The Three Quiz Funnels — what each one actually does
Deep analysis · every number re-pulled live 10 Aug 2026 (Stripe both rails · GA4 · Google Ads API · Meta Marketing API · customer360) · every headline number verified against two independent sources or explicitly tagged CHECKED ESTIMATE INFERRED
🔄 Updated 10 Aug 2026 — what changed since this page was built
Aga asked for every section refreshed with the sales that have landed since 7 Aug, every number re-verified, and one question answered head-on: where did every single sale come from, and did paid ads create any? That answer is the new §0b. Everything below has been re-pulled — nothing was carried over on trust.
| What | Said 7 Aug | True 10 Aug | Read this |
|---|---|---|---|
| v1 sales (provable, stamped) | 8 · $691.48 | 15 · $1,298.46 (11 Aug) | +7 sales since the page was built. Avg $86.56. 15/15 still active, zero refunds. §10 |
| The Meta leak — Variant 2 ads deep-linking v2-2 | 🔴 OPEN | ✅ FIXED | Both ads now point at the quiz root — read off the live creatives. Meta traffic to v2-2 hit zero on 8 Aug. Now in §19's ✅ done table. §3, §19 |
| Unintended traffic missing the $97 | 15% | 1.6% | Post-fix window 8–10 Aug. What remains is the Control — NIC-106). §3 |
| Did paid ads create sales? | "1 of 8, unclear" | YES — 3, all Google | 🔴 And all three are BRAND searches ("movement athlete"). Meta: 0 sales in August, agreed by two independent sources. §0b |
| YouTube traffic on v1 | "robot, stops ~9 Aug" | ✅ STOPPED 9 Aug — and excluded at source | 4,828 sessions, 84% of v1's window, 0 sales — but the daily series shows it died with the retrofit: 483 on 8 Aug, 0 on 9 Aug, 1 on 10 Aug. The morning's "still running" was a lump-window average, retracted 10 Aug PM. Both cockpit pullers now subtract it inside its fixed 3–8 Aug window. §3, §7, §19 done table |
| Competitive Non-Branded keyword surgery | to-do row 1 | 🔴 NOT DONE | Another £89.89 spent 1–10 Aug, still 0 conversions. Same 3 equipment keywords. §19 |
/register/ passes its name into checkout (part of NIC-91) | not shipped | 🟡 first fired 10 Aug | One subscription now carries checkout_version=register. It still carries no UTMs, so the old funnels remain unattributable. §1 |
What did NOT change: the ruling. v1 is still the one to scale, v2-2 still gets no traffic, the Control stays on Meta only. Three more days of data made that case stronger, not weaker.
🛠️ Nic's paid build landed — 11 Aug 2026, evening every line below read live from the platforms, not from the report
Nic reported three things: NIC-97 built, keyword surgery on Competitive Non-Branded | US, and six remarketing audiences started. Each was checked against the account itself — Meta Marketing API v21.0 (act_710789699289834 "TMA NEW", GBP) and Google Ads API v24 (customer 6909169207, GBP), pulled 11 Aug ~21:40 UTC. 🔴 Both ad accounts bill in GBP — the figures below are £, and are not converted.
| What Nic reported | What the account shows | Read this |
|---|---|---|
Build NIC-97 — the one geo-locked lead line | ✅ LIVE. New campaign QLG | ADV+ ON | MC | Variant 2 | Quiz Root — ACTIVE, OUTCOME_LEADS, £9.62/day, created 11 Aug 20:19 UTC. Destination read off the live creative's CTA link: https://quiz.themovementathlete.com/ — the ROOT. url_tags preserved. First spend £1.74 · 237 impressions. | The spec's three hard requirements are all met — root destination, leads objective, tracking tags kept. Verified from the creative, not the campaign name. |
| Geo-lock to buyer countries | 🟡 Covered, but split across two campaigns. The new line targets NL · CA · SE · GB · AU · FR · DE (plus PF/GF/AN/TF/BL — the French and Dutch dependent territories Meta bundles in automatically). US is not on it; it stays on QLG | ADV+ ON | US | Variant 2, ACTIVE at £9.62/day, whose live ad also points at the quiz root. | Fine — the spec left the implementation to Nic. All 8 buyer countries are covered and nothing points at a never-buying geo. Just remember it is two lines when the CPL is read on 17 Aug. |
| Pause MOF | ✅ DONE. MOF - Lead Form PAUSED 11 Aug 20:14:17 UTC (it spent £7.15 earlier that day before the pause). | The 99-leads-that-never-reach-the-quiz line is off. The D3 tripwire does not fire — NIC-97 was live before the 14 Aug deadline. |
| — (not reported, found in the check) | 🟡 QLG | ADV+ OFF | MC | Control is still ACTIVE — and still points at quiz.themovementathlete.com/2026/. Three live ad sets (£9.62 + £9.62 + £7.00/day = £26.24 of the account's £45.48 daily budget), £20.69 spent on 11 Aug. It is the account's biggest live line. | The budget itself moved exactly as the spec asked — MOF's £9.62/day became the new line's £9.62/day, and the account total is £45.48/day before and after. an earlier version of this row said spend "went up" — withdrawn, the arithmetic is flat What the clone branch left open is narrower: MC Control was the line the row named for re-aiming, and it is untouched, so 58% of Meta's daily budget still lands on the offer-less page. ✅ RULED 11 Aug — Nic repoints the 4 ads to the quiz root. Our own notes had disagreed: 7 Aug recorded this as the "deliberate Meta A/B control arm, kept on purpose for cheap multi-country leads" with an explicit "nobody re-flags them later as a defect"; 10 Aug Aga ruled "RE-AIM into the ONE geo-locked lead line". Aga settled it on 11 Aug: the 10 Aug ruling stands, the 7 Aug note is OVERTAKEN and now stamped as such in §3 and §6. The campaign keeps its 129 leads/30d at £2.43 — they just start landing on the quiz that sells. NIC-106 |
Keyword surgery on Competitive Non-Branded | US | ✅ DONE — and more thoroughly than first reported. Google's change log, re-pulled once it caught up, stamps two blocks on 11 Aug: 14:36:02 — bid changes on 5 keywords in this campaign (calisthenics bars, calisthenics equipment, calisthenics gear, calisthenics kit, calisthenics bars for home) plus 2 on Branded; 20:15:17–20:15:24 — the 3 keywords still spending (home calisthenics equipment, calisthenics home equipment, home gym calisthenics) PAUSED. That is both halves of the instruction — bid down AND pause. Six of 66 now paused in total (3 on 7 Aug, 3 last night). | 🔴 CORRECTION — this page briefly said the surgery "did not work" on the strength of 11 Aug's £9.11. That reading is WITHDRAWN and it was wrong. The changes landed at 14:36 and 20:15; almost all of that day's spend predates them. Judging a change with data older than the change is exactly the error our own laws forbid. The 11 Aug round is not yet judgeable — the first clean day is 12 Aug. |
| — (so what IS known) | Three things that hold regardless of last night's timing: (1) 30-day totals £235.89 · 32 clicks · 0 conversions AND 0 all-conversions, while both siblings converted (Competitors Branded £311.57 → 24; Non-Branded £241.33 → 6). (2) The 7 Aug round IS judgeable — it had four clean days after it, and spend simply continued: £7.61, £7.41, £6.22, £9.11. (3) The structural picture: all 60 still-enabled keywords are equipment phrases, and the campaign carries 0 negatives and 0 shared negative lists. Search terms bought 1–11 Aug: backyard calisthenics setup £9.94 · outdoor calisthenics gym £8.46 · calisthenics bar £8.30 · wooden parallettes calisthenics £7.91 · parallettes £6.00 · gornation pull up bar £4.77 · dip bar for home £3.26. | People shopping for hardware; we sell a programme. That is a targeting-class problem, and it is why the 7 Aug round moved the spend without changing the result. |
| — (the honest next step) | 🎯 A registered prediction with a read date, instead of a verdict. The structural case says spend leaks to the next equipment phrase, because 60 of them are still enabled behind zero negatives. Nic's round is the fair test of that. READ 13 Aug (two clean days, 12–13 Aug, with the pauses and bids in force): • Spend stays ~£6–9/day on hardware search terms → the set-level fix is needed, and AGA-157 is the move.• Spend drops materially and holds → Nic's round worked, AGA-157 stands down, and the campaign gets its clean fortnight to the 24 Aug gate like PMax and Non-Branded. | Either way the campaign has one more read before anyone touches it again — and the £235.89-for-zero record means the 24 Aug gate applies to it regardless of which way 13 Aug goes. |
| 6 remarketing audiences — web visits 7/14/30D and app installs 7/14/30D, non-purchasers | 🟡 Not on the account yet — which matches his own word, "started creating". Full pagination of all 212 custom audiences: zero created on or after 1 Aug 2026 (the newest is 3 Jul), and zero saved audiences. | Nothing wrong — it is in progress. But two minutes before he builds: the account may already hold most of this ladder. Sitting there since 3 Jul are Users that Logged-In No Purchase 1D/7D/14D/30D/60D/90D, Quiz Users→Paywall→Didn't Purchase (Control) and (V2-2), App Opens Last 30D, App Open 180D, Purchase Last 730D and 180days - webhit. Worth checking those before minting six more — duplicates fragment the audience and none of them will ever exit learning phase at £5/day. |
Where this leaves us: everything asked for was built, one thing is measurable now, one is not measurable until 13 Aug, and one question was left open by a fork the spec allowed.
Measurable now: the lead line is live and correctly built — root destination, leads objective, tracking kept, buyer geos — and the budget moved exactly sideways as specified (MOF's £9.62/day → the new line's £9.62/day; account total £45.48/day before and after). NIC-97 is closed as delivered. Open by design: the spec let the line be a clone or a re-aim; it was cloned, so MC Control — £26.24/day, 58% of Meta's budget — is untouched and still lands on /2026/. Somebody has to say what happens to it: NIC-106, 5 minutes. Not yet measurable: the keyword work landed at 14:36 and 20:15 last night, so the first clean day is 12 Aug and 13 Aug is the read; AGA-157 is staged behind that read, not ahead of it.
🔴 And an error of ours on the record, in the same voice we would use for anyone else's: the first version of this block called the keyword surgery a failure, citing 11 Aug's £9.11. The changes landed part-way through 11 Aug, so that day's spend mostly predates them — we judged a change with data older than the change. Withdrawn above. The same pull also showed we had under-reported the work: the 14:36 bid changes on five keywords were missing from our first read because Google's change log had not caught up when we pulled it.
🎯 12 Aug — THE REAL PROBLEM, AND YOUR LIST Aga: "should we scale Apple Search Ads? should we do affiliate? should we test the landing page? what shall we improve next?"
Written in plain English on purpose. Every number re-pulled live 12 Aug 2026 from the money systems, the ad platforms and the live queue — nothing carried over on trust.
THE REAL PROBLEM — and it is not which advert to buy.
1. We are filling a leaky bucket. In the last 30 days we won 150 paying customers and lost 129. So for every 100 people we bring in, 86 are just replacing someone who walked out the back door. Only 14 are actual growth. Every idea on your list — more Apple ads, affiliates, testing the landing page — is a way to pour water in faster. Not one of them plugs the hole.
2. Nobody in this company is looking at what we SPEND. This is the big one, and it is embarrassing. We have 17 agent teams and every single one of them is pointed at bringing money IN. Not one is pointed at money going OUT. So the largest and most certain pile of cash on the entire board — about $1,650–2,180 every month, for roughly an hour of cancelling subscriptions we do not use — has been sitting untouched. The report that found it is in a folder called RESCUED_TEMP_DOCS, was never added to the document index, and produced zero tasks in a month. Six of the cancellations are clean kills with no strings: Kajabi, Atlassian, The One Percent, Envato, Paddle Retain, Typeform. One line item — AWS at £296/month — is money we may simply be burning, because the app runs on DigitalOcean and nothing in our code mentions AWS at all.
3. You are the bottleneck, and the machine is making it worse. There are 51 open items waiting on you right now. The agent teams are excellent at finding problems and poor at finishing them, so everything they find becomes your inbox. On ~10 hours a week, roughly six of those 51 actually move the money. The rest are real, but they are tidying.
🔑 Put simply: we are arguing about how to pour faster, while a hole in the bucket goes unplugged and $2,000 a month of certain cash sits one hour of clicking away. That is why choosing between Apple ads, affiliates and a landing page test felt unsatisfying — none of them was ever the answer.
✅ YOUR LIST — six things, in this order, and what each one actually is
Ranked by money moved per hour of your time. Everything else on the board parks until these are done.
| # | What it is, in plain words | Your time | What it's worth |
|---|---|---|---|
1AGA-103 | THE CUT LIST — cancel the software we pay for and don't use. Six clean cancellations with nothing depending on them (Kajabi $187, Atlassian $130, The One Percent $74, Envato $71, Paddle Retain $37, Typeform $34 — all per month). Then one question each on a second tier, of which AWS at £296/month is the best five minutes on this whole page — our app runs on DigitalOcean and nothing in our code refers to AWS. Confirm every cancellation on the next statement, not on the cancellation screen — the screen lies. | ~1 hour | $1,650–2,180 every month CERTAIN |
2AGA-146AGA-147 | Say yes or no to emailing people we already have and have never asked. 6,813 people finished our survey and never did a workout. 14,600 people sit in an email sequence that has never once shown them the quiz. You are only saying "yes, draft it and show me" — you'll see the finished email and the safety check before anything sends. The first one deliberately tests 500 people before touching the rest. | 10 min | $0 to run the cheapest audience we own |
3AGA-150 | 🚨 A broken pipe has been throwing away Facebook leads since 1 December 2025. The connection between the Facebook lead form and our email system died — around 1,100 people filled in the muscle-guide form and never reached us. Facebook deletes them after 90 days, so only ~264 are still rescuable, and that window is closing. We are still paying ~£78/month into the broken pipe. Two clicks: export the leads, reconnect the integration. | 10 min | ~264 leads rescued today |
4AGA-161 | The email after someone's very first workout — the moment they're most excited, and we currently send nothing. About 48 people a day hit it. Everything on my side is already built and running nightly; what's missing is one small automation in the email builder (about six clicks), plus switching off "have you started yet?" emails for people who have obviously started. | 12 min | the highest-intent moment in the funnel |
5AGA-118 | Email the ~1,020 people who saw the price, clicked buy, and never paid. These are the warmest people we own and they have never been emailed. Not once. Same shape as #2 — you approve the finished email before it goes. | 10 min | best free audience we have |
6AGA-142 | 🔴 Three passwords are sitting in plain text inside our own files, left over from the Calisthenics Academy days, plus an old WordPress admin account that should not exist. Change the three passwords, delete the old admin. | 15 min | closes a real security hole |
That is roughly two hours of your time, and it is worth more than everything else on the board combined. Note on the other 45 open items: they are real, they are mostly small defects in emails and pages, and several are mine to do rather than yours. They wait.
⚠️ Read the 30-day, not the 7-day — the cockpit's own headline is noise right now
The cockpit currently leads with "churn is out-running sales — 57 lost vs 35 won in 7d". That window shows net −22 subs / −$552 MRR. The 30-day window shows net +21 / +$301. One week does not establish a trend — at these volumes a single bad week is inside normal variance. The business is roughly flat, not collapsing. Never let the 7-day line drive a decision; quote the 30-day beside it.
The four questions, answered — and what each one actually produces
| The question | Verdict | What it produces | Why |
|---|---|---|---|
| "Apple Search Ads are doing well — should we scale?" | 🔴 WE CAN'T. It is already as big as it can get | $0 more the money would have nowhere to go | All three of those ads only show to people who type our own name into the App Store. That is why they look brilliant — those people were already looking for us. There are only about 7,000 such searches a month, so if you tripled the budget there would be nothing extra to buy. And the same report shows 452 people found the app with no advert at all and 25 of them bought — so some of what the ads are taking credit for was coming anyway. Keep it running (it costs $146/month and stops competitors buying our name) — that's already decided. The only version that could grow is ads on general searches like "calisthenics app", which is switched off and has never been tried — that's a test, not a scale-up. |
| "Should we do affiliate?" | 🔴 CLOSED. Already adjudicated | ≈£275/mo 4% of the gap | FLEET_CHARTERS.json, verbatim: "Hold merchant side ≈£275/mo net with zero build hours; complete the publisher harvest (~4h) and STOP per the standing verdict." It was explored, sized and closed. It costs partner ops — founder hours — which is the scarcest thing we have. Do not reopen it. |
| "Should we test the funnel landing page?" | 🟡 NOT YET. Readable on leads only | ≈+$740/mo MODELLED | 704 quiz-landing visitors/7d. A +20% lift on the 24.4% lead rate reads in ~3.5 weeks. A lift in sales takes months — not readable before the clock. And leads are not the constraint: this document's own finding is that the funnel fills, and what fails is converting the warm people we already own. A 20% front-door lift ≈ +8 sales/mo. Real, but not the answer. |
| ⭐ "What shall we improve next?" → ASK THE ~24,000 WE ALREADY OWN | ✅ DO THIS FIRST | $0 cost 2 × 5-min decisions | 6,813 finished the survey and never trained (AGA-146) · 14,600 in 758 Nurture Sequence 2 who have never been shown the quiz (AGA-147) · 2,805 trained, never trialled. No spend, no build, no new asset — and both sit in Aga's queue right now as GO/NO-GO rows at impact 5/5. Nothing else on the board has that shape. AGA-146 deliberately tests 500 first because the rates behind it are modelled — keep that discipline. |
⭐ AND FIX THE HOLE — NIC-101, currently PARKED | ✅ DO THIS SECOND | $78.50 vs $34 per saved annual member | 84 of 115 people in the cancel queue are ANNUAL, and every one of them is shown $9.99/mo = $119.88/yr — against our own standing annual-at-50% = $78.50. We are offering leaving members a deal that is worse for them and worse for us. Measured across 516 people who ever took $9.99: median life 2.0 months, $34/taker — it does not save, it defers. 🔴 It must ship as a COUPON on their existing subscription, never a checkout link (they still hold an active sub — a checkout double-bills). |
⇒ IN ONE SENTENCE: spend an hour cancelling software we don't use, then spend forty minutes saying yes to emailing people we already own — and stop offering our leaving yearly members a worse deal than our own price list.
All of it is free or saves money. All of it fits inside the 81 days. All of it aims at the hole in the bucket rather than the tap. Everything else waits — including the "general search" Apple ads test, which should not open until we fix the Android sign-in problem. Right now 97 out of 100 iPhone users get signed in successfully, but only 68 out of 100 Android users do, using the exact same method. Paying to push more people through a door that jams is the worst possible version of pouring faster.
🔴 And the thing underneath all of it: 51 items are waiting on you, on about ten hours a week. The agent teams find problems faster than one person can decide on them, and only about six of the 51 move the money. That is the problem worth fixing — not Apple ads versus affiliates.
🔴 A retraction of our own, on the record — the Stripe portal toggle
An earlier version of this recommendation said the highest-leverage retention fix was "turn on subscription_update — a Stripe config change, not development." That is WITHDRAWN. It had already been ruled out on 11 Aug, in a section we did not read:
- July already ruled it OFF until Nic confirms webhook consumption — enabling it would desync entitlements.
- 🔑 Our web cancel button is CUSTOM, not the Stripe portal — so portal toggles do not govern the screen members actually cancel on. The portal being cancel-only is a true fact about the wrong surface.
How it happened: §5 of the cancellation master states the live portal config; §2a retracts acting on it. We read the section we came for and not the whole document — the exact failure the CHECK-BEFORE-YOU-ASK rule exists to stop. NIC-101 above is unaffected and stands. Source: CANCELLATION_AND_WINBACK_STRATEGY_2026-08-11.md §2a.
The 60-second version
758 Nurture Sequence 2 · 2,805 never-trialled — the cheapest audience on the boardThe short version: the new $97 quiz works, and the way traffic is currently being bought does not.
The quiz the root was repointed to on 31 July is selling: 15 sales in 11 days at $86.56 each, and every one of those buyers is still a member. It is also the only funnel we can measure at all, because it stamps its own name onto every sale in Stripe. That one write is why this document can exist.
The problem sits upstream. Of ~$770 spent on ads in those same ten days we can trace exactly three sales — all from people who typed "movement athlete" into Google. Every Meta campaign, four of the five Google campaigns, 1,569 clicks — no sale we can find. Meanwhile our own homepage and two blog posts produced 8 sales for £0. That ratio, not the ad accounts, is where the next hour of work belongs.
★ The recommendation — for the CMO
Questions to answer: What do we do? How do we win this? What do we need to scale? What does the data gathered so far actually say? — This is the strategic read. §19 below is the task list that follows from it.
⚡ THE 80/20 — if we only do FIVE things to reach $15K the focus question: what is the most important use of our time?
