reason not recorded.
• What the deep dive must decide (inventory order, launch ads, pricing, budget), order constraints (minimums, budget) and timing are on hold until MJ has asked his own questions.
• The success metric was never written before spend; the 20% / 2x yardstick above is the after-the-fact standard.
• Wall message ("Sold out, this drop is gone for now", flagged 08-27 as false scarcity) never changed and never resolved.
• The theme repo has no commits after 09-02; everything later (the price copy, the pop-out) was done in the admin by Arjun and is dated above from the theme file history. The live site has not changed since 09-11 19:22.
• Abandoned-checkout export (Shopify deletes after 3 months), never scheduled.