Ranked by expected dollars per hour of OUR time, from everything verified in this document. The gap is ≈$6,193/mo. Everything not on this list is either already running, already scheduled on a calendar date, or not worth hours this month.
| The action | Whose time | Cost | Why it is in the five | |
|---|---|---|---|---|
| 1 | GO the 758 Nurture Sequence 2 broadcast — today, send by 14 Aug. Short, one link, no price; the paywall closes. | Aga: the GO + 5 min on copy · fleet does the rest | $0 | 4,400 warm people, and the deep join just proved this exact pool produces buyers (the 3.6-year Nurture2 Completed win-back). The quiz closes 13% of price-viewers in the same session — it only needs arrivals. $1,500–3,000 modelled, this week. |
| 2 | Fix what the 9-of-15 already proved: the organic surfaces. CTA lift on the homepage hero + top 2 blog pages, by 17 Aug. | Fleet | $0 | 60% of all sales came from pages nobody optimises or owns. The cheapest permanent lift on the board — it compounds every week after. |
| 3 | Protect the machine that multiplies everything else: 993 SALES SEQUENCE. Enrolment-integrity check by 12 Aug; then leave it alone until the reads. | Fleet | $0 | Every lead — bought, organic, broadcast — flows into this arc. If its enrolment is broken today, the 10 Sep gate (which the whole paid plan waits on) reads garbage. |
| 4 | Nic's 45 minutes on paid: keyword surgery + build NIC-97 + pause MOF — one sitting, this week. | Nic, ~45 min | frees ~$1,050/mo | Stops the only live waste, starts the only paid test that matters, and the freed money is HELD for the gates — not respent. |
| 5 | BUILD Black Friday — the offer is already DECIDED: 70% off lifetime, yearly and quarterly (Aga, 11 Aug — corrected the fleet's "lock by 15 Sep": the decision was never open). What remains is build: the October warming arc, the BF pages + checkout links from the v3 map, the retargeting brief (~15 Nov). | Fleet builds · Aga approves copy | $0 prep | BF is the biggest cash event of the year ($20–35K, n=2 real events) and the payout of the whole lead bank. With the offer settled, warming can be READY early — 15 Sep becomes the build-done date, not a decision date. Context on record, not relitigated: prior 50%-off on recurring plans netted ~£688; the lifetime instrument added ~$13.8K — the lifetime leg is historically where BF cash comes from. |
💸 AD SPEND — WHAT WE DO RIGHT NOW, campaign by campaign the one box that answers it · settled 11 Aug, consistent with every section below
| Campaign | Now ≈/mo | ACTION THIS WEEK | Why |
|---|---|---|---|
| Google · Branded | ~£55 | KEEP, unchanged | The only paid line with verified sales (~£6/sale). Never scale it — it harvests. |
| Meta · Variant 2 | ~$230 | KEEP — it IS the leads test | Feeds the selling quiz since 8 Aug; its leads enter 993 SALES SEQUENCE. |
Meta · the NIC-97 lead lineQLG | ADV+ ON | MC | Variant 2 | Quiz Root | £9.62/d ≈£293/mo | ✅ BUILT AND LIVE — 11 Aug 20:19 UTC. Quiz ROOT, OUTCOME_LEADS, url_tags kept, buyer geos. | Verified off the live creative. First spend £1.74/237 impressions. Nic reports its real CPL 17 Aug — count BOTH this and the US Variant 2 line. |
| Meta · MC Control | £26.24/d 58% of budget | ✅ RULED 11 Aug — REPOINT to the quiz root (NIC-106, Nic, 5 min) | 7 Aug (this page): "deliberate — the Meta A/B control arm", "kept on purpose for cheap multi-country leads", and Nic's brief says "nobody re-flags them later as a defect." 10 Aug (Aga's NIC-97 ruling + the row above): "RE-AIM into the ONE geo-locked lead line → quiz root." The later ruling probably wins, but nothing stamps the supersession, so only Aga can say. It makes 129 leads/30d at £2.43 — exactly the job the 7 Aug note describes. NIC-106 |
| Meta · MOF Lead Form | £0 was ~£215/mo | ✅ PAUSED 11 Aug 20:14:17 UTC (inside NIC-97) | 99 "leads"/10d that never reached the quiz. 0 purchases in 77 days. Done. |
| Google · Competitive Non-Branded | £235.89 30d · 0 conv | ✅ SURGERY DONE 11 Aug — 5 keywords bid down (14:36) + the 3 remaining spenders paused (20:15). 🟡 READ IT 13 Aug, after two clean days. Set-level negatives are staged as AGA-157 behind that read, not ahead of it | Both halves of the instruction were executed. Last night's result cannot be read from last night's spend — the changes landed mid-day and late evening. What stands regardless: £235.89 / 32 clicks / 0 conversions and 0 all-conversions over 30 days, 60 still-enabled equipment keywords behind 0 negatives, and the earlier 7 Aug round which had four clean days and did not move the spend. 13 Aug decides whether this needs the set-level fix or just its clean fortnight to 24 Aug. |
| Google · PMax + Non-Branded | ~£420 | 🟡 LEAVE, read 24 Aug | Unproven (PMax's 3 claimed purchases have no Stripe match). The clean fortnight decides them. |
| Meta · warm retargeting | $0 → ~$150/mo | 🟡 AMENDED 11 Aug (Aga: "I don't think retargeting should be off") — TURN ON A SMALL TEST NOW: warm audiences only (site visitors 30d · quiz-starters · abandoners), OUTCOME_SALES objective, the $97 offer, ~$5/day from the freed budget. The full promo push still arms ~15 Nov for BF.🛠️ 11 Aug — Nic has started building the audience ladder: 6 audiences (web visits 7/14/30D and app installs 7/14/30D, non-purchasers). Not on the account yet — 0 of 212 custom audiences created since 1 Aug. 🔑 Check the existing ones first: Users that Logged-In No Purchase 7D/14D/30D, App Opens Last 30D, 180days - webhit and the two Quiz Users→Paywall→Didn't Purchase lists already exist from 3 Jul. | Her instinct has data behind it: every Meta sale ever came from warm retargeting, and the blocker the fleet named — "bench until attribution confirms" — is now met (the quiz rail stamps 6/6 sales). Honest caveat: it has only ever sold during dated promos (full-life £67.48/purchase); an evergreen-$97 retargeting test is genuinely NEW, so it is judged on stamped sales with a 4-week read. |
Net effect: spend drops from ≈$2,310/mo to ≈$1,150–1,250/mo, all inside the envelope — roughly $1,050/mo is freed and HELD, not respent, until the gates say where it earns: the 10 Sep 993 SALES SEQUENCE read (scale the lead line if ≥2–2.7%) and the BF sprint (15 Nov).
NIC-97 went live 11 Aug 20:19 UTC, so cold Meta is not pause-forced · 17 Aug — Nic reports the real CPL (count both the new Quiz Root line and US Variant 2 — the lead line is two campaigns) · 24 Aug — first clean read on PMax/Non-Branded + the 993 early look · 10 Sep — the gate · 15 Nov — scale retargeting into the BF push (the small $97 warm test runs from now, per the amended row above). Nothing else about paid changes before those dates.The full summary — everything this document found, and what I recommend rewritten 11 Aug on Aga's instruction; every figure verified in the sections below
1 · The machine we have — measured, not assumed
The $97 quiz is the only thing making a new sale on the web (§BQ: 15 of 15 new-subscription sales carry its stamp). It converts 13% of everyone who reaches its price (n=102), nets $86.56, and has kept 15 of 15 buyers with zero refunds. The deep join (§0b) added the fact that changes how to think about it: 13 of 15 buyers paid within ~20 minutes of first giving us their email — median 14 minutes. This is a same-session conversion machine: the buyer does not need weeks of nurture, they need to arrive and see the price screen. The corollary: every lever below is really one lever — put more of the right people in front of the quiz.
2 · Where its buyers actually come from — and where they don't
| Source of the 15 | Sales | What the deep join adds |
|---|---|---|
| Surfaces we own (homepage ×7, blogs ×2) | 9 | £0 spent. Unowned as a channel — nobody's job. The single cheapest growth surface on the board. |
| Google brand search | 3 | Harvesting people who already knew us (£17.87 total). Keep; never scale on its flattering ratio. |
| Untagged | 3 | 🔑 One decoded by the join: a 2022 customer who cancelled, completed 758 Nurture Sequence 2, and returned 3.6 years later to pay $97 — the warm pool demonstrably buys. |
| Meta · Email | 0 · 0 | Meta cold has never sold here (98 days). Email only began ASKING on 10 Aug (993 SALES SEQUENCE) — unproven, not failed. |
One buyer was lead-magnet-born (free Handstand guide → quiz → $97 in 14 minutes) — the magnet lane works end-to-end. And overnight, a quiz-checkout abandoner took the free trial instead of the $97, unprompted — the downsell path exists in the wild (§BQ trials table, and decision D4 below).
3 · Yes — we KEEP BUYING LEADS. The settled position, in Aga's terms
993 SALES SEQUENCE ask · the promo events, TMA's biggest cash days, sold to the existing list (July's $297 pull ≈$20K; BF 2025 ≈$31.6K) · renewals (§BQ: subscriptions from 2019 still paying). The gates never decide WHETHER to buy leads — only how many and from where.The operating shape: the geo-locked line (
NIC-97, GO'd) is the floor — never zero. The scale trigger is the linear break-even: 2.7% email conversion at $3.50/real-lead, 1.2% at $2 — identical at every budget level. So the highest-leverage paid work before 10 Sep is the creative that finds the $2 lead (no price, no offer — the assessment promise onto the quiz root). And every lead banked before ~24 Nov is a seat at Black Friday.4 · The honest state of each engine
| Engine | State · evidence | Next read |
|---|---|---|
| The quiz | ✅ Working — 15 sales, 13% price-view close, converts in-session | 17 Aug (CI tightens) |
| Owned surfaces | ✅ 60% of sales for £0 — but unowned and unoptimised | CTA lift this week |
Email — 993 SALES SEQUENCE | 🟡 Armed 10 Aug 13:12, fed via 750 TMA | Welcome Sequence After Quiz. Unproven, not failed | 24 Aug early · 10 Sep gate |
| Paid — brand | ✅ Harvests at ~£6/sale. Keep, don't scale | — |
| Paid — cold | 🔴 0 verified sales in 98 days. Now a maintenance LEAD line only | 14 Aug repoint check · 24 Aug |
| v3 trials | 🟡 8 real people; an APP-SIGNUP door, not a quiz door; +1 abandoner-downsell | 13–18 Aug first charges |
| The warm pools | 🟡 ~24,000 banked; one proven buyer already surfaced from them | the 758 broadcast — D1, needs Aga's GO |
| The app | 🔴 Partly unmeasured — see the honesty box | Play uninstall read by 12 Aug (D5) |
5 · 🔴 The app funnel — what I actually know, and what I do not
play_install_reports.py — decision D5, due 12 Aug); ② iOS needs the ASC analytics API; ③ the in-app journey needs the AppsFlyer tier call (Aga's, a cost). Until then no app number drives a budget decision here — and none does.6 · The recommendation, in order
| Move | Why this order | |
|---|---|---|
| 1 | Put warm people in front of the quiz. The 758 Nurture Sequence 2 broadcast (D1) · the organic CTA lift · protect 993 SALES SEQUENCE | The quiz converts in-session, so the constraint is arrivals — and the cheapest arrivals are the ~24,000 we already hold. The 3.6-year win-back proves that pool buys. |
| 2 | Keep the lead bank filling. NIC-97 runs; the $2-lead creative gets built against the real CPL (D2); every lead is a BF seat | Leads are the lifeblood — and BF (~24 Nov; the offer is DECIDED: 70% off lifetime/yearly/quarterly) is when the bank cashes. |
| 3 | Let the calendar decide the rest. 12 Aug Play read · 13–18 Aug trials · 14 Aug repoint check · 17 Aug CI + real CPL · 24 Aug Meta + 993 early look · 10 Sep gate · 5 Oct refunds · 15 Nov arm retargeting · ~24 Nov BF | Every remaining disagreement has a date on which it stops being an opinion. |
🔑 The single sentence for the CMO: this is a same-session selling machine with an audience problem and a full warehouse — the fastest path to $15K is marching the audience we already own past the price screen, while the lead line quietly fills the Black Friday bank behind it.
The arithmetic that decides it
Current run-rate, from the verified 10-day window: ~45 quiz sales/month ≈ $3,895/month. The KPI is $15K/month: that needs ~175 sales/month — 3.9× today — which is why the fleet plan below closes most of the gap through the warm list and BF rather than through this funnel's volume alone.
Where the QUIZ's sales come from — all 15, and they ARE the quiz's re-verified 10 Aug evening
Every sale in this table is a $97-quiz sale — each one carries checkout_version=quiz_checkout in Stripe, and the §BQ audit shows the quiz is 100% of new web sales since 1 Aug, so there is no other bucket these could blend with. (The count moved 14 → 15 when a sale landed on the 10 Aug afternoon pull.)
| Source | Sales | Share | What it cost |
|---|---|---|---|
| Surfaces we own — homepage hero ×5, nav ×2, TMA blog, BWTA blog | 9 | 60% | £0 |
| Google brand search — people typing our name | 3 | 20% | £17.87 |
| Untagged — origin unknowable | 3 | 20% | unknown |
| Meta | 0 | 0% | $292 |
| 0 | 0% | — never asked until today |
🔑 And the audience we already own, that costs nothing to reach
| Pool | People | If 1% buy | If 2% buy |
|---|---|---|---|
| Finished the survey, never did one workout | 6,813 | $5,847 | $11,694 |
Sitting in 758 Nurture Sequence 2, sent the quiz nothing | 14,600 | $12,530 | $25,059 |
| Trained, never started a trial | 2,805 | $2,407 | $4,815 |
One percent of just those two pools is $18,377 — more than five months of the current run-rate, from people we have already paid for. MODELLED — 1% and 2% are illustrative rates, not forecasts. Reactivation cohorts typically underperform new traffic, and none of these pools has been asked yet.
How we win — three moves, in order
| The move | Why it is first | Cost | |
|---|---|---|---|
| 1 now | Harvest what we already own. Let 993 SALES SEQUENCE run and protect it. Point 758 Nurture Sequence 2 at the quiz. Send an activation email to the 6,813 who never did a workout — starting with 500 and reading the click rate. | The largest audience we have, the only one that costs nothing, and the one asset that has never actually been asked for the sale. Until today no arc did. | £0 |
| 2 next 30 days | Make the rest measurable. The app-door click event, the keyword surgery, the /register/ stamp for the legacy rail. | ✅ CORRECTED 10 Aug — Aga challenged "70% untraceable" and she was right. As of today, 100% of August-created sales are stamped and traceable. What remains dark is inherited history only — renewals of pre-July subscriptions ($1,310/10d) and July trials maturing ($312/10d) — which no code change can retro-tag and which shrinks by itself. The go-forward funnel is fully visible. | ~2 h |
| 3 only after 1 and 2 | Then, and only then, buy traffic — warm first. Retarget quiz visitors with the $97 offer. Keep Google brand search. Treat cold Meta as an experiment with a kill date, not a budget line. | Every pound Meta ever returned came from retargeting a warm audience. We have never made cold acquisition work — 98 days of evidence. | inside the envelope |
The sequencing is the recommendation. Doing 3 before 1 and 2 is what we have been doing, and it is why $770 bought three sales.
What we need to scale — and what would have to be true
| To scale this… | …this must first be true | Status today |
|---|---|---|
| Email as a revenue channel | An arc actually asks for the sale, and converts ≥2% of quiz leads | 🟡 993 SALES SEQUENCE armed today — clock starts now, read ~10 Sep |
| Owned content (already our best source) | We know which pages produce sales, and can make more of them | 🟡 8 sales traced, but nobody owns this channel — the biggest gap on the board |
| Paid acquisition | A traceable sale at under ~$76, from a repeatable audience | 🔴 Only brand search qualifies, and it harvests demand rather than creating it |
| The app as a funnel | People who install actually train — and we can SEE whether they do | 🔴 Partly unmeasured: 44% of signed-up-with-email contacts never do a workout, installs-who-never-sign-up and deletions are invisible today (Aga's Slack: ~half delete — plausible, unverified). Play uninstall read due 12 Aug (D5). |
| Any of it, honestly | We can attribute a sale to its source | ✅ DONE for everything new — 100% of Aug-created sales are stamped (Aga's correction, verified fresh 10 Aug PM). Only pre-Aug history stays dark, and it shrinks monthly. |
🔑 The single sentence I would put in front of the CMO: we do not have a growth problem, we have a harvesting problem — roughly 24,000 people we already paid for have never been properly asked, and the machine that asks them started running today.
💰 The paid decision — Aga's three questions, answered with the experts re-run 10 Aug PM · both experts bound to the mental-models doctrine · every claim carries its n
✅ First, a retraction that strengthens the case: the leads-for-email loop IS wired
993 SALES SEQUENCE. Aga challenged it — "we only get leads from the quiz, they go straight to AC" — and she is right. Verified on the AC API: 993 SALES SEQUENCE has no direct entry trigger, but 750 TMA | Welcome Sequence After Quiz carries an enter-block that pushes its contacts into it — and the entrant log shows it running live (82 loaded at arming, 18 more arriving in the 13:00 hour, after it armed). So the pipe exists today: quiz lead → AC → 750 (activation) → 993 SALES SEQUENCE (the ask). The experts' "wire it first" gate below is therefore already satisfied — struck where it appears.Q1 · "What if we optimise Meta for LEADS — that's what we did in the past?"
The model is now rational to TEST, for the first time. What made it irrational until this week is gone: the quiz captures the email before the price, every August sale is stamped, and — as of today — a bought lead flows into an arc that actually asks for the sale. What it still lacks is the one number that decides it: email's conversion rate, which has existed for a few hours.
| If email converts quiz leads at… | $3k/mo buys (~2,200 leads @ $1.35) | Verdict |
|---|---|---|
| 1% | ~22 sales ≈ $1,900 | 🔴 losing ~$1,100/mo |
| 2% | ~44 sales ≈ $3,800 | 🟡 ~break-even |
| 3% | ~66 sales ≈ $5,700 | ✅ ~1.9× return |
MODELLED — the 2% break-even assumes the $85.82 average holds and the $1.35 CPL holds at scale (it will drift up as budget rises — marginal cost exceeds average cost). The rate is measurable by ~10 Sep (early look ~24 Aug).
Q2 · The campaign-by-campaign verdict (the Meta expert's, amended for the wiring correction)
| Campaign | Spend/10d | Expert verdict | Why |
|---|---|---|---|
| QLG ADV+ ON | US | Variant 2 | $76.11 | KEEP — the test vehicle | The only campaign feeding the selling quiz, on-regime just 2 days (since the 8 Aug repoint). Its leads now flow into 993 SALES SEQUENCE. This trickle IS the leads-for-email test, already running at $228/mo. |
| QLG ADV+ OFF | MC | Control | $145.86 | AGA'S CALL — the expert said pause; the wiring correction weakens that | The expert's pause rested on "its quiz has no $97 offer, so leads cannot convert." True at the paywall — but under the LEADS model its leads also enter AC → 750 → 993 and get the email ask. Multi-country leads at ~$1.14, feeding the same email machine. If the model being tested is leads-for-email, this is a second (cheaper, non-US) arm of the same test. If the model is paywall-conversion, it is dead spend. The choice of model decides this row. |
| MOF — Lead Form | $70.49 | PAUSE — both readings agree | In-Facebook form: the person never reaches the quiz, so no paywall AND (unverified whether form leads even enter 750's flow — they carry no quiz tags). 0 purchases in 77 days (n=0). Structurally outside both models. |
Q3 · "Do we have enough data?" — the analyst's split (accepted in full)
| DECIDABLE NOW (no more data needed) | MUST WAIT — for what, until when |
|---|---|
|
✅ Kill the 3 equipment keywords (£89.89/10d, 0 conversions of any kind, wrong intent by design — the error costs are asymmetric) ✅ Pause MOF — Lead Form (structural, not statistical: n=0 in 77d and the leads never reach the funnel) ✅ Keep Google Branded (3/3 Stripe-corroborated at ~£6/sale — cheap harvest) ✅ Harvest the owned pools (6,813 + 14,600 + 2,805 — needs no paid data at all) ✅ |
⏳ ~17 Aug — +7d of quiz sales: the 13% price-view close (n=102, CI roughly 8–23%) narrows; decides the allowable-CAC model ⏳ ~22–24 Aug — the first clean Meta fortnight since the 8 Aug repoint: does cold Meta produce ANY stamped sale at $97? ⏳ ~24 Aug early look · 10 Sep gate — 993 SALES SEQUENCE conversion: ≥2% scale the leads model · <1% kill it (pre-registered now, before anyone peeks)⏳ 5 Oct — the refund window closes; settles net LTV for any real budget increase |
Q4 · So what do we do this week — act or wait?
Both. Act on the structural items (they need no data); wait on the scaling decisions (they have dates). This week: kill the equipment keywords, pause MOF, keep Variant 2 trickling as the live test, harvest the owned pools, protect 993 SALES SEQUENCE. Then the calendar does the deciding: 17 Aug → 22–24 Aug → 10 Sep → 5 Oct, each date unlocking a named decision.
🫀 "Leads are the lifeblood" — the paid decision at the COMPANY-SYSTEM level Aga's challenge, 10 Aug evening — and it changes the kill line
Aga: "Does this take into consideration that we NEED leads — they are the lifeblood? The paid decision needs to take into consideration the whole company system." It did not, fully. Here is the system view, and it amends the recommendation.
What a lead is actually worth — all four payouts, not just the first
| Payout | When | Evidence |
|---|---|---|
| 1 · The paywall, same session | day 0 | 13% of price-viewers buy (n=102) CHECKED |
2 · The email ask — 993 SALES SEQUENCE | days 1–30 | armed 10 Aug; rate readable ~10 Sep NOT KNOWABLE YET |
| 3 · PROMO EVENTS — the one the gate ignored | 2–3× a year | The July $297 lifetime pull took ~$19,997 in 11 days, and every traceable sale went to the EXISTING list: 20 × $297 through ActiveCampaign email + 6 × $297 through Meta retargeting of warm audiences. Black Friday 2025 took ~$31.6K the same way. Those buyers were leads bought months or years earlier. CHECKED — Stripe metadata |
| 4 · Renewals | for years | §BQ: subscriptions born 2019–2025 paid $1,310 this ten days CHECKED |
The 10 Sep gate prices a lead only on payout 2. But payouts 3 and 4 are where TMA's biggest cash days have always come from — roughly $50K/year of promo pulls land on the list that lead-buying built.
What the system view CHANGES — three amendments to the recommendation above
| # | Before (funnel view) | After (system view) |
|---|---|---|
| 1 | <1% email conversion on 10 Sep = kill all lead campaigns | 🔴 AMENDED: never kill to zero. <1% = shrink to a maintenance lead line (the cheapest arm that passes the geo-quality bar) and let Black Friday 2026 price the list. Why BF is in this sentence at all: TMA's biggest cash days are promo events sold to the EXISTING list (July $297 pull ≈$20K in 11 days, all traceable sales to people already on it; BF 2025 ≈$31.6K the same way) — and those buyers were leads collected months or years earlier. A weak 993 SALES SEQUENCE read proves the everyday drip is weak, not that a lead is worthless; the November event is the real market test of a lead's value, and every lead bought before it is a seat in that audience. Killing lead-buying in September quietly shrinks November's payday. |
| 2 | Leads judged on CPL and 30-day conversion | Leads judged on CPL × GEO QUALITY. The app-install lesson applies with full force here: 6,700 installs from IN/PK/PE produced zero buyers. A cheap lead from a never-buying country is a seat in an audience that never buys promos either. Whatever campaign carries the lead line, it geo-locks to buyer countries. |
| 3 | Value of paid = traceable sales ÷ spend (21¢/$) | That ratio ignored payouts 3+4 — which is exactly the mistake in reverse of the one this page kept correcting. The honest statement: fast payback is measurable and poor; system payback is real, slower, and partly modelled. Both belong in the decision. |
🎯 THE MC CONTROL DECISION — the one Aga asked to see on its own
QLG | ADV+ OFF | MC | Control — $145.86/10d (~$437/mo), multi-country, ~$1.14 leads, feeding the /2026/ Control quiz (7-day trial → $157, no $97 offer).The three readings, honestly:
🔴 Paywall view (the Meta expert): pause — its landing page cannot sell the $97, so day-0 revenue is structurally zero.
🟡 Leads view (Aga's model): keep — its leads land in AC, enter
750 TMA | Welcome Sequence After Quiz → 993 SALES SEQUENCE, and get the email ask like everyone else, at the cheapest CPL we have.🟢 System view (this section): keep the JOB, fix the GEO. The lead line earns its place through payouts 2+3 — but "MC" is the risk: multi-country leads carry the app-install failure mode. Its lead geography was measured on 7 Aug: only ~20% US (CA 24% · UK 22% · AU 15% · FR 9%) — those ARE buyer countries, so it passes the bar the app installs failed. India-class geos are not in it.
🟢 RULED BY AGA, 10 Aug evening — "I need this." Filed as
NIC-97 with her GO and the full spec. One lead line worth ~$150–200/mo, geo-locked to proven buyer countries (US·UK·CA·AU·DE·SE·NL·FR), pointed at the quiz root (so its leads see the $97 AND enter the email machine — the Control page gives up payout 1 for no reason). That is MC Control's budget doing MC Control's job, minus its two defects (the geo tail and the offer-less landing page). Whether that line is the re-aimed MC Control or a widened Variant 2 stays Nic's implementation call — the decision itself is MADE. One consequence folded into the same row: MOF — Lead Form pauses (its budget funds the line). 🔴 Judge this line at the REAL quiz-lead CPL (~$6.57 GA4-lens, see §BQ), never the $1.35 platform figure.
🔑 The one-line answer to "does the paid decision consider the whole system": it does now — leads stay bought in every scenario; what the gates decide is only HOW MANY and FROM WHERE.
🏁 THE FLEET'S RANKED PLAN TO $15K — the full re-run on the corrected data tma-growth-lead · 10 Aug evening · bound to the mental-models doctrine · Aga's dual objective: $15K pace AND the BF lead bank
Re-run after two of its earlier inputs were corrected (email "0%" was a machine measured before switch-on; $1.35 was the platform CPL lens). KPI framing per Aga: $15,000/mo ASAP — the gap is ≈$6,193/mo.
The verdict: the gap closes through the owned list, 993 SALES SEQUENCE, and Black Friday — not through paid. Paid runs exactly as Aga decided (NIC-97) as lead-pipeline maintenance while the email machine proves itself.
993 SALES SEQUENCE's conversion rate.The ranked plan
| # | Lever | Owner | Cost | $/mo impact | Time to cash |
|---|---|---|---|---|---|
| 1 | Warm-pool broadcast — 758 Nurture Sequence 2's engaged ~4,400 → the quiz root (utm 758broadcast; collision + segment-integrity gates MUST pass first) | Aga GO · fleet executes this is AGA-129, now concretely spec'd | $0 | $1,500–3,000 one-off + $300–400/mo MODELLED | 3–7 days |
| 2 | 993 SALES SEQUENCE at steady state — new leads flowing through the ask | fleet monitors · Aga gates 10 Sep | $0 | $650–2,611/mo at 2–8% MODELLED | 24 Aug early · 10 Sep gate |
| 3 | CTA lift on the top organic surfaces — homepage hero + top 2 blog pages (9 of 15 sales came from these, unowned, unoptimised) | fleet | $0 | $300–600/mo MODELLED, n=9 | 3–7 days |
| 4 | BF 2026 — offer locked by 15 Sep, October email warming, the pulse ~28 Nov | Aga (offer) · fleet (arcs) | $0 prep | $20–35K in Nov CONFIRMED, n=2 events | Nov |
| 5 | v3 trial pipeline — email + blog CTAs at the trial checkout (8 clean trials in 4 days post-fix, first charges 13 Aug) | fleet | $0 | $500–1,200/mo MODELLED, n=8 | 13 Aug first read |
| 6 | The paid geo-locked lead line — runs exactly as Aga decided (NIC-97) | Nic | $150–200/mo | $50–230/mo + ~25 buyer-geo leads/mo into the BF bank | 30+ days |
| 7 | Churn diagnosis — every 1% off churn ≈ $88/mo on the current base; closes the Q4 residual | fleet | $0 | DARK until measured | 60+ days |
| 8 | Apple Search Ads pilot | Aga, only after 10 Sep | $300–500/mo | UNVERIFIED | 60+ days |
This week's three concrete actions
- ① By 17 Aug: the
758 Nurture Sequence 2engaged-segment broadcast to the quiz root — after the collision gate andsegment_guard.pyPASS, on Aga's GO. Measured on its own UTM, purchases not clicks. - ② By 12 Aug: the upload-integrity gate verifies all current
993 SALES SEQUENCEentrants receive the arc correctly (plaintext twins, the enter-block on750 TMA | Welcome Sequence After Quiz) — an enrolment defect today would corrupt the 10 Sep gate entirely. - ③ By 17 Aug: CTA audit + lift on homepage hero and the top two blog pages — expect +15–25% organic quiz-starts MODELLED.
The decision calendar — now including BF
| Date | What arrives | What it decides |
|---|---|---|
| 13–17 Aug | the 8 clean v3 trials hit their first charges | <6 converting = checkout-friction diagnosis |
| 17 Aug | quiz rate CI tightens (~30 sales) | below 10% = look at quiz screens first |
| 22–24 Aug | 993 SALES SEQUENCE first opens/clicks + the clean Meta fortnight | zero opens = delivery/enrolment diagnosis, not copy |
| 10 Sep | the 993 gate | ≥2% → Aga may expand NIC-97 · <2% → single worst-eligible slot per the Evolution Law |
| 15 Sep | BF build READY (the offer is already decided: 70% off lifetime/yearly/quarterly — Aga, 11 Aug) | Warming arc written, pages + v3 checkout links staged, retargeting brief drafted — so October warming starts on day one |
| 5 Oct | refund window closes on Aug sales | first clean monthly cash picture |
| ~28 Nov | BF 2026 | target $20–35K n=2 prior events, sold to the bank everything above has been filling |
NIC-97's ~$150–200/mo yields roughly 25 real leads and ~$146/mo — pipeline maintenance, not a growth engine. The fleet would not scale paid before 10 Sep at all, calling an earlier step-up a proxy-optimisation error. This page had proposed a softer ladder (step to ~$800/mo on a positive 24 Aug early-look). Both positions are recorded; the 24 Aug read is Aga's decision point between them. If 993 SALES SEQUENCE proves ≥5%, the budget conversation reopens on 10 Sep with a real multiplier case.🎯 THE DUAL-OBJECTIVE UPGRADE — the fleet's second pass, on Aga's directive ("$15K while maximising leads for BF")
993 SALES SEQUENCE must convert 2.7% for lead-buying to be profit-positive — at $200/mo, $800/mo or the full $1,671/mo alike. At $2.00/lead the break-even drops to 1.2%; at $5.00 it rises to 4.3%. So CPL matters more than budget, which means the most valuable paid work before 10 Sep is not choosing a number — it is the CREATIVE that might find the $2 lead: no price, no offer, just the assessment promise ("find out exactly where your calisthenics starts — 5 minutes, free") landing on the quiz root, where the email is captured naturally and the $97 appears only after the results.| Meta lead spend | CPL scenario | Leads banked by BF (~24 Nov) | Incremental BF revenue | Break-even 993 rate |
|---|---|---|---|---|
| $200/mo — NIC-97, live now | $3.50 | ~223 | $138–245 | 2.7% |
| $200/mo | $2.00 if the creative finds it | ~390 | $242–429 | 1.2% |
| $800/mo | $3.50 | ~891 | $553–980 | 2.7% |
| $1,671/mo full Meta envelope | $3.50 | ~1,862 | $1,155–2,048 | 2.7% |
| $1,671/mo | $2.00 | ~3,258 | $2,020–3,584 | 1.2% |
MODELLED — BF yield $0.62–1.10/contact from the two prior events (n=2, CONFIRMED); CPL scenarios span the platform/GA4 lens tension. And the baseline nobody pays for: organic is already banking ~336 leads/month — ~1,310 by BF for £0, worth $812–1,441 of BF revenue before a single paid lead is counted.
993 SALES SEQUENCE converting above the break-even, which is why 10 Sep is the gate. (2) 🔴 Before anything scales, verify the pipe: this week, prove NIC-97's leads actually arrive in AC carrying NEW quiz-* tags (quiz_lead_signal_check.py, by 15 Aug) — the MOF lesson, applied in advance. Two new calendar entries: 15 Aug (NIC-97 AC-arrival check) and 19 Aug (first BWA magnet-strip read — switch-on law) · 15 Nov (arm Meta warm retargeting for the BF window — 🔴 its previously-quoted £2.41 CPA is WITHDRAWN: full-life truth is £67.48/purchase; the £2.41 was a window artifact covering 1.7% of the spend).🔑 The 80/20, in one line: the vital 20% is the $97 quiz (100% of new web sales) and the 14,600 warm contacts who have never once been sent to it — and the wasteful 80% (cold paid, the equipment keywords, the form-fills) is already stopped or ruled. The constraint is not volume of leads; it is how many warm people are shown the price. That is exactly what lever 1 does first.
📋 THE DECISION SHEET — 11 Aug the growth fleet run on the FINAL data (the deep join, the trials join, the app concern) · decision standard applied
| # | Decision | Who | The fleet's recommendation | Tripwire / date |
|---|---|---|---|---|
| D1 | The 758 Nurture Sequence 2 broadcast (~4,400 engaged → quiz root) | AGA — GO pending | GO, send by 14 Aug — and the in-session finding changes the DESIGN: the email's only job is session creation — short, ONE link, zero price mention; the paywall does the closing (13/15 buyers close in ~14 min). The 3.6-year win-back is the proof this pool buys. | <0.5% session rate on send → pool is cold, no repeat. Signal readable 21 Aug. |
| D2 | Lead-line scale (creative pre-build?) | Nic · Aga above cap | Hold $200/mo until NIC-97's ACTUAL week-1 CPL lands — the break-even formula's denominator is currently assumed, not measured. CPL ≤$2.50 → build the $2-lead creative test; >$4 → hold. | Nic reports real CPL by 17 Aug — a 5-minute pull. |
| D3 | Cold Meta: pause now vs hold to 24 Aug | Nic · Aga if reallocating | Hold to 24 Aug ONLY IF the NIC-97 repoint is live by 14 Aug — otherwise pause immediately; spend without attribution has no measurement case. Read = total stamped quiz sales by source, never Meta's own lead count. | 14 Aug repoint check · 24 Aug decision |
| D4 | The abandoner→trial downsell (it happened organically overnight) | AGA — strategy ruling | n=1 is an existence proof, not a rate. Add a soft P.S. to the abandonment note offering the free trial — $97 stays the primary ask. Costs nothing, surfaces the organic mechanism, doesn't displace recovery. Economics: trial→retained nets more than $97 over 6 months IF retention holds; cannibalising a would-be $97 costs ~$72/instance. | Wire before 17 Aug · watch the $97-vs-trial split on the next 30 notes · read at the 10 Sep gate |
| D5 | The app install/delete question (Aga's Slack: ~half delete) | Fleet (Android) · Aga (iOS/AppsFlyer cost) | Run the Play uninstall read NOW — a ~30-min extension of play_install_reports.py; park iOS until it lands. This matters before 10 Sep: if D30 retention is very low, the 993 SALES SEQUENCE gate is mailing people who are already gone, and the gate's read changes. | Play read by 12 Aug · escalate to Aga if D30 <40% |
| D6 | What the new evidence killed | — | 🔴 The £2.41-CPA premise (above) · 🔻 any "nurture drip before the paywall" design — a 14-minute conversion event cannot be accelerated by an email; email's job is creating the session, not closing it. Unchanged: 993 SALES SEQUENCE runs · 750 stays activation-first · NIC-97 at $200. | — |
AGA: the GO on the
758 Nurture Sequence 2 broadcast — $0, the only lever that can produce sessions this week, aimed at the pool that just produced a proven buyer. The fleet builds it; you approve the copy before send (gates: collision + segment_guard.py PASS attached).NIC: report NIC-97's actual first-week CPL — one number, five minutes, and it unlocks or defers the whole lead-scale decision.
The risk in this recommendation, stated plainly
The main case against it: owned-audience harvesting is finite. You can convert the 24,000 once. If that is all we do, the run-rate rises and then flattens, and we will have spent 60 days not solving acquisition — which is the thing that actually compounds.
Why I still recommend this order: the harvest is what funds and informs the acquisition work. It produces cash in weeks rather than months, and — more importantly — it tells us what a converting message looks like before we pay Meta to show it to strangers. Trying to buy cold traffic while our own list has never been asked is the expensive way round.
Your three questions, answered first
/2026/, Meta multi-country, ACTIVE) is ~100% Facebook/Instagram-fed and still cannot be shown to have earned anything in August — six August payments came from July trials maturing on the shared checkout ($291.87 total), and none of them records which page sent the buyer. That is not the same as $0; it is unknowable. See §0b./2026-v2-2/, so the root redirect could never catch them.
Re-read off the live creatives on 10 Aug, both now point at https://quiz.themovementathlete.com/ — the root.CHECKED — Meta Marketing API, 10 Aug
The traffic agrees: Meta sessions to v2-2 ran 5–25/day to 7 Aug, then 0 on 8 Aug and 0 on 9 Aug. The page itself is still up (HTTP 200, no redirect — checked on the wire 10 Aug), which is correct: it is the rollback.
v2-2 is now dark — live but unfed. That job is in §19's ✅ done table.✅ The measurement problem is now closed (10 Aug evening): the YouTube crawler put 4,828 sessions on v1 in 12 days — 84% of everything v1 received, 3 leads, 0 sales — but the daily series shows it stopped when the retrofit completed 9 Aug (0 sessions 9 Aug, 1 on 10 Aug), and both cockpit pullers now subtract it inside its fixed 3–8 Aug window. Every v1 rate on this page has it removed, and from today the cockpit's own numbers agree. §19 done table.
First — the three names used all the way through this page
| We call it | It is | Its offer, in one line |
|---|---|---|
| v1 · "the new one" | /2026-quiz-checkout-v1/ — what the quiz root sends everyone to since 31 Jul | Pay $97 now for 12 months. No trial. |
| v2-2 · "the old one" | /2026-v2-2/ — what the root used to send everyone to | Free 7-day trial, then $24.97/month. |
| Control | /2026/ — the oldest build, only ever used by one Meta campaign | Free 7-day trial, then $157/year. |
Every number below says which of the three it belongs to. Where a figure is an estimate rather than a direct measurement it is tagged ESTIMATE; where the underlying system can't tell us directly and we inferred it, INFERRED.
The five numbers that matter — in plain words re-pulled 10 Aug
2026-quiz-checkout-v1 has taken in its first 11 days2026-quiz-checkout-v1NIC-106).🃏 THE MAVERICK TAKE — the divergent bench, in its own voice run 11 Aug 2026 on Aga's instruction
What this is: the Maverick bench (/maverick-fleet) run against this document. Six generators worked blind — each saw only a one-page pack of verified facts, never the plan above, never our strategy docs — each with one thinking lens and one deliberately ordinary vantage point (a 1990s catalog merchandiser, a referral-only wedding photographer, a country-club membership director, a farmers-market stall owner, a hotel revenue manager, a marching-band director). A seventh seat, the Challenger, DID read the plan and attacked it. They produced 87 ideas; a curator then checked each one's logic (did it work elsewhere for a reason that also exists here?), kept the ones whose worst case is small and capped but whose best case is big, weighed each against your ~10 hours a week, and advanced the top 5. 🔴 By law this section is never blended with the growth fleet's voice — you are always shown both: what the growth fleet recommended, and what the mavericks think. Nothing here has passed the skeptic or the decision standard; if you want one pursued, it becomes a decision memo like anything else.
What the growth fleet recommended (their view, one line): march the ~24,000 warm contacts we already own past the price screen (the
758 Nurture Sequence 2 broadcast, organic CTA lift, protect 993 SALES SEQUENCE), keep the geo-locked lead line filling the Black Friday bank, and let the decision calendar rule the rest — 24 Aug early look, 10 Sep email gate, no paid scaling before the gates read, BF 2026 cashes the bank.🃏 What the mavericks see instead (their view, one line): the same evidence — every big cash pull came from the existing list plus a dated, priced offer — says the list eats meals, not drips: run the dated-event machine as a calendar instrument starting in September (before the 1 Nov break-even clock, not 23 days after it), ask every buyer for the second yes at the receipt, and turn the assessment result itself into distribution.
The top 5 — each causally vetted, each with its cheapest test and its downside cap
🃏 M1 · The September Meal — run the dated-event machine as a calendar instrument, not a holiday
Where it came from: four seats converged on this blind — the cataloger (the Lands' End 8–13-drops-a-year calendar), the hotel revenue manager (fenced rates + invented occasions), the club director (a numbered "charter class", not a markdown) — and the Challenger independently made it his centre.
The mechanism, vetted: every large cash event in TMA's own record was the existing list + a dated, priced offer (July 2026 lifetime-$297: ≈$19,997 in 11 days; BF 2025: ≈$31.6K). Catalog cadence and hotel rate-fencing show a list absorbs far more asks when each event is fenced to a segment that has never seen one (fenced = the offer is visible ONLY to that chosen group — the public price never moves) and wears a fresh reason. The Memorial-Day counterexample (≈+£688 net) was a 50%-off subscription discount; July won as a lifetime product event — a different mechanism, so the pull-forward fear generalises from the wrong case.
What it could 10x: the one proven cash machine goes from ~2 events/yr to 6–10 — and a meal lands before the break-even clock.
Cheapest test: design ONE fenced September event (lifetime family, ~half July strength, capped units, audience = a slice of the 14,600 never-asked nurture pool) — through segment_guard.py, the broadcast-collision check, and a decision memo. Black Friday's decided offer stays untouched.
Downside cap: pull-forward measured at Memorial scale ≈ £688, capped further by the fence and the unit cap. Honest collision: the plan above holds email asks until the 10 Sep gate — this is the mavericks' explicit disagreement with it, and it is Aga's call, nobody else's.
🃏 M2 · The Bounce-Back — ask for the second yes at the receipt
Where it came from: the cataloger (the Fingerhut pack-in — the bounce-back catalog in the parcel out-pulled every prospect mailing they ever ran) + the hotel revenue manager (the trade-in upgrade desk).
The mechanism, vetted: the hottest name in any file is the buyer of ten seconds ago. At TMA it sits unused twice: all 15 quiz buyers paid same-session and none was asked a second question; and the renewal base (45% of web cash, some subscriptions since 2019) has never been offered the suite. Shapes: a $97 buyer gets a 72-hour "your $97 counts toward lifetime" certificate; a 2019-tenured renewal gets "$200 of what you've already paid credits toward lifetime — pay once, never renew again."
What it could 10x: revenue per buyer, on zero new traffic.
Cheapest test: ONE slot in the post-purchase buyer welcome (the tma-buyer-welcome-specialist seat, via the email fleet + the upload-integrity gate). Honest read time ≈4–6 weeks at ~1.5 buyers/day. The credit mechanics are a pricing decision — Aga's memo.
Downside cap: seen only by people who already paid; worst case it is ignored.
🃏 M3 · The Rank Card — the assessment result becomes a shareable status object
Where it came from: the marching-band director (the "chair placement card" — "Pull section · Level 7 of 19 · top 22% of assessed adults"); the Challenger picked the same move independently; and the SEO bench's own M1 ("the assessment IS the product") recorded it a third time. Three independent convergences is itself signal.
The mechanism, marked honestly: a computed rank is a status object that invites exactly one question — "what chair would I be?" — and the only answer is the quiz. It works at the source because rank is legible and flexing has a native audience. WHETHER PEOPLE WILL ACTUALLY SHARE IT IS UNPROVEN — this test exists to answer exactly that
What it could 10x: the funnel's top grows with its own throughput — every completer becomes a distribution node at $0, versus $2–5 per bought lead.
Cheapest test: a fleet-built share-card image emailed to the next ~100 quiz completers (opt-in share, UTM'd); count arrivals from shared cards in 14 days.
Downside cap: an ignored email. Privacy capped hard: opt-in share only — the "public results by default" variant is parked, not advanced.
🃏 M4 · The Status-Preserved Win-Back ladder — "we kept your progress"
Where it came from: the hotel revenue manager (elite win-back preserves tier status; lapse-tenure pricing) + the wedding photographer (the anniversary rail) + the club director (the reinstatement window).
The mechanism, vetted: hotels win back elite churn by preserving status, not by discounting — loss-aversion on something the person already owns. TMA's existence proof: a 2022 cancel returned after 3.6 years unprompted and paid $97 (§0b). ~100k historical users sit behind a $9.99 win-back rail that has 8 payers and zero distribution. The doctrine-clean frame: "your level history is still on file — resume where you stopped," priced by lapse tenure. (The coercive siblings — fake reactivation fees, expiring assessments — were demoted: unenforced claims.)
What it could 10x: the largest unasked audience the company owns becomes a laddered system instead of an 8-payer accident.
Cheapest test: 500 lapsed ever-paid contacts through segment_guard.py (this audience is ever-paid by design), tenure-laddered copy, 14-day read.
Downside cap: unsubs among long-lapsed contacts, watched by the deliverability seat; public price integrity untouched — the ladder is only ever seen by the lapsed.
🃏 M5 · The Equipment-Box Channel — stand the quiz in someone else's high-intent foot traffic
Where it came from: the farmers-market stall owner (the QR card in every pull-up-bar/rings box, rev-share to the seller) + the wedding photographer (the vendor web: physios, chiropractors, climbing-gym coaches discharge people who need exactly this and hand them a photocopied sheet).
The mechanism, vetted: TMA's own measured truth is that only warm/intent traffic has ever converted (own surfaces + brand search; cold Meta ≈$444/purchase). A person unboxing a pull-up bar, or being discharged from rehab, is at peak intent — and the partner has zero post-sale relationship, so a rev-share QR card is found money on their side. Jesse's rotator-cuff story is the sales letter to the physio.
What it could 10x: a standing warm-intent channel at $0 media that compounds with partners instead of budget.
Cheapest test: the outreach fleet pitches 10 equipment sellers + 10 physio clinics with a one-page rev-share offer (through the outreach gate); count replies in 14 days. Zero spend.
Downside cap: outreach emails + a rev-share that pays only on sale; brand exposure capped by the outreach gate.
🃏 The Challenger's counter-take — the one maverick seat that saw the plan UNEDITED, by law
THE PLAN'S HIDDEN ASSUMPTIONS: (1) 🔴 "The list must be warmed, not asked" — the 14 Aug broadcast is deliberately price-free, and the only automation that asks waits for a 10 Sep gate. (2) 🔴 "Promos borrow demand" — one Memorial-Day datapoint (a 50%-off subscription) is generalised to all events, so the next dated offer is 24 Nov, 23 days AFTER the 1 Nov break-even clock. (3) 🔴 "Held budget is prudence" — $1,050/mo freed and parked while the single proven paid mechanism (warm retargeting during dated promos, ~£67/purchase) gets $5/day. Also silent: "the quiz is the funnel" (it's a 14-minute checkout for pre-sold intent, not a converter); "lifetime is a promo-only SKU" (it's what the list actually buys); "the 6,813 never-worked-out and 2,805 never-trialed stay unasked until the gate clears."
IF WE REVERSE THEM: (1) The list eats meals, not drips. Every large cash event in the record — July $297 → ~$20K/11 days, BF'25 → $31.6K — was the existing list + a dated, priced offer. Reverse: the 14 Aug email is the warm-up; a priced, deadlined sequel follows within 72h. (2) A September event, not promo abstinence. Memorial failed as a subscription discount; July won as a lifetime product. Run the July mechanism at half strength in early September — even half closes the entire $6.2K gap in one event, before the clock, with BF untouched. (3) Spend where it's already proven. Don't test retargeting at $5/day in a vacuum; deploy the freed $1,050 as the retargeting layer OF that September event — the only context in which Meta has ever produced a purchase.
HOW I'D BEAT US: With $50K I'd clone the assessment quiz in four weeks (the funnel's behaviour is measured and copyable: 20% capture, 13% price-viewer conversion, 14-minute buys), sell a $49 no-trial first year, and farm TMA's own open distribution — 869 video comment sections, the BWA-shaped queries, calisthenics Reddit. What I cannot buy at any price: the 83k list, 8 years of renewals (45% of cash), 1.9M impressions of owned content. The moat is the graveyard, not the quiz — and the plan defends the quiz while leaving the graveyard unasked until mid-September. Monetising and locking that base (September event + lifetime/annual upgrades of the renewal book) is the moat-build; a funnel is not one.
THE 10x FRAME: $90K/mo ≈ 1,000 sales/mo — 3.3× the entire quiz traffic at 100% conversion. So 10x cannot come from improving this funnel; the swap is "sell to our list" → "own the free-assessment category": the quiz result as a shareable artifact ("push 4/12 · pull 2/19") making every completer distribution, feeding a monthly July-style event machine. Three tests under $500 combined, runnable now: the shareable results card (fleet-built, $0); a $97-with-deadline email to 500 of the 6,813 dead app signups ($0, 72h answer); $150 behind July's best retargeting creative on a 7-day lifetime window.
THE STEP THAT SHOULDN'T EXIST: The no-ask layer. Zero buyers have EVER come from email; 13/15 buyers paid within 20 minutes of giving an email. Eliminate the offer-free broadcast doctrine; Raise the dated event to a quarterly first-class instrument; Reduce cold acquisition to ~zero (the plan's one frame-complete element — credit); Create price-at-the-moment-of-signal everywhere the quiz already proves it: app signups and magnet leads see the $97 immediately, not after an arc.
ONE SENTENCE FOR AGA: The plan assumes drips are food and events are a drug, but eight years of this company's own cash says the list eats exactly one meal — a dated offer — and the plan schedules the next one three weeks after the break-even deadline.
993 SALES SEQUENCE). It is recorded here as a maverick position; only Aga can reopen that ruling. ② "Zero buyers ever from email" is true of the measured window, but the first automation that ASKS only switched on 10 Aug 13:12 — its read arrives ~10 Sep. The claim indicts the old doctrine, not the new arc.All 87 ideas + Aga's 2 additions — in plain words, with effort vs money added 11 Aug on Aga's ask: "I want to see all ideas"
Effort: 🟢 = the agent fleet can do it now — costs you a GO, almost none of your hours · 🟡 = needs a decision from you (usually a price ruling) and/or a few hours of Nic · 🔴 = needs real development or real-world operations — weeks, not days.
Money: 💰💰💰 = could close a big slice of the $6.1K/mo gap by itself, anchored to something we've actually measured · 💰💰 = meaningful — hundreds to a few thousand dollars · 💰 = small or slow · ❓ = honestly unknown until tested.
Where it stands: ⭐ = in the top 5 above · 🧪 = strong next-test candidate · 🅿️ = parked with a revisit date in the ideas ledger · ⛔ = demoted, kept visible with the reason — knowing why we said no is worth as much as the idea.
Numbers discipline: every $ figure in these tables is MODELLED — arithmetic from a measured anchor (July's $19,997 event · BF 2025's $31.6K · the 15-buyer cohort), never a promise.
If you only look at one table — the 14 best uses of effort, ranked by likely result per hour of your time
| # | Move | Effort | What it could realistically produce | First read |
|---|---|---|---|---|
| 1 | The September event (⭐M1 · A2) | 🟡 one GO + the email fleet | $5–10K one-off if July repeats at half strength MODELLED — and it proves whether events can run 6–10×/yr instead of 2 | ~2 weeks after the send |
| 2 | The receipt upsell + tenure trade-in (⭐M2 · B1–B2) | 🟢 one email slot + your ruling on the credit | ~$500–1,500/mo steady MODELLED: ~45 buyers/mo, 1 in 15 upgrading ≈ $1,200/mo — plus a possible $3–8K one-off from the long-tenure trade-in MODELLED, wide | 4–6 weeks |
| 3 | The win-back 500-test (⭐M4 · C1) | 🟢 fleet, through the safety gate | the test itself: $250–2,500 MODELLED: 0.5–5% of 500 at ~$97 — but the real prize is the READ: the full past-member file is ~100k people | 2 weeks |
| 4 | "Alumni" instead of "cancel" (C8) | 🟡 Nic changes the cancel flow + your GO | attacks the biggest bleed — churn is out-running sales (52 lost vs 38 won in 7 days); every cancel diverted to a $9.99 alumni tier keeps the door open | ~1 month after live |
| 5 | Smarter email segmentation — value cells + goal-matched streams (G6–G7) | 🟢 an email-fleet brief | lifts every send we ever make; the magnet→quiz→$97-in-14-minutes sale already proved goal-matching works | next broadcast |
| 6 | The $2-lead ad creative — sell the assessment, never the subscription (E16) | 🟢 a creative brief; feeds the lead line you already GO'd (NIC-97) | cheaper leads into the machine that closes 13% of price-viewers — the plan already wants this creative; the cataloger just named the angle | 2–4 weeks of spend |
| 7 | The Rank Card (⭐M3 · D1) | 🟢 fleet builds it | honestly unknown — but if even 1 in 10 shares it, the quiz's top grows for $0 | 14 days |
| 8 | Black Friday mechanics bundle — charter class + early member preview + $25 deposit book (A5–A7) | 🟡 mechanics only — the decided 70%-off offer is untouched | BF 2025 made $31.6K; mechanics that lift show-up 10–30% are worth $3–9K MODELLED | 24 Nov |
| 9 | Newsletter barter + co-branded guides (E5–E6) | 🟢 outreach fleet | rents other people's warm lists for £0 — and warm audiences are the only kind that has ever converted here | replies in 14 days |
| 10 | The equipment-box + physio channel (⭐M5 · E1–E3) | 🟢 outreach fleet | a slow-building standing channel at $0 media; nothing is owed until a partner's customer actually buys | replies in 14 days |
| 11 | Gift memberships for Q4 (D5) | 🟡 Nic builds a simple gift flow | every gift = cash now + a brand-new member at $0 acquisition, on Christmas timing | Q4 |
| 12 | The 90-second mini-assessment widget (F13) | 🔴 a dev build | the big one IF built: BWA's 1.9M search impressions get a taste of the product on the page instead of a link to a form | after build |
| 13 | Wake the affiliate machine we already own (H1) | 🟢 affiliate + outreach fleets — £0, commission pays only on sale | the program exists (Tapfiliate + the affiliate checkouts) and nets ≈£275/mo, mostly dormant; one cycle wakes dead partners AND signs the bench's E-band targets as tracked affiliates | replies in 14 days |
| 14 | Paid traffic behind the blog posts that measurably convert (H2) | 🟡 $5–10/day inside the existing envelope | a warm-up step for the cold traffic that never bought when asked directly — judged on cost per real buyer-geo lead vs the $2–5 the quiz line pays | 🔒 gated: the winners list lands ~19 Aug |
A · Sell to the list we already own — dated events, the "meals" (13 ideas)
| # | The idea, in plain words | Effort | Money | Where it stands |
|---|---|---|---|---|
| A1 | Run a small dated sale 6–10× a year instead of 2 — each to a different slice of the list that hasn't seen one, each with a fresh reason (New-Year posture, the member's own birthday…). — the catalog boss | 🟡 | 💰💰💰 July made ~$20K in 11 days; this makes that repeatable | ⭐ inside M1 |
| A2 | The September event itself — a half-strength July-style lifetime sale, early September, BEFORE the 1 Nov break-even date. — the challenger | 🟡 | 💰💰💰 $5–10K one-off MODELLED | ⭐ M1's centre |
| A3 | Invent small occasions for the quiet months (the app's birthday, "Assessment Week") — rotating who sees them so nobody gets more than ~2 a year. — the hotel manager | 🟡 | 💰💰 | ⭐ inside M1 |
| A4 | "Members-only price" — sale prices that exist only in a chosen segment's inbox; the public page never shows a discount. — the hotel manager | 🟡 | 💰💰 sells without cheapening the public price | ⭐ inside M1 (the "fence") |
| A5 | Sell Black Friday's 70%-off lifetime as 500 numbered "charter memberships" — a founding class that closes forever, not a markdown. — the club manager | 🟡 mechanics only | 💰💰 lifts BF + protects price integrity | 🧪 BF mechanics |
| A6 | Longest-paying members see Black Friday a week early, framed as a privilege. — the catalog boss | 🟢 | 💰💰 | 🧪 BF mechanics |
| A7 | $25 refundable Black-Friday reservations from September — deposit-holders become pre-committed cash AND the perfect warm ad audience. — the photographer | 🟡 a simple payment path (Nic) | 💰💰 turns one weekend into a 90-day pipeline (BF'25 = $31.6K) | 🧪 BF mechanics |
| A8 | Once a year: 48 lifetime seats auctioned live, price starts $497 and drops $10 every hour until they're gone. — the hotel manager | 🟡/🔴 | 💰💰 one-off theatre | 🅿️ |
| A9 | "$50 TMA credit, expires Sunday" instead of "50% off" — losing something you already own beats deciding to want a discount. — the hotel manager | 🟢 an email copy test | 💰💰 same price math, stronger psychology | 🧪 |
| A10 | Finish your assessment → entered to win a lifetime membership. — the catalog boss | 🟡 competition rules need care | 💰 | 🅿️ |
| A11 | Enrollment happens in monthly "seasons" of 250 places that genuinely close, with a waitlist. — the photographer | 🔴 changes the always-open funnel | ❓ | 🅿️ |
| A12 | Price the calendar — full price for January/September starts, cheaper named cohort starts in the dead months. — the photographer | 🟡 a pricing ruling | ❓ | 🅿️ Aga-only |
| A13 | The $97 price "holds for 72 hours pending membership approval", then reverts to $157 — with a monthly cap on places. — the club manager | 🟡 and the cap must be REAL | 💰💰 87 of 102 price-viewers currently walk away | 🅿️ honest only if the cap is real |
B · Ask the buyer again — money at and after the receipt (9 ideas)
| # | The idea, in plain words | Effort | Money | Where it stands |
|---|---|---|---|---|
| B1 | The receipt certificate: "your $97 counts toward lifetime for 72 hours." The hottest buyer in the world is the one from ten seconds ago — and we've never asked them a second question. — the catalog boss | 🟢 one email slot | 💰💰 ~$1,200/mo if 1 in 15 of ~45 buyers/mo upgrades MODELLED | ⭐ M2 |
| B2 | The tenure trade-in: "you've paid since 2019 — $200 of it counts toward lifetime. Pay $297 once, never renew." — the hotel manager | 🟢 email + 🟡 your price ruling | 💰💰💰 a one-off pull from the safest audience we have MODELLED, wide | ⭐ M2 |
| B3 | One cheap add-on tick-box at checkout (printed roadmap, mobility module) — like monogramming on an order form. — the catalog boss | 🟡 Nic | 💰 a few extra $ per order | 🅿️ |
| B4 | A $2,997 "Documented Year" above lifetime — quarterly filmed reassessments, a printed portfolio, a coach call. — the photographer | 🔴 real service delivery — Jesse's hours | 💰💰 ten buyers ≈ the whole gap, but it costs founder time | 🅿️ |
| B5 | 25 numbered "Member-Owner" seats at $1,997 — lifetime, name in the app, a quarterly roadmap vote. — the club manager | 🟡 deliberately uncomfortable — your call | 💰💰 ~$50K if it sells MODELLED | 🅿️ flincher |
| B6 | A $49/yr member-only extra: quarterly video form-review + a yearly progress report. — the club manager | 🔴 a real service | 💰 | 🅿️ |
| B7 | Lifetime on installments — $41 × 12, "lifetime layaway." For many households the monthly figure, not the price, is the decision. — the catalog boss | 🟡 Stripe config | 💰💰 | 🅿️ pricing |
| B8 | Add your partner or teen for +$49/yr — their own assessment, one bill. — the club manager | 🔴 accounts build | 💰💰 | 🅿️ |
| B9 | "$29 locks your renewal price forever" offered to long-tenured members. — the club manager | 🟡 | 💰 | ⛔ charging loyal members to keep their own price reads as a shakedown |
C · Bring past members back (13 ideas — ~100k historical users sit behind a win-back rail with 8 payers)
| # | The idea, in plain words | Effort | Money | Where it stands |
|---|---|---|---|---|
| C1 | "Your progress is still on file — resume exactly where you stopped." No discount needed; the thing they already own is the offer. — the hotel manager | 🟢 500-person test via the safety gate | 💰💰💰 the 3.6-year returner proves the file is alive | ⭐ M4 |
| C2 | Different comeback offers by how long they've been gone — 90 days vs 3+ years get different doors. — the hotel manager | 🟢 | 💰💰 | ⭐ inside M4 |
| C3 | The deadline version: "former members rejoin with no joining fee — by Friday." — the club manager | 🟢 | 💰💰 | ⭐ inside M4 (copy variant) |
| C4 | On the anniversary of their first assessment, auto-send their own old results: "this was your level that day. Retake it." Personal, non-promotional, runs forever. — the photographer | 🟡 an automation build | 💰💰 | 🧪 |
| C5 | A yearly "movement statement" of THEIR data — with what the people who started alongside them can now do. — the hotel manager | 🟡/🔴 data build | 💰 | 🅿️ |
| C6 | Three escalating goodbye letters — "we're closing your file" — ending in one deep offer. — the catalog boss | 🟢 | 💰💰 the catalog trade revived 5–15% of dead names this way (their number, not ours) | 🅿️ tone risk — after M4 reads |
| C7 | A real paper letter to churned payers' home addresses — physical mail converts again precisely because everyone stopped sending it. — the catalog boss | 🔴 print + address data | ❓ | 🅿️ |
| C8 | Nobody "cancels" — they become $9.99/mo "alumni", keeping their history, levels and streaks. — the club manager | 🟡 cancel-flow change (Nic) + your GO | 💰💰💰 churn is the binding constraint: 52 lost vs 38 won in the last 7 days | 🧪 strong — retention material |
| C9 | A $4.99 "paused" tier — the app goes read-only, your levels stay current, one tap resumes. For injuries, travel, busy seasons. — the club manager | 🟡 | 💰💰 catches cancels the $9.99 rail is too blunt for | 🅿️ |
| C10 | A $29 "reactivation fee"… that is always waived. — the hotel manager | — | — | ⛔ a fee that exists only to be waived is a fake claim — banned class |
| C11 | "Your assessment placement expires in 14 days." — the hotel manager | — | — | ⛔ not mechanically true — unenforced claims are banned |
| C12 | "Your program slot is released to a waitlist on Friday — reply KEEP." — the hotel manager | — | — | ⛔ same law — unless the scarcity is made mechanically real |
| C13 | "Your first workout has been rebuilt and gone unused 213 times" — the waste made personal and visible. — the hotel manager | 🟢 | 💰 a striking angle | 🧪 candidate copy for the dormant-pool test |
D · Members recruit members (9 ideas)
| # | The idea, in plain words | Effort | Money | Where it stands |
|---|---|---|---|---|
| D1 | The shareable Rank Card — "Pull section · Level 7 of 19 · top 22%." A rank invites one question: what would I be? — and the only answer is the quiz. — the band director | 🟢 | ❓→💰💰 if people share it, the quiz's top grows for $0 | ⭐ M3 |
| D2 | Three named guest passes, granted at the emotional peak — the moment you unlock your first skill. — the photographer | 🟡 | ❓ base is ~45 buyers/mo — chains need volume first | 🅿️ |
| D3 | Every paying member gets 3 × 30-day guest passes a year, each with the guest's own assessment. — the club manager | 🟡 | ❓ | 🅿️ |
| D4 | A gift registry — partner, parents and friends fund your lifetime membership for a birthday or milestone. — the photographer | 🟡 | 💰 | 🅿️ |
| D5 | Gift memberships: "give a year of strength — $97", delivered Christmas morning to a different name. Cash now + a new member at $0 acquisition. — the catalog boss | 🟡 Nic — a simple gift flow | 💰💰 | 🧪 Q4 |
| D6 | Bring-a-rookie: your recruit's 3 free workouts sit inside your training group and count toward its standing. — the band director | 🔴 app | ❓ | 🅿️ |
| D7 | Level-matched 7-day duels — challenge a friend; the invite ends "get placed first so it's fair" (the quiz link). — the band director | 🔴 app | ❓ | 🅿️ |
| D8 | Verified skill badges people paste into dating profiles — "Front Lever: Unlocked · Movement Age: 26." — the market-stall owner | 🟢 an image | ❓ | 🅿️ (yes, really — parked, not killed) |
| D9 | "Host a park session" kits — a generated group workout + a QR to the quiz, for members who train outdoors. Calisthenics is performed in public; every park session is a shop window. — the photographer | 🟡 light ops | 💰 | 🅿️ |
E · Other people's audiences — borrowed distribution (16 ideas — only warm, high-intent traffic has ever converted here)
| # | The idea, in plain words | Effort | Money | Where it stands |
|---|---|---|---|---|
| E1 | A rev-share QR card inside every pull-up-bar and rings box — the person unboxing equipment is at peak intent, and the seller has no post-sale relationship to protect. — the market-stall owner | 🟢 outreach | 💰💰 slow-building standing channel | ⭐ M5 |
| E2 | Physios hand discharged patients a co-branded assessment link instead of a photocopied sheet — Jesse's rotator-cuff story is the sales letter. — the market-stall owner | 🟢 outreach | 💰💰 | ⭐ M5 |
| E3 | The wider "vendor web": chiropractors, climbing-gym coaches, personal trainers — each with a referral link and a courtesy account. — the photographer | 🟢 | 💰💰 | ⭐ inside M5 |
| E4 | White-label the quiz for calisthenics YouTubers — "The [Creator] Level Test", their name on the results, rev-share. Creators with 10× our views and no product get a finished funnel overnight. — the market-stall owner | 🟡 outreach + a build per creator | 💰💰 | 🧪 top of the partnerships queue |
| E5 | Let partners give away OUR free guides as THEIR lead magnets — co-branded, with the quiz baked into chapter one. — the market-stall owner | 🟢 | 💰 the handstand guide already turned a stranger into $97 in 14 minutes | 🧪 |
| E6 | Swap dedicated email sends 1-for-1 with similar-sized fitness newsletters — rent warm lists for £0 instead of buying cold clicks. — the market-stall owner | 🟢 | 💰💰 | 🧪 |
| E7 | Our quiz on other fitness apps' cancel screens; their app on ours — churners quit the modality, not fitness. — the market-stall owner | 🟡 | ❓ | 🅿️ |
| E8 | GLP-1 / weight-loss clinics prescribe TMA against muscle loss; the re-assessment doubles as the clinic's adherence report. — the market-stall owner | 🟡 long sales cycle | 💰💰 slow | 🅿️ |
| E9 | Get listed where employers pay (Wellhub, Forma, lifestyle-spending accounts) — the employee's price becomes $0 without discounting a penny. — the market-stall owner | 🟡 application processes | 💰💰 | 🅿️ |
| E10 | Weatherproof QR plaques on public calisthenics rigs, reached through the equipment manufacturers who install them. — the market-stall owner | 🔴 ops | ❓ | 🅿️ |
| E11 | A QR appendix in the classic calisthenics books — "the book tells you why; this places you where you are." — the market-stall owner | 🟡 | 💰 slow | 🅿️ |
| E12 | Consign lifetime-$297 to deal marketplaces (AppSumo-style) — permanently-warm deal-hunter lists with built-in deadlines. — the market-stall owner | 🟡 | 💰💰 but ~70% marketplace cut + price exposure | 🅿️ |
| E13 | Sponsor a Twitch self-improvement arc — the streamer assesses live, chat follows the skill tree. — the market-stall owner | 🟡 | ❓ mechanism unproven | 🅿️ |
| E14 | QR stickers in hotel gyms — a captive, bored, equipment-free moment. — the market-stall owner | 🟢 | ❓ mechanism unproven | 🅿️ |
| E15 | Sell placement in our emails to other brands ("package inserts") — the catalog trade's dirty secret. — the catalog boss | — | — | ⛔ mostly demoted — the list's trust IS the moat; survives only as reciprocal barter (see E6) |
| E16 | Paid ads sell the free assessment as the desirable object — "get your movement age and your personal skill map" — never the $25/mo subscription. Peterman's ads sold the catalog, and the catalog closed. — the catalog boss | 🟢 a creative brief | 💰💰 this IS the $2-lead creative the plan already wants | 🧪 feeds NIC-97 directly |
F · The product does the marketing (15 ideas — mostly app builds; ~100k people currently train alone, in private, and never once perform)
| # | The idea, in plain words | Effort | Money | Where it stands |
|---|---|---|---|---|
| F1 | Small named "sections" by your weakest movement (Handstand section, Pull section) with a visible roll-call of ~8 peers — nobody quits mid-season because the section would notice. — the band director | 🔴 app | 💰💰 retention mechanics | 🅿️ product roadmap |
| F2 | A scheduled, filmed "performance date" — your current skill auto-cut against your day-1 assessment footage. — the band director | 🔴 app | ❓ every member becomes a before/after ad | 🅿️ |
| F3 | At milestones, prompt the member to perform the skill FOR someone — spouse, kid, coworker — and capture the reaction. The real audience is the member's household. — the band director | 🟡 a push/email prompt | 💰 | 🅿️ |
| F4 | Quarterly graded certificates — Division I/II/III per movement, dated and signed by the head coach. An external verdict outranks a self-tracked stat, and the cadence doubles as a renewal ritual. — the band director | 🟡/🔴 | 💰💰 | 🅿️ |
| F5 | A challengeable, named "first chair" at every level — film the standard, take the chair, the deposed gets notified. — the band director | 🔴 app | ❓ | 🅿️ |
| F6 | Monthly named classes — "Class of Aug '26, 112 strong" — with a one-year reunion re-assessment. — the band director | 🟡 | 💰 | 🅿️ |
| F7 | A weekly 20-minute livestream: Jesse conducts everyone through the SAME movement, each at their own level — one ensemble, and native YouTube content in the same 20 minutes. — the band director | 🟡 20 min/wk of Jesse | 💰💰 activation + content in one | 🅿️ founder-time |
| F8 | An annual qualify-to-attend live event — a season with a destination. — the band director | 🔴 ops | ❓ | 🅿️ |
| F9 | One city, one park, one Saturday: every local member performs at once, filmed as one show. — the band director | 🔴 ops | ❓ the most shareable footage we'd ever own | 🅿️ |
| F10 | Quiz results become public profile pages by default, linkable from all 869 videos. — the band director | — | — | ⛔ privacy — the opt-in version lives inside M3 |
| F11 | An earned shirt that states your rank ("PULL-UPS · LVL 12/19") — and quietly goes stale if you stop training. — the band director | 🔴 merch ops | 💰 | 🅿️ |
| F12 | "Senior Night" — auto-built tribute artifacts for members with 5–7 YEARS of logged data: day-1 vs today, every session, years spanned. The 2019 cohort finally does marketing work. — the band director | 🟢 the fleet can hand-build the first few | 💰 the deepest proof wall we own | 🧪 cheap and lovely |
| F13 | A 90-second embeddable mini-assessment — one movement, instant partial placement ("your push-up chair: Level 4-ish of 12") — planted on the 318 magnet pages, both websites, inside YouTube. A partial rank is an itch only the full quiz can scratch. — the band director | 🔴 a widget build (dev) | 💰💰💰 BWA's 1.9M impressions get a product taste on the page | 🧪 the big build — same thesis as the SEO fleet's "the assessment IS the product" |
| F14 | Charge $9 for the assessment, credited back at purchase — a paid sitting qualifies harder and always shows up. — the photographer | — | — | ⛔ for now — leads are the lifeblood (your ruling); this would strangle the top of the funnel |
| F15 | Auto-edited "reveal" films at each milestone — day-1 footage against the new skill, watermarked, built to be posted. — the photographer | 🔴 app | ❓ | 🅿️ (same family as F2) |
G · Wake the dormant pools + smarter email (12 ideas — 6,813 surveyed-never-trained · 2,805 trained-never-paid · the 83k list)
| # | The idea, in plain words | Effort | Money | Where it stands |
|---|---|---|---|---|
| G1 | A $9 one-shot month — no subscription, no card kept on file — for the 2,805 who trained but never trialled. They said no to a commitment, not to the product. — the hotel manager | 🟡 a product/price | 💰 | 🅿️ |
| G2 | A $5/mo "social membership" — community + monthly reassessment, no daily workouts — for the 6,813 who finished their survey and never trained. Full membership one tap away. — the club manager | 🔴 | 💰 | 🅿️ |
| G3 | Join at $197/yr, get $100 back as credit after 12 workouts — the member funds their own activation bounty against the 44% never-work-out rate. — the club manager | 🟡 pricing | ❓ | 🅿️ flincher |
| G4 | Age-priced doors: under-25 at $57/yr; a "Longevity membership" for 60+ at full price with its own framing. The quiz already captures age. — the club manager | 🟡 pricing | 💰 | 🅿️ |
| G5 | $1 first month, renews at full price — the Columbia House front door, for the 87% who see the price and walk. — the catalog boss | 🟡 pricing + churn math needs care | 💰💰❓ | 🅿️ pricing |
| G6 | Goal-matched email streams (the Handstand book, the Over-40 Comeback book) instead of one generic stream — the quiz already knows each person's goal. — the catalog boss | 🟢 email fleet | 💰💰 the magnet→quiz→$97-in-14-min sale proved the mechanism | 🧪 an email-fleet brief |
| G7 | Treat the 83k as recency/value cells — the "saw the price, didn't buy" cell gets re-mailed within 10 days; 2019 payers get different mail from 3-year dormants. — the catalog boss | 🟢 email fleet | 💰💰 lifts every send | 🧪 an email-fleet brief |
| G8 | A live touch at minute 15 for price-viewers who haven't paid — a one-time rate that dies at minute 30. Check-in is when you upsell the guest. — the hotel manager | 🟡 SMS infra | 💰💰 87 of 102 walk today | 🅿️ collides with your abandonment rulings — revisit with the recovery read |
| G9 | A 15-minute "placement consultation" calendar link for people who didn't buy within 24h — a human reads their results back to them. — the photographer | 🟡 whose calendar? | ❓ | 🅿️ |
| G10 | A framed print of YOUR OWN skill map as the annual-plan bonus — a gift instead of a discount; premiums historically out-pull equal-value discounts. — the catalog boss | 💰 | 🅿️ | |
| G11 | A printed year-one portfolio shipped to lifetime buyers — the album that sits on the coffee table and gets shown to every visitor for a decade. — the photographer | 🔴 print ops | 💰 a referral artifact in the house | 🅿️ |
| G12 | Lifetime becomes a transferable certificate — giftable, bequeathable, resellable once. A purchase becomes an asset. — the club manager | 🟡 legal/policy | ❓ | 🅿️ |
H · Aga's additions after reading the bench (hers, not the bench's; added 11 Aug)
| # | The idea, in plain words | Effort | Money | Where it stands |
|---|---|---|---|---|
| H1 | Reactivate the affiliate machine we already own. Tapfiliate tracking and the affiliate checkout family (normal/ — the one with the coupon box) already exist and net ≈£275/mo, mostly dormant [from the affiliate hub docs, not re-verified live today]. One affiliate-fleet cycle: wake the dead partners, and sign the bench's E-band targets — equipment sellers, physios, calisthenics YouTubers — as tracked affiliates on infrastructure we already pay for. The mavericks reinvented affiliate recruitment without knowing the program exists; this is the version that uses it. — Aga, 11 Aug | 🟢 affiliate + outreach fleets, through the outreach gate | 💰💰 commission-only — nothing is owed until a sale lands | 🧪 runnable now |
| H2 | Small paid budget behind the blog posts that measurably convert. Blogs already created 2 of the 15 quiz sales for £0, and the magnet strip on 318 pages produces ~336 free leads/mo — but the capture layer only went live 5–6 Aug, so the first clean list of winner pages lands ~19 Aug. Then: $5–10/day behind the top 2–3 pages, buyer countries only, judged on cost per real lead vs the $2–5 the quiz line already pays — a warm-up step for the cold traffic that never bought when asked directly ($444/purchase). Really a lander variant of the NIC-97 lead line, not a new budget. — Aga, 11 Aug | 🟡 inside the existing ad envelope | 💰💰 if a warmed lead comes in at or under the quiz line's cost | 🧪 gated on the 19 Aug winners read — buying before it is guessing |
| H3 | The old "really cheap leads" — the muscle-guide Facebook forms, dug up and measured. Aga remembered cheap leads from a muscle-building-guide lead form; here is what the account actually shows (all pulled live 11 Aug). The campaign (MOF - Lead Form, running since Nov 2025, retargeting last-28-day site visitors with the "Build Muscle With 40+" PDF): £1,939 → 2,905 leads = £0.67 each — the cheapness is real. But three things travel with it: ① geography — buyer-country leads cost £1.08; the £0.56 majority is Syria (£0.06), Algeria (£0.03), India (£0.08) — the same never-buying-geo class as the 6,700-install lesson; ② 🔴 the pipe has been broken since 1 Dec 2025 — the ActiveCampaign sync died 3 weeks in, so ~1,100 leads since never reached email, Meta wipes stored leads after 90 days (~2,100 gone forever, ~264 still rescuable today — AGA-150), and the campaign still spends ~£78/mo into the dead pipe; ③ quality of the 250 that DID reach email: exactly 1 ever paid (0.4%), 9% went on to the quiz or app — though that cohort was mailed by the old estate that never asked for money, so the 0.4% partly indicts the no-ask doctrine, not only the leads. The honest re-test, in order: rescue the 264 · fix the sync · geo-lock to buyer countries · THEN judge £1.08-class form leads against the $2–5 quiz lead, now that 993 SALES SEQUENCE asks. A £1.08 lead that clears quality would meet Aga's own "$1–2 and I'll take them" bar. — Aga's question, measured 11 Aug · Meta Marketing API + TMA ActiveCampaign, ever-paid join n=16,079 | 🟢 rescue + sync are UI-quick (AGA-150); re-test rides the existing envelope | 💰💰 IF quality holds at £1.08 — unproven until the 993-era re-test | 🧪 gated: fix the pipe BEFORE any more spend |
🃏 "The £0.67 is a denominator illusion" (the catalog boss) — a name's price means nothing without its response rate. In the only unit that matters, cost-per-quiz-progressor, these names run ≈£12 (£1.08 ÷ the 9% who progressed) against $2–5 to buy a quiz entrant directly. Aga's bar has two clauses — "$1–2" is met; "high-quality" has never been tested.
🃏 "The 0.4% payer rate convicts nobody" — those 250 names entered an estate that never asked for money; their non-buying partly indicts the retired no-ask doctrine, not the lead. The honest test only became possible on 10 Aug when the estate learned to ask — so run it, don't remember it.
🃏 "Rescue outranks judgment" (the cataloger's nightmare: paying to mint names, then burning the file) — ~2,100 leads are already gone forever; ~264 perish on Meta's 90-day clock while ~£78/mo keeps buying fruit that rots in a dead pipe. Whatever the quality verdict, pulling the rescuable file and fixing-or-killing the pipe comes first — a no-regret ops move.
🃏 "The 2,300 cheap-geo names are a different SKU, not garbage" (the hotel manager: distressed inventory wants a fenced rate) — never mail them with the main file (they drag sender reputation), but held in their own cell they cost ~nothing and are pre-banked Black-Friday seats — 70%-off yearly ≈ $47 may clear in some of those markets. The fuller fenced-regional-price play is an Aga-only pricing door.
🃏 "Autofill strips the exact mechanism that makes the quiz close" (the photographer + band director) — the 13%-in-14-minutes close is built on 14 minutes of self-qualifying effort; an instant form is a guest signing someone else's guest book. Form leads are raw material for the TOP of the quiz, never substitutes for quiz leads — the bridge to test: "your guide is chapter 0 — now find your exact level."
🃏 The Challenger's angle — if buyer-geo £1.08 survives the test, its natural home is not a standing drip audience; it is the retargeting layer of the next dated event (the only context in which Meta has ever produced a purchase), minted cheap in the weeks before the event fires.
Cheapest test: rescue the ~264 before deletion → their own AC tag → the guide→quiz bridge into the asking estate, buyer-geo vs cheap-geo cells split — read cost-per-price-viewer and cost-per-payer in 14 days, through
segment_guard.py + the deliverability watch. £0 new spend. Downside cap: a few hundred emails to people who opted in for the guide; worst case the leads are confirmed lookers and the £78/mo gets killed on evidence instead of memory — which is itself worth the test.[Honesty flag, curator-style: the bench says "rescue via the Meta API" — that path is 403'd for our credentials (no
leads_retrieval permission), so the rescue is the 10-minute UI export + integration reconnect, filed as AGA-150.]
Where all of this lives permanently: these tables, plus the ideas ledger EXPERT_PROMPTS/STRATEGY/MAVERICK_IDEAS_LEDGER.md — every 🅿️ idea sits there with a revisit date (11 Sep – 11 Nov 2026) so nothing dies quietly, and 4 of the strongest mechanisms were added to the pattern library so every future run inherits them. The ⛔ rows are kept on purpose: they are the ideas that fail our honesty rules (no fake fees, no unenforced claims, no selling the list's trust), and writing down WHY we said no stops a future session from re-proposing them.
0b · The new quiz vs the past quizzes — every sale, and whether paid ads made any
Scope, set by Aga: this is about the new quiz that started selling on 1 August, measured against the quizzes that came before it. Everything here is quiz revenue only — TMA's other money (the app, renewals, the July lifetime promo, direct /register/ links) is excluded from the comparison and shown separately at the end, labelled as such. CHECKED — tools/finance/quiz_sales_provenance.py + GA4 per-version pull, both run 10 Aug
🏁 The head-to-head — same 10 days, same measurement, all four builds
1–10 August 2026, the new quiz's first 10 selling days. Visitors are real people with the YouTube crawler removed. Sales are Stripe's, not GA4's.
THE NEW QUIZ2026-quiz-checkout-v1$97 today, no trial | THE PAST QUIZZES — all sell a 7-day free trial | |||
|---|---|---|---|---|
/2026/Control | 2026-v2-2 | 2026-v2-3 | ||
| Real visitors (crawler removed) | 561 | 143 | 71 | 9 |
| Gave us their email | 112 (20.0%) | 35 (24.5%) | 11 (15.5%) | 4 |
| Reached the price | 102 (18.2%) | 36 (25.2%) | 11 (15.5%) | 5 |
| Started a free trial | — no trial exists | 1 | 2 | 0 |
| PAID SALES | 15 | 0 | 0 | 0 |
| CASH TAKEN | $1,298.46 | $0 | $0 | $0 |
| Cash per real visitor | $2.31 | $0.00 so far — best case $0.82 if both pending trials convert in full | ||
In its first 11 selling days the new quiz took $1,298.46. The three older quizzes, over the same 10 days and with 223 real visitors between them, took nothing.
So we checked what is actually in flight, rather than assuming. Across all three old builds, those 223 visitors started 3 trials in total — one already cancelled, two still running (a $24.97/month ending 13 Aug and a $157/year ending 17 Aug). Even if both convert at full price, the past quizzes will have earned $181.97 from those 10 days — $0.82 per visitor against the new quiz's $2.31. The gap narrows; it does not close or reverse. CHECKED — every subscription created 1–10 Aug on the legacy rail, read individually
🔑 And the reason this comparison is possible at all is the same reason the rest of the page is full of "not knowable": the new quiz stamps its own name onto every sale. The old ones do not — so where their sales can be counted, it is only because there are three subscriptions to look at by hand. At any real volume they go dark again.
🔴 Read this first — what "where it came from" can and cannot mean
A sale only tells us where it came from if the visitor arrived carrying a tag, and that tag survived all the way to the payment. Three honest limits, each of which changes how you read the tables below:
- No tag ≠ "direct". When a sale carries nothing, the correct answer is UNKNOWN. It may well have come from an ad whose tag was stripped on the way. We never fill that gap with a guess.
- 🔴
utm_source=tmais OUR OWN internal link tagging, not a source. When someone clicks a button on our homepage, that tag overwrites whatever brought them to the site in the first place. So "TMA homepage" means "the last thing they clicked was ours" — it does not mean the homepage found them. This is the single easiest number on this page to misread. - A tag proves a click, not a cause. Someone who saw a Facebook ad on Monday and searched our name on Google on Friday shows up as Google. Neither platform's own conversion count is used as truth here.
The answer, in one line
Paid ads created 3 of the 14 sales on the new checkout — all three from Google, and all three from people typing our own brand name into the search box.
Meta created none. Not one August sale on any funnel carries a Facebook or Instagram click ID, and Meta's own reporting agrees — it counts 0 purchases across all three of its live campaigns for 1–10 Aug. Two independent sources, same answer, which is as strong as this gets. $286.90 of Meta spend in those 10 days produced 206 leads and no traceable sale.
Every one of the 15 sales on 2026-quiz-checkout-v1, in order 15th added 11 Aug — the magnet-born buyer
The full list, nothing omitted. "Last click" is the tag stored on the payment itself.
| When | Paid | Plan | Last click before paying | Persona | Is this paid ads? |
|---|---|---|---|---|---|
| 1 Aug 14:06 | $97.00 | annual | no tag — UNKNOWN | MUSCLE_BUILDER | cannot say |
| 1 Aug 19:32 | $97.00 | annual | Google Ads · term "movement athlete" | AGE_STRONG | ✅ YES — brand search |
| 2 Aug 14:56 | $12.48 | monthly 50% | our own site · nav link | STRENGTH_BUILDER | no |
| 3 Aug 04:55 | $97.00 | annual | our own site · nav link | JUST_START | no |
| 4 Aug 08:43 | $97.00 | annual | no tag — UNKNOWN | AGE_STRONG | cannot say |
| 6 Aug 08:21 | $97.00 | annual | BWTA blog · assessment landing page | MUSCLE_BUILDER | no |
| 6 Aug 20:05 | $97.00 | annual | TMA homepage · hero | FAT_BURNER | no |
| 7 Aug 11:33 | $97.00 | annual | no tag — UNKNOWN | JUST_START | cannot say |
| 8 Aug 11:40 new | $97.00 | annual | TMA homepage · hero | AGE_STRONG | no |
| 8 Aug 16:34 new | $97.00 | annual | TMA blog · the-best-calisthenics-program | SKILL_SEEKER | no |
| 9 Aug 08:22 new | $97.00 | annual | TMA homepage · hero | MUSCLE_BUILDER | no |
| 9 Aug 21:18 new | $97.00 | annual | TMA homepage · hero | JUST_START | no |
| 10 Aug 11:16 new | $97.00 | annual | Google Ads · term "the movement athlete" | STRENGTH_BUILDER | ✅ YES — brand search |
| 10 Aug 16:08 new | $24.98 | quarterly 50% | Google Ads · term "movement athlete" | FAT_BURNER | ✅ YES — brand search |
| 10 Aug 20:40 new | $97.00 | annual | 🔑 THE LEAD-MAGNET CHAIN — opted into the Handstand free guide at 20:26, landed /resource-thank-you/, clicked its hero CTA into the quiz, bought 14 minutes after the opt-in. Confirmed independently by Stripe UTMs and AC list membership (contact 378188). The ONE sale whose pre-quiz origin we truly know — the magnet list recorded what our internal utm_source=tma tag erased. | SKILL_SEEKER | no |
| 15 sales · $1,298.46 | avg $86.56 · 13 × $97 · all 15 still active · zero refunds | ||||
| Grouped by where the money came from | Sales | Cash | Share of cash | What it actually tells us |
|---|---|---|---|---|
| Our own website — homepage hero ×4, nav ×2 | 6 | $497.48 | 38% | 🔴 The last click was ours — that is all it proves. Something else brought them to the site first; this tag overwrote it. |
🔑 The lead-magnet page — /resource-thank-you/ hero Handstand guide opt-in → quiz → paid in 14 min | 1 | $97.00 | 7% | The one sale with a fully-known pre-quiz origin — the magnet lane sells. |
| Google Ads — brand terms | 3 | $218.98 | 17% | ✅ The only provable paid sales. They searched for us by name. |
| Our own blogs — TMA blog ×1, BWTA blog ×1 | 2 | $194.00 | 15% | Genuine owned-content sales. BWTA's assessment landing page earned one. |
| No tag at all | 3 | $291.00 | 22% | UNKNOWN, and it must stay unknown. Could include a paid click whose tag was stripped. |
| Meta (Facebook / Instagram) | 0 | $0 | 0% | 🔴 Zero. Confirmed twice — no fbclid on any sale, and Meta's own pixel reports 0 purchases. |
| 0 | $0 | 0% | 🔴 Zero, and consistent with §14: email sent the quiz almost nothing all week. |
🔬 The deep join — every buyer's FULL history, Stripe × ActiveCampaign all 15 contacts read individually in the TMA AC account (bwta), 10 Aug evening
Aga's ask, after contact 378188's chain was reconstructed: do the same for every sale. Each buyer's email was joined to their AC contact — when the lead was born, what lists and tags they carried before buying, and how long from first contact to paid.
| Sale | Who · AC contact | Lead born (AC) | Lead → paid | Prior history before the buying session |
|---|---|---|---|---|
| 1 Aug · $97 | akrifta@… · 377739 | same session | ~15 min | none — quiz-born |
| 1 Aug · $97 Google brand | brian_totten@… · 377756 | same session | ~6 min | none — quiz-born |
| 2 Aug · $12.48 | ebelnikola@… · 377787 | same session | ~16 min | none — quiz-born |
| 3 Aug · $97 | averyarh@… · 377808 | same session | ~14 min | none — quiz-born |
| 4 Aug · $97 was "no tag" | andrew@2purple.net · 191131 | 28 Dec 2022 | 3.6 YEARS | 🔑 A CANCELLED CUSTOMER, WON BACK. Lists: Current Customers + Customer Cancelled. Tag: Nurture2 Completed — he finished the very nurture pool (758 Nurture Sequence 2) this page proposes to broadcast to, then returned through the quiz and paid $97. The warm pool demonstrably contains buyers. |
| 6 Aug · $97 bwa/blog | unwantedalmondjoy@… · 377982 | same session | ~16 min | none — quiz-born |
| 6 Aug · $97 | hedvallerika@… · 377918 | 4 Aug | ~2 days | the one considered buyer — led on the 4th, returned via the homepage hero and paid on the 6th |
| 7 Aug · $97 | evekandiyoti@… · 378030 | same session | ~11 min | none — quiz-born (first buyer carrying the full NEW quiz-* tag chain — the signals went live 7 Aug) |
| 8 Aug · $97 | kieransmurphy@… · 378078 | same session | ~11 min | none — quiz-born |
| 8 Aug · $97 tma blog | alexpuckett501@… · 378087 | same session | ~20 min | none — quiz-born |
| 9 Aug · $97 | larsgschmidt@… · 378109 | same session | ~14 min | none — quiz-born |
| 9 Aug · $97 | cl@allxcl.com · 378142 | same session | ~4 min | none — quiz-born |
| 10 Aug · $97 Google brand | erik.swartzwelder@… · 378165 | same session | ~4 min | none — quiz-born |
| 10 Aug · $24.98 Google brand | jstouttx@… · 378176 | same session | ~3 min | none — quiz-born |
| 10 Aug · $97 sale #15 | ingridacherney@… · 378188 | same session | 14 min | 🔑 LEAD-MAGNET-BORN: opted into TMA LM [FREE GUIDE] HANDSTAND, hit /resource-thank-you/, clicked the hero CTA into the quiz, paid $97 — magnet → paid in 14 minutes, confirmed independently by Stripe UTMs and AC list membership |
① The quiz is a SAME-SESSION conversion machine. 13 of 15 buyers paid within ~20 minutes of first giving us their email — median ≈ 14 minutes. They did not need nurturing; they needed the price screen.
② The two exceptions are each worth a strategy. One buyer considered for 2 days and returned through the homepage. And one — a 2022 customer who cancelled, completed
758 Nurture Sequence 2, and came back 3.6 years later — is the live proof that the warm-pool broadcast (lever #1 of the fleet plan) is aimed at people who really do buy.③ The magnet lane works end-to-end (sale #15: free guide → quiz → $97 in 14 minutes) — and its acquisition source was erased by our own internal
utm_source=tma tag, which AC compounds by storing the literal string "none" in its own UTM fields (AGA-73, now with revenue attached).④ No other buyer carries ANY prior email or magnet history — so email's contribution to these 15 was one warm return + one magnet hand-off, consistent with §NP: the everyday email ask only began today.
🔴 The paid-ads verdict — what £359 of Google and $287 of Meta bought in 10 days
Google figures from the Google Ads API in the account's own currency (GBP); Meta from the Marketing API (USD). Sales from Stripe. 1–10 Aug. CHECKED
| Campaign | Spend | Clicks | Sales it can be shown to have made | Cost per real sale |
|---|---|---|---|---|
| Google · Quiz Funnel | Branded | US keywords: "movement athlete", "the movement athlete", "themovementathlete" | £16.27 | 12 | 3 · $218.98 | £5.42 |
| Google · Competitors Branded | US (mostly "betterme") | £113.15 | 65 | 0 | — |
| Google · Non-Branded | US | £91.29 | 24 | 0 | — |
| Google · Competitive Non-Branded | US — the one flagged on 7 Aug | £89.89 | 18 | 0 | 🔴 still zero |
| Google · PMax | US | £48.51 | 110 | 0 (the account claims 3 — see below) | — |
| Meta · ADV+ OFF | MC | Control | $142.30 | 249 | 0 | — |
| Meta · ADV+ ON | US | Variant 2 | $75.46 | 143 | 0 | — |
| Meta · MOF — Lead Form | $69.14 | 960 | 0 | — |
| TOTAL | ≈£359 + $287 ≈$770 combined | 1,581 | 3 sales · $218.98 | 🔴 ≈$257 per sale |
Read the top row against every row beneath it. £16.27 on our own brand name bought all three sales. The other £343 of Google and $287 of Meta — 1,569 clicks — bought no sale we can trace at all.
And one discrepancy, stated rather than smoothed: Google's account reports 7 purchase conversions for 1–10 Aug where Stripe evidences 3. The 3 it credits to Branded match Stripe's 3 exactly — a clean agreement. The other 4 (3 on PMax, 1 on Branded) have no matching Stripe sale carrying a Google click ID. Most likely view-through or cross-device credit, or the codeless "any page ending /success" trigger catching a purchase Google did not cause. Treat PMax as unproven, not as proven.
⚠️ And what Google's "82 conversions" actually are — they are mostly not sales
The Ads account reports 82 biddable conversions for 1–10 Aug, which reads like a healthy result. Split by what was actually counted: CHECKED — Google Ads API, conversion_action breakdown
| What the account counted | Count | What it really is |
|---|---|---|
| Leads — July 2026 — Google Tag Manager + generate_lead + Submit lead form | 77 | Someone gave an email address. Not a sale. |
| Purchase (July 2026) (All Success Pages) + Lifetime $297 Promo — Purchase | 7 | Claimed purchases — of which 3 are corroborated in Stripe. |
| Android installs / first_open / session_start | 8 | App events, unrelated to the quiz funnel. |
So the campaigns are optimising toward leads, not buyers. That is a legitimate strategy — but it means every "cost per conversion" figure in the Ads UI is a cost per email address, and reading it as cost per sale overstates performance by roughly 11×. Worth pairing with the separate finding of 10 Aug that the account also carries ADD_TO_CART and BEGIN_CHECKOUT as biddable goals — see REMARKETING_PLAYBOOK_2026-08-10.html.
Context — everything TMA sold, 8 May → 10 Aug NOT the quiz comparison
/register/ links, email-driven sales, everything. Most of it has nothing to do with the quiz. It is here for one reason only: to show how much of TMA's money carries no source at all, which is the constraint every ad decision runs into.224 new sales · $28,610.36 (renewals on subscriptions created before the window excluded — 322 of them, $14,451.11).
| Evidence carried on the sale | Sales | Cash | Detail |
|---|---|---|---|
| A paid-ad click ID | 9 | $2,000.98 | 6 Meta (all $297 lifetime, 5–10 Jul — the July promo) + 3 Google (all August, all the $97 quiz checkout, all brand terms) |
| An email-campaign tag | 20 | $5,940.00 | All ActiveCampaign, all the 4th of July LIFETIME SALE arc, all $297 |
| An internal (our-own-site) tag | 8 | $691.48 | 8 of the 14 v1 sales above — real source masked by our own tagging. (That this equals the 7 Aug v1 total is coincidence, not the same set.) |
| Nothing at all | 187 | $19,977.90 | 🔴 70% of the WINDOW'S money is untraceable — but 🟢 this is the May–July legacy era, not the current state: 100% of August-created sales are stamped (Aga's correction, 10 Aug PM). The old /register/ checkout stored nothing. |
① In three months, paid advertising can be shown to have produced 9 sales. Six of those were the July $297 lifetime promo on Meta — a promo, not the everyday funnel. In normal (non-promo) trading, paid ads have produced 3 traceable sales, all in the last 10 days, all on our own brand name.
② The 70%-untraceable figure belongs to the May–July window, NOT to today — ✅ corrected 10 Aug after Aga challenged it, and she was right: fresh Stripe pull shows 100% of August-created sales stamped; the dark residue is old renewals ($1,310/10d) and July trials ($312/10d), which nothing can retro-tag. What the legacy era still costs us: "paid ads made no PRE-AUGUST sales" must stay stated as "we cannot show that they did". For August onward, we CAN show it — and did, sale by sale, above.
BQ · Beyond the quiz — every other WEB sale since 1 August audited fresh from Stripe, 10 Aug PM · web only, no App Store / Play / RevenueCat
Question to answer: are we getting any other sales on the web — beyond the quiz — since August 1st, and can we analyse them? Both Stripe rails audited charge by charge. The ebook's ThriveCart checkout is a separate money rail invisible to Stripe and is out of scope here.
The headline: NO. Since 1 August, the quiz is the only thing making a NEW sale on the web. Every one of the 15 new-subscription sales this month carries checkout_version=quiz_checkout. The v3 checkouts have minted trials but taken no August money; there are zero one-off charges; the ThriveCart Stripe rail is empty.
| All web money taken 1–10 Aug | Charges | Cash | Share | What it actually is |
|---|---|---|---|---|
| NEW sales — all quiz | 15 | $1,298.46 | 44.5% | 13 × $97 annual + $24.98 + $12.48 (the 50% downsell). A 15th sale landed 10 Aug PM — the count on this page was 14 at the morning pull. |
| Renewals of pre-July subscriptions | 31 | $1,310.49 | 44.9% | 19 monthly · 7 quarterly · 5 annual — from subscriptions born 2019 ×2, 2022, 2023 ×2, 2024 ×7, 2025 ×4, 2026 ×15. The retention business. |
| July free-trials maturing | 7 | $311.84 | 10.7% | Mostly small monthlies + one $157 annual — the old funnel's tail, paying out. |
| TOTAL web | 53 | $2,920.79 | 100% |
The NON-quiz sales, one by one — every other web dollar, 1–10 Aug all 38 listed · read individually from Stripe, 10 Aug evening · the mirror of the quiz table in §0b
Question to answer: what other sales have we made that are NOT from the quiz. The answer in one line first: none of these is a new sale. All 38 are payments from subscriptions born BEFORE August — the old funnel's trials paying out, and the renewal base. Their origin predates all tracking, so a sub's "source" is its vintage.
| Charged | Paid | Plan | Subscription born | Status today |
|---|---|---|---|---|
| JULY FREE-TRIALS PAYING OUT — 7 charges · $311.84 (the old funnel's tail; these were "trials" 2 weeks ago) | ||||
| 2 Aug 05:52 | $49.97 | quarterly | 26 Jul 2026 | active |
| 2 Aug 13:40 | $24.97 | monthly | 26 Jul 2026 | active |
| 2 Aug 18:58 | $157.00 | annual | 26 Jul 2026 | 🔴 will cancel at period end |
| 4 Aug 20:38 | $24.97 | monthly | 28 Jul 2026 | active |
| 5 Aug 22:31 | $24.97 | monthly | 29 Jul 2026 | active |
| 8 Aug 21:20 | $9.99 | monthly | 9 Jul 2026 | active |
| 10 Aug 16:12 | $19.97 | monthly | 10 Jul 2026 | active |
| RENEWALS OF THE EXISTING BASE — 31 charges · $1,310.49 (sorted oldest subscription first — the lifeblood evidence) | ||||
| 6 Aug 18:29 | $14.97 | monthly | 30 Mar 2019 — paying for 7+ years | active |
| 6 Aug 00:48 | $14.97 | monthly | 28 Sep 2019 | active |
| 10 Aug 14:03 | $157.00 | annual | 3 Aug 2022 | active |
| 3 Aug 15:50 | $49.97 | quarterly | 26 Apr 2023 | active |
| 5 Aug 10:58 | $24.97 | monthly | 29 Aug 2023 | active |
| 3 Aug 19:03 | $49.97 | quarterly | 26 Apr 2024 | active |
| 5 Aug 01:00 | $9.99 | monthly | 5 Jul 2024 | active |
| 6 Aug 19:28 | $9.99 | monthly | 6 Jul 2024 | active |
| 2 Aug 00:18 | $157.00 | annual | 1 Aug 2024 | active |
| 8 Aug 20:13 | $49.99 | quarterly | 8 Aug 2024 | active |
| 5 Aug 22:32 | $49.97 | quarterly | 14 Oct 2024 | active |
| 5 Aug 12:15 | $49.97 | quarterly | 29 Oct 2024 | active |
| 10 Aug 19:04 | $9.99 | monthly | 10 Jul 2025 | active |
| 5 Aug 17:49 | $159.99 | annual | 5 Aug 2025 | active |
| 1 Aug 20:24 | $9.99 | monthly | 29 Oct 2025 | active |
| 2 Aug 21:27 | $24.97 | monthly | 25 Nov 2025 | active |
| 6 Aug 22:11 | $9.99 | monthly | 20 Jan 2026 | active |
| 1 Aug 06:03 | $24.97 | monthly | 25 Jan 2026 | active |
| 3 Aug 18:09 | $49.97 | quarterly | 27 Jan 2026 | active |
| 9 Aug 18:25 | $24.97 | monthly | 2 Feb 2026 | active |
| 6 Aug 01:39 | $9.99 | monthly | 5 Feb 2026 | active |
| 8 Aug 11:26 | $24.97 | monthly | 1 Mar 2026 | active |
| 6 Aug 18:09 | $24.97 | monthly | 30 Mar 2026 | active |
| 3 Aug 17:16 | $9.99 | monthly | 4 Apr 2026 | active |
| 8 Aug 12:33 | $9.99 | monthly | 15 Apr 2026 | active |
| 7 Aug 01:24 | $49.97 | quarterly | 30 Apr 2026 | active |
| 8 Aug 16:18 | $24.97 | monthly | 1 May 2026 | active |
| 8 Aug 17:18 | $19.97 | monthly | 8 May 2026 | active |
| 2 Aug 16:20 | $24.97 | monthly | 26 May 2026 | active |
| 9 Aug 20:55 | $132.13 | annual | 9 Jun 2026 | active |
| 9 Aug 18:55 | $24.97 | annual | 9 Jun 2026 | active |
| TOTAL — 38 charges | $1,622.33 | 37 of 38 payers still active · 0 renewals set to cancel · 1 July-trial will cancel | ||
WHERE the 38 originally came from — the origin signals Stripe still holds price family · product · coupon · trial-vs-direct, read per subscription, 10 Aug evening
These subscriptions predate UTM tracking, but three signals survive on every one: the PRICE it sits on (each price belongs to a specific checkout family), any COUPON on its first invoice (coupons identify promo emails), and whether it began as a trial. Grouped:
| Origin family | Payers | Cash/10d | The evidence |
|---|---|---|---|
| The $9.99 "Special Offer" rail — the win-back / downsell offer | 8 | $79.92 | All sit on price nine-ninetynine, product "$9.99/mo Special Offer" — the cancel-save / /30trial/-family offer reached through email and the cancellation flow. Born Jul 2024 → Jul 2026. The win-back machine's paying alumni. |
| Standard plans started as 7-DAY TRIALS at full price — the old quiz/site trial funnel | 21 | $1,164.40 | Prices Monthly V3 $24.97 · Quarterly V2 $49.97 · Yearly V3 $157, all trial-start, no coupon — the classic funnel this page's §2 calls "the past quizzes", working as designed across 2022–Jul 2026 vintages. |
| 50%-OFF COUPON buyers — promo emails | 5 | $384.04 | Coupons "promo", "first invoice 50% off" on standard plans (first payments $78.50 · $9.98 · $24.98 · $12.48 · $9.98) — born 1 Aug 2024, 8–10 May 2026 and 9 Jun 2026: the discount-email instrument's long tail, still paying full price now. |
| "Welcome Offer 50%" — the app's post-signup upgrade offer | 2 | $209.98 | Prices quarterly-welcome-50 / yearly-welcome-50 with the Welcome Offer 50% Coupon — the in-app welcome-expiring rail (the same family as the Play store's *_welcome_expiring SKUs). |
| 2019 legacy pricing | 2 | $29.94 | Price "Monthly (7day trial)" $14.97 — the original offer. Both born 2019, both still paying, 7 years on. |
① There are NO hidden non-quiz sales. Every non-quiz dollar is a subscription born before August paying again — nothing else is selling.
② The base is astonishingly loyal: 0 of 31 renewing subscriptions are set to cancel, and two have paid monthly since 2019.
③ This is the lifeblood table. These 38 people were once leads. Their $1,622/10d (≈$4,900/mo) is payout #4 of the lead's value — the one the CPL math never sees.
The trials, one by one — every subscription with a trial since 1 Aug, WITH where each came from 19 of 19 · Stripe × ActiveCampaign joined per person, re-run 11 Aug
Every trial's email was joined to its AC contact (the TMA account, bwta) — when the person first became known to us, what lists and tags they carried, and therefore which door they came through. Where the acquisition click (blog post? social?) is genuinely unrecoverable, the row says so instead of guessing.
| Created | Plan | Who | Where they came from — the journey AC + Stripe hold | Status today |
|---|---|---|---|---|
| 🔴 NOT PEOPLE — the phantom-mint cleanup (8 of the 10 "cancellations") | ||||
| 2–3 Aug 7 subs | 3 annual · 4 monthly | NO EMAIL AT ALL | ⚙️ Anonymous page-load mints — the v3 checkout minted a Stripe record with no customer email attached. Not visitors who left; records with nobody behind them. Bulk-cancelled in the 5 Aug 22:11 sweep. | cancelled (cleanup) |
| 3 Aug 22:21 | monthly | prod-defer-smoke+…@themovementathlete.com | ⚙️ Our own smoke test — an internal QA address, cancelled in its creation minute. | cancelled (test) |
| THE 2 GENUINE CANCELS | ||||
| 2 Aug 13:02 | monthly $24.97 | btw1194@… | AC 377782, born the same morning — same-day app signup, took the trial, cancelled 29 min later. Acquisition click: NOT KNOWABLE | 🔴 cancelled — real |
| 2 Aug 17:14 | monthly | 2002antokir44@… | AC 377624, born 30 Jul via the QUIZ (Stage-1 list + app registration) — a quiz lead who trialled 3 days later, cancelled 7 Aug. | 🔴 cancelled — real |
| ✅ THE LIVE TRIALS — 8 people (one holds two subscriptions twice) | ||||
| 6 Aug 08:43 + 08:54 | monthly ×2 (retry) | niall@… | AC 377984, born the same day 03:41 — same-day app signup → trial ~5h later. No quiz lists, no magnet lists. Acquisition click: NOT KNOWABLE | trialing → 13 Aug |
| 6 Aug 13:56 + 10 Aug 11:43 | monthly ×2 (retry) | brianlynch78@… | 🔑 AC 369367, born 7 JUNE 2026 — a TWO-MONTH-OLD lead (app signup, activated) who came back and took the trial in August. The considered-buyer lane, in the flesh. | trialing → 13/17 Aug |
| 7 Aug 23:40 | quarterly $49.97 | rslr.gehring@… | AC 378059, born the same day 18:33 — same-day app signup → trial ~5h later. click NOT KNOWABLE | trialing → 14 Aug |
| 8 Aug 23:40 | quarterly $49.97 | ericdauplaise@… | AC 378093, born the same day 18:37 — same-day signup → trial ~5h. click NOT KNOWABLE | trialing → 15 Aug |
| 10 Aug 09:36 + 09:39 | annual $157 ×2 (via /register/ AND v3!) | marco.chiaradia@… | AC 378162, born the same day 04:35 — same-day signup who hit both the legacy /register/ page (the first register stamp ever) and the v3 trial page minutes apart. | trialing → 17 Aug |
| 11 Aug 07:33 NEW overnight | monthly | achatnet43@… | 🔑 A QUIZ-CHECKOUT ABANDONER RECOVERED INTO A TRIAL. AC 378199: full NEW quiz-* chain + quiz-checkout-abandoned — she reached the $97 checkout, started the card, did not pay, and took the free trial instead. The abandonment → trial downsell path, working in the wild. | trialing → 18 Aug |
① The "10 cancellations" were 2. Seven were anonymous phantom mints with no email and one was our own smoke test. The v3 funnel's real record since the fix: 8 people trialling, 2 genuine cancels ever.
② The v3 trial door is an APP-SIGNUP door, not a quiz door — 5 of the 8 live trialists signed up and trialled the same day with no quiz history; one is a two-month-old June lead returning; and the newest is a quiz-checkout abandoner downselling herself to the trial.
③ "Blog post or social?" — honestly NOT KNOWABLE for these people: the v3 checkout writes no UTMs, AC's own source fields store the literal string
"none", and same-day signups leave no earlier trace. That is a one-line fix (the same UTM-flattening the quiz rail already has) — until it ships, per-trial acquisition cannot be answered, and this table says so rather than inventing it.💰 The price per lead — every campaign, both lenses 1–10 Aug · platform counts vs GA4 quiz-leads
| Campaign | Spend | Platform's own "leads" | Platform CPL | What that lead actually is |
|---|---|---|---|---|
| Meta · MOF — Lead Form | $71.91 | 99 | $0.73 | 🔴 An in-Facebook form-fill that never reaches the quiz — cheapest and worth the least |
| Meta · QLG Variant 2 (US) | $77.07 | 47 | $1.64 | A quiz email capture on the selling quiz ✅ |
| Meta · QLG MC Control | $146.80 | 73 | $2.01 | A quiz email capture on the Control quiz (no $97 page, but enters the email machine) |
| Google · PMax | £48.51 | 45 | £1.08 | GTM/GA4 lead events — Google-attributed |
| Google · Branded | £17.87 | 9 | £1.99 | Brand searchers — also the only 3 SALES |
| Google · Competitors Branded | £113.15 | 20 | £5.66 | "betterme" searchers |
| Google · Non-Branded | £91.29 | 2 | £45.65 | generic intent |
| Google · Competitive Non-Branded | £89.89 | 0 | ∞ | equipment shoppers |
| The honest blended lens | Spend | Leads | CPL |
|---|---|---|---|
| Meta — platform count (incl. MOF form-fills) | $295.78 | 219 | $1.35 |
| Meta — GA4 quiz leads only (the leads that can ever buy) | $295.78 | ~45 | ~$6.57 |
| Google — platform lead actions | £360.71 | 80 | £4.51 |
| Google — GA4 quiz leads only | £360.71 | ~25 | ~£14.43 |
① The quiz carries ALL new web revenue. There is no second funnel quietly selling — no email-driven
/register/ sales, no v3 money, nothing. Whatever we decide about paid and email, the quiz is the machine it feeds.② Renewals are 45% of the month's web cash — as big as all new sales combined. Subscriptions from as far back as 2019 are still paying. This is the quiet half of the business, and it is also the evidence for the lifetime-value argument in the next section: a lead bought years ago is still producing cash today.
PA · Paid advertising — the full picture every figure pulled live 10 Aug 2026
Questions to answer: where are we spending right now and what does it generate? And since Facebook shows no result, should we launch a version of the quiz that instead of having a paywall simply gets people to download the app and see their results." This section is the single source of truth for paid ads on this page. §15, §16, §17 and §18 keep their 7 Aug reasoning but their figures now point here. Sources: Google Ads API (account currency GBP) · Meta Marketing API (USD) · Google Play's own install and earnings reports · Stripe, both rails · GA4.
PA.1 · Where the money is going right now
Every campaign that spent in the last 10 days — what it is, who it targets and what it asks for, not just its name. CHECKED — campaign objects + ad-set targeting + keyword lists read from both APIs, 10 Aug
| Platform | Campaign | Status | Optimising for | 1–10 Aug spend |
|---|---|---|---|---|
| Meta | QLG | ADV+ OFF | MC | Control Cold acquisition. Multi-country, sends people to the /2026/ Control quiz. Running since May. | ACTIVE | LEADS | $142.30 |
| QLG | ADV+ ON | US | Variant 2 Cold acquisition. US-only. Was pointed at the old v2-2 quiz until 8 Aug, now at the quiz root. | ACTIVE | LEADS | $75.46 | |
| MOF — Lead Form Cold/mid-funnel. An in-Facebook lead form — the person never reaches our site. 0 purchases in 77 active days. | ACTIVE | LEADS | $69.14 | |
| Quiz Funnel | Competitors Branded | US Bids on rivals' brand names — mostly "betterme". £90 of it on that one word. | ENABLED | Sales obj. | £113.15 | |
| Quiz Funnel | Non-Branded | US Generic intent — "workout plan", "gym routine", "best workout programs". | ENABLED | Sales obj. | £91.29 | |
| Quiz Funnel | Competitive Non-Branded | US 🔴 Equipment shoppers — "street workout equipment", "calisthenics bar", "parallettes". We sell a programme, not equipment. Flagged 7 Aug, still running. | ENABLED | Sales obj. | £89.89 | |
| Quiz Funnel | PMax | US Google's automated placement — it chooses where to show. Most clicks, no verified sale. | ENABLED | Sales obj. | £48.51 | |
| Quiz Funnel | Branded | US People typing OUR name — "movement athlete", "the movement athlete". The only paid line that has produced a verified sale. | ENABLED | Sales obj. | £16.27 | |
| TOTAL, 10 days | $286.90 + £359.11 ≈ $770 · ≈ $77/day · ≈ $2,310/mo | |||
| Longer windows | Meta | |
|---|---|---|
| 30 days (11 Jul – 10 Aug) | $612.63 · 408 leads · 6 claimed purchases | £1,027.61 · 1,460 clicks · 128 "conversions" |
| 90 days (12 May – 10 Aug) | $6,415.92 · 1,482 leads · 45 claimed purchases · 5,526 app installs | £2,312.43 · 33,407 clicks · 4,929 "conversions" |
PA.2 · What it generates — the same 10 days counted three ways
The three columns disagree by two orders of magnitude, and each one is honestly reported by its own system. Only the right-hand column is money.
| Campaign · 1–10 Aug | Spend | What the platform calls a conversion | Leads GA4 | SALES Stripe-verified |
|---|---|---|---|---|
| Google · Branded brand-name searches | £16.27 | 12 (9 leads · 4 purch) | 3 | 3 · $218.98 |
| Google · Competitors Branded | £113.15 | 20 (all leads) | 10 | 0 |
| Google · PMax | £48.51 | 48 (45 leads · 3 purch) | 23 | 0 🔴 |
| Google · Non-Branded | £91.29 | 2 | 1 | 0 |
| Google · Competitive Non-Branded | £89.89 | 0 | 0 | 0 |
| Meta · all three live campaigns | $286.90 | 206 leads · 0 purchases | 206 | 0 |
| TOTAL | ≈$770 | 288 | 243 | 3 · $218.98 |
94% of what Google calls a "conversion" is an email address, not a sale. Of 82 biddable conversions in the window, 77 are lead actions and 7 are purchase-labelled — of which 3 are corroborated in Stripe. Every "cost per conversion" in the Ads UI is therefore a cost per email address, and reading it as cost per sale overstates performance by roughly 11×.
purchase: 0 on all three live campaigns for 1–10 Aug, and Stripe holds no August sale carrying a Facebook or Instagram click ID. Two independent systems, same answer: $286.90 bought 206 email addresses and no traceable sale.⚠️ The one number that does NOT reconcile: Google credits 3 purchases to PMax that have no matching Stripe sale with a Google click ID. Most likely view-through or cross-device credit, or the codeless "any page ending
/success" trigger catching a purchase PMax did not cause. Treat PMax as unproven, not as proven.PA.3 · Cost per real sale — and why one line flatters the whole account
| Spend | Real sales | Cost per real sale | vs $76 break-even | |
|---|---|---|---|---|
| Google · Branded only | £16.27 | 3 | ≈£5.42 | ✅ 14× inside |
| Everything else paid 4 Google + 3 Meta campaigns · 1,569 clicks | ≈$750 | 0 | ∞ | 🔴 no return |
| Blended | ≈$770 | 3 | ≈$257 | 🔴 3.4× over |
PA.4 · 🔴 What Meta has actually sold — and what those campaigns really were
The standard this table follows: no number without its context. Here is every Meta campaign that has sold anything in 90 days, what it was selling, to whom, and when. CHECKED — campaign + ad-set level, Meta Marketing API, 10 Aug
| Campaign | What it was actually selling | To whom | When it ran | Spend | Purch | Cost per purchase |
|---|---|---|---|---|---|---|
| Retargeting | July 4th 2026 Sale 7 campaigns | The $297 LIFETIME offer — one payment, forever | Warm retargeting. Ad sets read "All Users Ex Lifetime" and "All Users Ex All Plans" — people who already knew us and had not bought | 2–14 Jul (the lifetime sale window was 2–12 Jul) | $1,417.04 | 21 | $67.48 |
| Memorial Day 2026 Sale | The 50%-off sale | Retargeting | 28 May – 5 Jun (the sale ran 22–31 May) | $235.48 | 16 | $14.72 |
| QLG — quiz lead-gen 9 campaigns, incl. the 2 live today | Nothing. It asks for an email. The offer, if any, arrives later by email | Cold acquisition — new people who have never heard of us | ALWAYS ON — 98 active days | $3,555.26 | 8 | 🔴 $444.41 |
| MOF — Lead Form live today | Nothing. An in-Facebook lead form | Cold / mid-funnel | ALWAYS ON — 77 active days | $229.97 | 0 | 🔴 — |
Read down the "what it was selling" column and the pattern is the whole story: every purchase Meta has ever produced came from retargeting a warm audience with a hard-discount offer, during a sale that had an end date.
The real finding is the one underneath it: 🔴 TMA has never made COLD acquisition work on Meta. The always-on lead-gen has now run 98 days and $3,555 for 8 purchases — $444 each, against a break-even of about $76. That is not a campaign that needs optimising; on this evidence it is a channel that does not do this job.
Inside the July sale — which audiences actually bought, and which burned money
Same offer, same fortnight, same creative pairs. The only variable is who was targeted. CHECKED — ad-set level
| Audience (Nic's naming) | Creative | Spend | Purchases | Cost per purchase |
|---|---|---|---|---|
| WW · API | Video | $127.13 | 6 | $21.19 |
| WW · EXU | Video | $128.10 | 5 | $25.62 |
| WW · EXU | Image | $125.67 | 4 | $31.42 |
| NA-EU-AU-NZ · EXU "All Users Ex Lifetime" | — | $207.46 | 5 | $41.49 |
| WW · LLU · Hot lookalike | Video | $121.67 | 1 | $121.67 |
| WW · QUP | Video + Image | $218.71 | 0 | 🔴 — |
| WW · LLU · Warm + Cold lookalikes | Video + Image | $235.49 | 0 | 🔴 — |
| WW · API | Image | $136.75 | 0 | 🔴 — |
① Video beat image on the same audience, twice — API video 6 purchases vs API image 0; EXU video 5 vs image 4. Same offer, same people, same fortnight.
② The lookalikes and
QUP took $454 and returned nothing — that is 32% of the sale's budget on audiences that did not buy. The money was made by people who already knew us and had not yet paid.🔴 Honest limit:
EXU, API, QUP, LLU are Nic's own audience abbreviations and we have not seen them defined anywhere. The ad-set descriptions are quoted verbatim above; the campaign-level codes are reported as-found rather than decoded. Worth asking Nic once, so this table can be read without a glossary.🔴 And the offer that "worked best" on cost-per-purchase was the one that made the least money
reference_promo_calendar_permanent: the 50%-off instrument netted roughly +$688 across May and June combined, because June came in below the clean April baseline — the sale borrowed its own demand forward — and it permanently cut cash-per-sale while teaching the list to wait for discounts. The $297 lifetime added ≈$13,800 by comparison.The lesson for any future Meta test: judge the campaign on the money the OFFER made, never on the platform's cost-per-purchase. A cheap CPA on an offer that cannibalises full-price sales is the most expensive number on this page.
PA.5 · The app-download experiment we have already run — and what the store said about it
Aga's question assumes this is untested. It is not. Between 25 June and 7 July we spent about £1,800 across both platforms buying app installs, then switched it all off.
| What ran | Spend | Installs claimed | Cost per install |
|---|---|---|---|
Meta — 10 × App Install | … | Google Play | $972.29 | 5,526 | $0.176 |
Google — 10 × UAC | … (Play + iOS) | £1,096.01 | ~4,372 Android + ~857 iOS opens | ≈£0.21 |
| Meta's two iOS campaigns spent $20.02 and produced 0 installs. | |||
Those are the ad platforms' numbers. Here is Google Play's own install report — the store counting its own installs, which is the only source that can be trusted about install quality: CHECKED — tools/finance/play_install_reports.py
| Month | Play installs | Where they came from | From countries that have NEVER produced a Play buyer | Play buyers that month |
|---|---|---|---|---|
| June ads on from 25th | 4,109 | IN 1,632 · ZA 596 · PK 430 · PE 312 · US 270 · GB 164 | 69.0% | 14 |
| July ads on to the 7th | 5,221 | IN 1,734 · ZA 914 · PK 570 · PE 538 · US 279 · GB 199 | 67.2% | 23 |
| Aug 1–10 ads OFF | 48 | GB 8 · US 7 · FR 5 · DE 3 | 29.2% | — |
India, Pakistan and Peru supplied about 6,700 of those installs and produced ZERO Play buyers. South Africa's 1,510 produced one. The people who actually bought on Play in those months were in the US, Germany, Britain, Canada and Sweden — which is not who we bought.
Did those 5,526 people ever open the app? Yes — and then they left within days.
Amplitude daily active users, four matched 13-day windows either side of the blitz. CHECKED — Amplitude segmentation API, 10 Aug
| Window (13 days each) | Active users/day | What it tells us |
|---|---|---|
| BEFORE — 12–24 Jun | 570 | baseline, no install ads |
| DURING — 25 Jun – 7 Jul the blitz | 822 | +252/day (+44%). The installs were real and the people did open the app. |
| AFTER — 8–20 Jul | 600 | 🔴 Straight back to baseline. Of ~5,526 installs, the lasting daily lift is ≈30 users. |
| NOW — 29 Jul – 10 Aug | 1,038 | higher than any of the above, with the install ads off — this is the quiz and organic, not them |
Honest limit: Amplitude counts all app users, so this measures the whole app's activity rather than a tagged install cohort. Isolating the cohort itself needs AppsFlyer raw data — see PA.9.
Did it produce revenue? The most generous honest answer is "roughly half its cost".
| Google Play NET payout | May (no ads) | June | July |
|---|---|---|---|
| NET revenue | £115.78 | £378.37 | £730.65 |
| Charges (units sold) | 6 | 14 | 23 |
Play revenue did rise, and it is not the July promo — the SKUs are quarterly_welcome_expiring, quarterly, yearly; there is no lifetime SKU in any month, so the $297 sale is cleanly excluded. Against May as the no-ads baseline, the June + July lift is ≈£877, against ≈£1,816 spent — and that credits every penny of Play growth to the ads, which is the most favourable assumption available.
PA.6 · The app funnel it would feed — measured before we point traffic at it
Live census of everyone on the app side with an email, seen in the last 90 days. CHECKED — funnel_stage_census_2026-08-10.json, 15,523 contacts
| Where they stopped | People | Share |
|---|---|---|
| Signed up, never finished the survey | 4,353 | 28% |
| Finished the survey, never did a single workout | 6,808 | 44% |
| Trained, never started a trial | 2,799 | 18% |
| Payers | 1,130 | 7.3% |
The app gives three free workouts, then asks for money. 6,808 people never reached workout one — the single biggest leak anywhere in this document. That is the funnel an app-first ad would pour into, and it is the strongest argument for fixing activation before buying more of it.
PA.7 · 🔑 The door Aga is describing is already built — and nobody is counting who walks through it
Before proposing anything new we checked whether it exists. It does. Read out of the live v1 quiz bundle index-yC00xO94.js (1.34 MB, canary-checked): CHECKED 10 Aug
| What is in the live build | Already tagged? |
|---|---|
apps.apple.com/…/id1357148593?ct=paywall_exit | ✅ Apple campaign token |
play.google.com/…com.themovementathlete.app&referrer=utm_source=quiz&utm_medium=paywall_exit | ✅ Play install referrer |
So "finish the quiz, don't buy, download the app free" is already coded and already instrumented at the destination. It sits on the paywall's exit/expired state.
page_view · screen_view · session_start · scroll · first_visit · user_engagement · generate_lead · add_to_cart · Lead · view_promotion. There is no click or outbound event of any kind, and no separate exit screen renders (R10Paywall is the only paywall screen, 174 views). Take-up of the app door is NOT KNOWABLE today.And it cannot be recovered from the store either: Google retired the bulk
acquisition/ reports after June 2021, so install-by-channel is unavailable for any recent month, and Apple's ct= needs App Store Connect app-analytics — a different API from the finance one we use. Adding one click event is the cheapest measurement in this whole document.PA.8 · ✅ Aga's question, answered — and how to test it properly
Your instinct is right. The mechanism has a trap, and we now know exactly what it is.
What failed in June and July was a raw app-install campaign — we told Meta and Google "get installs", and they did exactly that by finding the cheapest installers on earth, in India at four cents each. That is not what you described. Yours is quiz → results → app, which makes someone spend five minutes answering questions before the app is ever mentioned. The install experiment does not refute your idea — it hands us the failure mode to design against.
So: yes, worth testing. With four conditions the data insists on, and one free step first.
| The decision | Why the data forces it | |
|---|---|---|
| STEP 0 free | Add a click event to the two app links that already exist, and read two weeks of it. No build, no spend. | The door is live and untracked (PA.7). If people already take it, we know the appetite before designing anything. If nobody does, the idea is answered for free. |
| 1 | Geo-lock to US · UK · CA · AU · DE · SE · CH · NL · FR. Hard-exclude IN, PK, PE, BD. | ~6,700 installs from IN/PK/PE produced zero Play buyers; 67–69% of all paid installs came from never-buying countries. |
| 2 | Optimise for the quiz lead. Never APP_INSTALLS or OUTCOME_APP_PROMOTION. | That objective is precisely the thing that fetched India at $0.04. The objective chose the audience, not the creative. |
| 3 | Judge it on payers — Play NET + iOS NET + the funnel census. Never on Meta's install count. | Meta reported 5,526 installs; the store reported about one buyer from 72% of them. |
| 4 | Fund it by reallocation — ~$20/day × 14 days ≈ $280, taken from the live lead-gen campaigns currently producing 0 sales. Stay inside the ~$3,175/mo envelope. | There is ~$865/mo of headroom, but exceeding the envelope is an Aga decision, not an optimisation. |
The gate, written before it runs so it cannot be argued about afterwards:
✅ ≥3 attributable payers in 14 days → the offer is the problem, not the traffic. Extend and widen.
🔴 0–2 payers → the constraint is Meta cold-traffic quality, not the paywall. Then the recommendation becomes stop buying Meta cold traffic altogether and move that budget to retargeting, which is the only Meta spend that has ever sold anything (PA.4).
🔴 And the case against, stated as plainly as the case for: a Play sale nets £31.77 against the web quiz's $85.82, so an app-first path must convert about twice as well just to draw level — and it feeds a funnel where 6,808 people have signed up and never done one workout (PA.6). If activation is the real constraint, more traffic into it makes the leak bigger, not the revenue.
PA.9 · What this section cannot tell you
| Question | Why not |
|---|---|
| Did the 3 brand-search sales need the ad? | No holdout has ever been run. The click is certain; the incrementality is not — some share would have clicked the organic result below the ad. NOT KNOWABLE |
| Which channel caused a recent app install? | Google retired the bulk acquisition/ reports after June 2021. Install-by-country is available; install-by-channel is not. NOT KNOWABLE |
| How many people take the paywall-exit app door? | No click event exists on the quiz (PA.7). One line of code fixes it. NOT KNOWABLE — today |
| Do PMax's 3 claimed purchases exist? | No Stripe sale carries a matching Google click id. Reported as claimed, not as fact. UNCORROBORATED |
| What did the 5,526 installs do inside the app? PARTLY ANSWERED 10 Aug | Partly knowable, and now measured. Amplitude shows daily actives rising 570 → 822 during the blitz and falling back to 600 after (PA.5) — so they installed, opened, and left. What is still missing is the cohort in isolation, and the reason is not a bug: the AppsFlyer raw-data export returns HTTP 400 with "your current subscription package doesn't include raw data reports". 🔴 That is a plan limitation, not a broken integration — closing it costs money, so it is Aga's call, not a fix. |
Every one of these is a measurement gap, not an analytical one. Three of the five close with small engineering changes — the click event, the AppsFlyer export, and the attribution one-liner in §19 #3.
PA.10 · The identity-resolution proposal — the growth fleet's verdict 3 seats + adversarial skeptic gate · 11 Aug 2026 · DEC-2026-08-11 registered, review 15 Dec
The proposal analysed: pilot US-only consumer identity resolution (Retention.com-class, not RB2B) — a script resolves a share of anonymous US visitors to email addresses via the vendor's third-party consent network and delivers them into ActiveCampaign for an educational arc ending at the free assessment. Hypothesis: recovering an already-paid-for visitor by email beats re-buying them through Meta/Google remarketing. The proposal was unusually rigorous by our own laws — legal as a hard gate, a randomized holdout, incrementality-first metrics, explicit kill criteria. The fleet's job was to test it against TMA's measured reality.
UPDATED 11 Aug PM. This section, and the three specialist reviews behind it, did not engage the proposal's competitive evidence — Legion Athletics, MUD\WTR and Cymbiotika appeared zero times, and Retention.com was never named. Raised by Nic; correct on both counts. Those case studies have since been read at source, sized against their own public revenue and traffic denominators, and transplanted onto TMA three ways — they imply $19–$85/mo like-for-like, reproducing the chain below from an independent direction. Full working: the response page, §2.
The decision: NO to the vendor pilot now — not a permanent veto of the idea, a NO on this pipe, this window, this vendor path
The skeptic's ruling after attacking all three seats (including in the pro-proposal direction): the NO is grounded three independent ways at once, which is the opposite of a fragile verdict —
| Seat | The independent kill | The load-bearing fact |
|---|---|---|
| Deliverability (veto seat) | 🔴 The AC route is closed. Loading vendor-resolved, never-opted-in identities into the ActiveCampaign ACCOUNT that holds the 83k list account + domain both CONFIRMED 11 Aug — see the note below | ActiveCampaign's own acceptable-use policy prohibits third-party-sourced addresses — one abuse report risks suspension of the whole account (a policy fact, not a probability). 🔴 Answered by Nic: send from hello@ instead and the addresses never enter AC, so this exposure does not exist — the trade moves to the transactional/payment-recovery domain instead (below) |
| Economics | 🔴 Can't afford what it actually costs. The vendor fee (~$0 free tier) is a decoy — the real costs are setup, founder attention, and list-reputation expected loss | Chain arithmetic at generous assumptions: 3,000 free-tier contacts/90d → ~1.2 trials ≈ $1,667 per trial vs the $20 north star; break-even needs a cold-contact→trial rate 4–33× industry base rates MODELLED — and robust: even 5× optimistic it fails by 83× |
| Attribution / measurement | 🔴 Can't measure it. The proposed 4-arm 30-day holdout produces noise, not evidence, at our volume | Realistic volume ≈ 810–1,350 identified emails/mo MODELLED → 275 per arm at 30 days vs 640–7,738 needed; powering the metric that pays the bill (purchases) needs 19,600+/arm. A test you cannot power is a spend, not an experiment |
hello@, like we already mail people outside AC" — is right on its own terms: if the addresses never enter ActiveCampaign, the acceptable-use exposure does not exist. That finding is removed. What it does NOT change, now read live off the wire: the AC marketing estate, the SendGrid transactional mail and the hello@ 1:1 rail all send from the same mailbox — hello@themovementathlete.com (AC API from-address on msgs 13674/13661/13662 = the 993 SALES SEQUENCE arc, plus 11460/13071; SendGrid's only authenticated domain is themovementathlete.com and its only verified sender is that address; the 1:1 rail sends from it via the Gmail API). So the route changes which system sends, not which reputation carries it. 🔴 We published a wrong correction in between: a census of the whole AC message table returned legacy other-brand domains and we briefly concluded the estates were separate — the wrong instrument, because that table holds 13,132 records spanning years of other brands. Only the read off the live-wired messages could answer it. Unchanged either way: the legal position (law attaches to the data, not the pipe) and the volume/cost/measurability findings. Full write-up: the response page.And the premise doesn't match our books: the proposal's frame is "stop repeatedly paying Meta/Google to re-acquire the same visitor" — but TMA is not doing that today. Everything that ever sold on Meta is PAUSED; live spend is direct lead-gen (people who give an email immediately) plus one ~$5/day warm test. There is no remarketing spend to replace — so the pilot must stand as a NEW cold lane, and as a new cold lane it fails the three tests above. the live-spend figure itself is [UNRECONCILED] — $897/mo vs $2,310/mo in the 10 Aug pull, FLEET-57; the verdict holds at either figure
What the fleet ADOPTS from the proposal — its good ideas have a cheaper, legal home we already own
The skeptic's strongest finding cuts the other way: the proposal's one genuinely valuable idea — "own more of the relationship you paid to create, and suppress paid delivery to people you already own" — does not need the vendor. We already hold the emails that matter (magnet + quiz capture, in the tens of thousands), collected with real consent. Adopted now:
- The 4-group audience architecture for paid suppression, built on FIRST-PARTY captured emails — wired into the November BF retargeting build, when there will finally be remarketing spend worth suppressing. Zero legal exposure.
- Holdout/incrementality-first as standing law for any new lane — the estate already runs touched-vs-holdout on abandonment; this makes it non-optional.
- A monitoring canary on every data sync — priced by today's finding that the Facebook lead-form→AC sync sat dead for 8 months (~2,100 paid leads expired unnoticed).
- 🔑 The honest test of the proposal's whole frame runs in November for £0: the BF first-party own+suppress holdout. If owning the relationship lifts ROI there, the frame is proven — on opted-in data, with no vendor, no legal gate, no list risk.
FLEET-56 · DEC review 15 Dec): the vendor lane is reconsidered ONLY if (a) the November first-party own+suppress holdout shows a real ROI lift AND (b) legal counsel clears US-only anonymous-visitor resolution for a UK-established company (UK GDPR applies by establishment; Art 14 notify-within-a-month; DPIA required; the ICO is hostile to third-party lists; a live US class-action wave targets exactly this pixel class on the site operator). If both clear, the ONLY buildable shape is the deliverability seat's 10-condition isolated pilot — separate ESP, separate cousin domain, cold-list doctrine, graduation into the main estate only by an explicit act — earliest Dec 2026/Jan 2027, which is also the best fitness-intent month to read it. The list is the moat; nothing rents a hole in it.
PA.11 · Apple Search Ads — "is this only on branded terms?" — and the 5-check compliance sweep, re-measured every figure pulled live 12 Aug 2026 · Apple Search Ads API v5 · Google Ads API v25 · Meta Marketing API v21 · AppsFlyer aggregate partners report
🍎 Aga's question, answered: "but this is ONLY on branded terms right? so people ALREADY looking for TMA, NOT from any other keywords"
Yes — and it is stronger than that. Not only did the money come from our own brand terms, every penny of it did. Over the June–July window 61% of the spend went to competitor and non-branded terms and returned $0.00 and zero buyers. The five campaigns that did it are all now paused.
| Apple Search Ads · 1 Jun – 31 Jul 2026 | Spend Apple's own billing | Taps | Installs | Buyers | Revenue AppsFlyer attributed | ROAS |
|---|---|---|---|---|---|---|
OUR OWN BRANDBranded - US - SR $78.47 · Branded - EU $55.54 · Branded - UK $12.86 — all three still RUNNING | $146.87 38.9% | 265 | 116 | 6 | $1,098.82 | 7.5× |
NOT our brandCompetitive - Branded $78.16 · Non-Branded $65.57 · Competitive - Non-Branded $43.08 · Non-Direct Competitive - Demand Gen $27.83 · Competitors_MC_CA-EU $15.81 — all five PAUSED | $230.45 61.1% | 213 | 15 | 0 | $0.00 | 0.0× |
| TOTAL | $377.32 | 478 | 131 | 6 | $1,098.82 | 2.9× |
Revenue by campaign, AppsFlyer, same window: June $99.98 / 1 buyer — Branded - US - SR (it appears in AppsFlyer under its raw id 2144133585, resolved back through the Apple API). July $998.84 / 5 buyers — Branded - EU $763.89 (3) + Branded - US - SR $234.95 (2). Every non-brand campaign: $0.00, 0 buyers. August 1–11 continues the pattern — $1,178.34 from 3 buyers, 100% branded.
⚠️ And the 5.6× itself does not reconcile — the spend does, the revenue does not
The figure in circulation: $404 spent across June and July → $2,277 revenue → a 5.6× ROAS. The spend holds up. The revenue does not reproduce from the attribution data, and the gap is roughly 2×.
| The quoted figure | Measured here | Verdict | |
|---|---|---|---|
| Spend | $404 | $377.32 (1 Jun–31 Jul) $408.24 (12 Jun–12 Aug) | ✅ Reconciles. Apple's own billing, same order — the window boundary explains the difference |
| Revenue | $2,277 | $1,098.82 | 🔴 Does not reconcile — roughly 2× apart |
| ROAS | 5.6× | 7.5× branded · 0.0× non-brand · 2.9× blended | The single blended figure averages a working line with a dead one |
The most likely explanation, so we know what to look for: Apple's own dashboard attributes on Apple's window, which counts view-through installs. Over 12 Jun – 12 Aug Apple recorded 165 tap-installs but 305 view-installs — a view-inclusive basis roughly doubling credited revenue is exactly the shape of this gap. If that is what it is, both numbers are "right" and we simply state which basis we quote. The source report and its exact date range need naming — NIC-109. The underlying export is not reachable from the repo, so that is the one part this analysis cannot close itself. until then the $2,277 is [UNRECONCILED] and should not be quoted
What this does not change — already settled, do not re-litigate
- 🔴 Causation is NOT KNOWABLE. Branded search buys clicks from people typing our own name — most would have arrived anyway. The control sits in the same report: organic delivered 891 installs, 37 buyers and $13,495.97 over the same two months, for free. ASA branded added 6 buyers for $146.87. A ROAS here is an ATTRIBUTED ROAS and is never evidence the spend created the revenue.
- It cannot scale. Branded is bounded by how many people search our name — ~7,084 impressions/30d. Tripling the budget will not spend it. DEC-2026-08-11-keep-the-3-branded-apple · review 30 Sep
- The non-branded experiment has already been run and it failed — $230.45 across five campaigns → 15 installs, 0 buyers, $0.00. That is why they are paused. Any future non-branded test is a new experiment that must argue against this result, not a fresh idea.
- Keep the three branded campaigns running. The asymmetry favours it: worst case is ~$75/month wasted, best case is the attributed figure. Nothing to do.
The 5-check paid compliance sweep (AGA-104, minted 5 Aug) — re-measured 12 Aug: 4 of 5 answered with no ad-UI visit, and 2 of its recommendations were wrong
| # | What the row asked for | Measured 12 Aug | Outcome |
|---|---|---|---|
| a | Verify Competitive Non-Branded + Competitors Branded + PMax are PAUSED "per the settled category-search-only ruling" | 🔴 The ruling was mis-stated. google-category-search-only is a CHANNEL rule — "Category Search ONLY. No to Demand Gen and YouTube." Competitive Non-Branded and Competitors Branded are both SEARCH campaigns, so neither breaches it. The one live breach is PMax (PERFORMANCE_MAX serves Display/YouTube/Discover), ENABLED at £5/day. The Demand Gen breach is already paused.🔴 And pausing PMax blind would have killed the account's best line: 30d it took £150.32 → 78 conversions, ALL from its SEARCH arm at £0.53/click. Its non-Search arms took £60.85 across 280 clicks for ZERO conversions. | Real, but re-aimed. The fix is PMax's placements, not the campaign → NIC-108. Competitive Non-Branded's separate performance problem (£245/30d, 34 clicks, 0 conv, £7.21/click) is already staged as AGA-157 for 13 Aug — not duplicated. |
| b | Meta → Manual placements → untick Audience Network as a permanent lock | AN served £0.45 of £628.73 over 30 days — 0.07%. It is already unticked on the only manual ad set (MC Control = facebook, instagram, messenger, threads). The two ad sets on automatic placements are the ADV+ ON arm of the live A/B test against ADV+ OFF Control. | ✅ NO ACTION — and the proposed action would have destroyed a live experiment. Unticking AN on an Advantage+ ad set is what makes it no longer Advantage+. Monitor instead. |
| c | Revert the "+43% budget step on 1 Aug ($35→$50/day)", or record the measurement gate as cleared | The step never reached spend. 20–31 Jul averaged £34.84/day; 1–12 Aug averaged £34.98/day — +0.4%. Enabled daily budgets sum to £36.00/day today. | ✅ NOTHING TO REVERT. A budget cap is not spend. (The gate itself is still uncleared — 86% of Google "conversions" are still email addresses, not sales — but no budget rode on it.) |
| d | Is any YouTube/Video campaign live? — 69,076 "ADVERTISING" views at 4.4 sec each (T14, open since 28 Jul) | No. Zero VIDEO-type campaigns exist; all 17 are Search / PMax / UAC (paused) / Demand Gen (paused). 🔑 And the mystery is solved. In Jun–Jul YouTube was bought — by Remarketing | 4th of July Promo (Demand Gen: £77.35 → 137,955 YouTube impressions, 2,454 clicks) plus the ten UAC app campaigns (~£23.53, ~83,600 impressions). 4.4 seconds is exactly what a skippable YouTube ad view looks like. All of them are now paused. | ✅ Nothing live to switch off. T14 CLOSES — the views were bought advertising, not garbage traffic, and the buying stopped when those campaigns were paused. |
| e | DECIDE: ASA brand defence at ~$8/day — "we currently bid $0 on our own brand terms"; confirm its real August state | 🔴 Premise refuted. We do not bid $0 on our own brand — our own brand is the only thing ASA runs. Three Branded campaigns RUNNING (US-SR, EU, UK), seven paused. Live and billing throughout. | ✅ Already decided — DEC-2026-08-11-keep-the-3-branded-apple, review 30 Sep. Keep, do not scale, causation NOT KNOWABLE. |
NIC-110 reassigned by Aga, 12 Aug: "this needs to go on the to-do list"Quiz Funnel | Competitors Branded | US — cut from £10/day to £5/day. It bids on rivals' brand names (mostly "betterme"), is the biggest line in the account at £314.77/30d = 32% of Google spend, runs at its £10/day cap, and returns 24 conversions at £1.58/click. The problem is cost per output: £13.12 per conversion, against PMax Search £1.19, Branded £1.74 and Meta's cheapest lead line £2.43 — 3rd of 5 Google lines, better than Non-Branded (£40.07) and Competitive Non-Branded (no conversions at all), but 7.5× worse than the two that work. And 86% of this account's conversions are not sales — 105 leads + 4 carts + 4 installs against 18 purchases — so those 24 are leads, and all 3 Stripe-verified Google sales in the 1–10 Aug audit came from Branded, none from this line.Why cut and not pause: competitor-brand search is genuinely in-market intent — "betterme alternative" is a person shopping — which a cold lead-form fill is not, so the two lead types are not interchangeable at any price. And switching a line off before knowing what its output feeds is the mistake the MC Control round already cost us.
Why after
NIC-108: the freed £5/day has one good home — PMax, the cheapest converter and the only line actually at its cap (£150.32/30d against £5.00/day). Branded cannot absorb it: £40.00 in 30 days against a £6/day cap means it is volume-bound, not budget-bound, so money moved there will not spend. But PMax leaks 40% of its budget into zero-return placements until NIC-108 lands, so moving money in first just enlarges the leak. Also hold until after 13 Aug — AGA-157's read lands then, and two simultaneous Google changes make the account-level read unusable.Kill condition, pre-registered so it is not re-argued: at the 10 Sep gate, if this line still shows zero Stripe-verified sales AND cost-per-conversion above £10, pause it. Everything else on the old sweep is closed.
🔧 Paid-ads jobs — a filtered view of §19, not a second list folded in here on Aga's instruction, 12 Aug
Aga: "if there are any tasks that you cannot do and he needs to do in regards to paid ads, fold them in here." These are the paid-ads items the fleet cannot execute itself — ad-account changes ship through the account owner, and capability is not authority (Aga's ruling, 11 Aug: "do not do it yourself"). Every one below was measured by the fleet first, so each is one sitting rather than a research task. 🔴 §19 is the ONE to-do list — this table is a convenience filter showing only its paid-ads rows, so a reader in this section does not have to scroll. Every row below also appears in §19 with its full detail, and §19 is the one that governs. If the two ever disagree, §19 wins.
| Row | What gets done | Effort |
|---|---|---|
NIC-110new, 12 Aug ✅ Aga approved | Cut Competitors Branded from £10/day to £5/day and raise PMax to £10/day — after NIC-108, and after 13 Aug. £13.12 per conversion against £1.19 on the line the money moves to. Kill condition set for the 10 Sep gate. | 10 min |
NIC-106✅ Aga approved | Repoint QLG | ADV+ OFF | MC | Control to the quiz root. Its four live ads still land on /2026/, which does not show the $97. Already specified ad-by-ad in §19 row 1a. | 5 min |
NIC-108new, 12 Aug | Stop PMax buying placements that have never converted — £60.85/30d across Display, Search Partners, YouTube and Discover for 280 clicks and zero conversions, while its Search arm carries the account's cheapest conversions. Do not pause the campaign. Three implementation options offered; he picks and reports which. | 15 min |
NIC-109new, 12 Aug | Name the report and date range behind the $2,277. The spend reconciles, the revenue is ~2× apart from AppsFlyer, and the likely cause is view-through attribution. Five minutes of naming a source settles which basis we quote. | 5 min |
NIC-102 | Ship the staged iOS App Store keyword field — ~86 of 100 characters currently wasted. Adjacent to paid, cheapest ASO change available. | 10 min |
NIC-105 | Remove the third-party GTM container from the checkout — GTM-PPGJXHH is not ours and loads on the page holding a live Stripe key. | 15 min |
AGA-150Aga's, in Ads Manager | Rescue the rotting Facebook lead-form leads — the AC sync died 1 Dec 2025, ~264 leads are still retrievable, and Meta deletes them at 90 days. Export the CSV and reconnect the AC↔Facebook Lead Ads integration. | 10 min |
PA.12 · The "Build Muscle With 40+" lead magnet — tested twice, 513 leads, zero buyers measured 12 Aug 2026 · Meta Graph API · ActiveCampaign API · LIVE STRIPE, both accounts
🔴 The verdict: do NOT restart this campaign. It is already paused — leave it paused.
The Facebook lead form Build Muscle With 40+-higher intent ran under campaign
MOF - Lead Form from Nov 2025 until it was paused on 11 Aug 2026. Lifetime:
£1,940.21 for 2,907 leads — about 67p a lead, which is why it looked attractive.
It has now been tested twice, in the two conditions that matter, and the natural experiment was free: one cohort reached ActiveCampaign and received the full email treatment; the other reached nothing at all, because the AC ↔ Facebook Lead Ads sync died on 1 Dec 2025 and nobody noticed for eight months.
| Cohort of the SAME lead magnet | Leads | Received the email sequence? | Attributable payers |
|---|---|---|---|
Nov 2025 (AC tag 63570) | 250 | ✅ YES — full treatment | 0 |
| 26 May – 11 Aug 2026 (the CSV export, 12 Aug) | 263 | 🔴 NO — sync dead | 0 |
| TOTAL | 513 | 0 |
🔑 This is the whole argument. The obvious hypothesis was "the leads were fine, the broken pipe is what cost us the revenue." The nurtured cohort is the control that refutes it — 250 people got every email we send and converted no better than the 263 who got nothing. The dead sync cost us the chance to nurture them; it is not what stopped them buying.
⚠️ How this nearly went the other way — a tag name read as a payment fact
The first pass on the export reported "6 of these leads became paying customers" and was
briefly used to argue the magnet's quality was fine. That was wrong. The ActiveCampaign tag
stripe-integration means a Stripe customer OBJECT exists — it is created at
checkout, not at payment.
Checked against live Stripe on both accounts: all 7 suspected payers show zero paid
charges and no active subscription; the only one carrying a subscription at all reads
incomplete_expired — a failed checkout. The single apparent payer in the Nov-2025
cohort has been a Stripe customer since 2019-11-09 with his first paid charge 2024-10-01,
a year before the form existed — his later charges are ordinary renewals, so the magnet did not
acquire him either.
Three lessons, each cheap to reuse: (1) before any claim that
someone paid, hit the payment processor — a tag, a flag, a list membership and a derived JSON
are all proxies. (2) ever_paid_emails.json flagged one of them too, so a second
proxy agreed with the first — two proxies agreeing is not verification. (3) the original
"6 of 26" used the wrong denominator: those 26 are precisely the people who had already found
their own way in. The honest base was all 263. selection effect
What happens to the 263 leads — one email, as a closing test
237 of the 263 are absent from ActiveCampaign entirely (the other 26 arrived by other routes —
20 found the quiz on their own). Segment-integrity pre-check against ever_paid_emails.json
returned zero payers among the 237, so all of them are mailable.
They still get one email — and the reason is not that the channel works. The money is already spent, one send costs effectively nothing, and these people asked us for a guide and received silence for up to 78 days. That is a promise worth closing.
- Deliver the guide they were promised — the asset must be confirmed to exist before the email is written. Promising it twice and not delivering is worse than the original silence.
- Acknowledge the gap honestly. A standard welcome would read as though nothing happened.
- New tag, never the old one.
facebook-lead-ads-integrationnow means "arrived before 1 Dec 2025" and is the only evidence of when the sync died. - Pre-registered read: quiz-starts and purchases from this cohort's own UTM at 14 days — never opens.
🔑 Set expectations honestly: given 0 payers from 513, the prior is that
this returns little. It is a loop being closed, not a channel being reopened — and if it returns
nothing, that is the final answer on this magnet. FLEET-73 · nothing sends without
Aga's GO.
1 · The decision table — which one to keep
🔴 Read this before any number below — how much each one can be trusted
Aga challenged two figures on 7 Aug and was right both times. Both came from the same root cause, so it is stated here once, plainly:
The two older checkouts do not record where the buyer came from. The /register/ checkout used by 2026-v2-2 and /2026/ is also used by email promos and direct links. Stripe saves nothing that says "this person came from the quiz". So for those two funnels we are inferring, not measuring.
| Funnel | Can we prove a sale came from it? | What that means here |
|---|---|---|
2026-quiz-checkout-v1 | YES — definitively. Stripe stores quiz_variant=quiz_checkout_2026_v1 on every sale. | Its 14 sales are certain. The only funnel where that is true. (Re-verified 10 Aug: it also stores the persona, the offer taken, and any ad click ID — which is what made §0b possible.) |
2026-v2-2 | No — inferred from the plan and the timing. | Trustworthy in June and July, when the quiz was clearly the main source. Not before. |
/2026/ Control | No — and the least reliable of the three. | Trustworthy in June and July only, and even then it rests on just 15 sales. |
The proof, and why May is now excluded everywhere: the Control's checkout took 40 sales in May while its page had only 138 visitors all month. Day by day, 34 of those 40 landed between 22 and 31 May — every one at exactly $78.50, half of $157. That is the Memorial Day email promo, not the quiz. Aga said the Memorial sale never went through the quiz; the dates prove it.
What changed as a result: every May figure is out of the money comparisons, and the decision table is now built on June and July only for the two old funnels. An earlier version of this page showed v2-2 at $420 and the Control at $1,550 — both wrong. Corrected figures below.
Everything on one row per funnel, in the same units. Each funnel measured in its own fair window (v1: 31 Jul–6 Aug · v2-2 and Control: July, their last full month).
2026-quiz-checkout-v1the NEW page — pay $97 now, no trial
- Sales we can prove came from it
- 15 · $1,298.46 (1 – 11 Aug)
- Visitors reaching the price
- 17.8% — ex-YouTube-crawler, in line with the other two
- What it sells
- A year, paid up front. No trial to leak out of.
- How sure are we?
- 15 sales — still small, but they are the only provable quiz sales we have, and the rate has held for 11 straight days.
2026-v2-2the OLD page — 3 plans, cheapest offer of the three
- Sales we can prove came from it
- None — its checkout is shared. See the box above.
- Visitors reaching the price
- 22.4% — actually the best of the three
- What it sells
- Three cards. The $0 free trial is pre-selected; the dearest option is $24.98.
- How sure are we?
- *Its top card is $24.98 vs v1's $97 — that much is on the page. The revenue split is not knowable.
/2026/ — the Controlfree week, then full price $157/year
- Sales we can prove came from it
- None — its checkout is shared.
- Visitors reaching the price
- 19.1% — in line with the other two
- What it sells
- A year at full price, after a 7-day trial. Verified on the live checkout.
- How sure are we?
- Least is known about this one. Keep it because it costs little and draws a different audience — not because it out-earns v1.
🔴 STOP — the money cannot be split by funnel. Here is the proof.
Aga challenged the Control's pricing on 7 Aug. Checking it broke a bigger assumption, so this replaces the revenue comparison that stood here before.
All three quiz builds link to the exact same set of old checkout pages. We fetched and searched all three live bundles — they contain an identical list, in identical quantities:
| Old checkout page linked in the build re-read from the live bundles, 10 Aug | 2026-v2-2 | /2026/ | …checkout-v1 |
|---|---|---|---|
/register/yearly3/ | 7 | 7 | 1 🔴 |
/register/yearly_2026/ — v1's dominant annual link | 1 | 1 | 8 🔴 |
/register/monthly_2026/ | 2 | 2 | 2 |
/register/quarterly_2026_promo/ | 1 | 1 | 1 |
/register/monthly_tripwire_2026/ | 1 | 1 | 1 |
/register/monthly_2026_promo/ | 1 | 1 | 1 |
/2026/ and /2026-v2-2/ are byte-for-byte identical in their checkout links — 13 refs, same six paths, same counts. v1 is different: its dominant annual link is yearly_2026 (8 refs) not yearly3 (1), and it additionally carries 8 references to its own payments/quiz-checkout, which the other two have none of.This makes the core point stronger, not weaker. The two builds whose money we actually need to separate — the Control and v2-2 — are the two that are identical. Nothing distinguishes a Control sale from a v2-2 sale, and nothing ever will without a code change.
So a sale on yearly3 could have come from either old build — or from an email, or a direct link. Stripe records nothing that separates them. The split used earlier on this page (yearly3 = the Control, monthly3/quarterly2 = v2-2) was inferred from Nic's written description, not measured — and the code shows it is not true.
| Figure published earlier on this page | Status now |
|---|---|
| Control "$92.72 average per sale", "$1,990 per 1,000 visitors" | 🔴 WITHDRAWN — built on sales that cannot be shown to come from that page |
| v2-2 "$25.65 average per sale", "$506 per 1,000 visitors" | 🔴 WITHDRAWN — same reason |
| "Control buyers paid $105.35 each, 82% still paying" | 🔴 WITHDRAWN — that is the checkout's buyers, not the page's |
2026-quiz-checkout-v1: 8 sales · $691.48 · $86.44 average (7 Aug) | ✅ STANDS, and re-verified 10 Aug — the same query reproduces 8 · $691.48 · $86.44 exactly for that window. Now 15 · $1,298.46 · $86.56. |
And on the $78.50 specifically — Aga was right that no buyer can pay it from that page. The Control's trial button leads to a full-price $157/year 7-day trial. 🔴 CORRECTED 10 Aug — the URL named here was wrong: the destination is /register/yearly3, not checkouts-v3/trial/12month/. The Control's live bundle contains zero checkouts-v3 references (canary-checked read of index-BqqEpqKj.js, 1.31 MB) — and §13 said so, so this page contradicted itself. The $157 conclusion is unchanged and is now confirmed a better way: by the Stripe price of the trial the Control actually creates. $78.50 is exactly half of $157 — a coupon, which means a promo link, not this page's offer.
The honest position: we have revenue for one of the three funnels. Not for the other two.
Part 1 — what IS measured, for all three
Google Analytics sees everything up to the moment someone leaves for a checkout, and it knows which build they were on. These figures are solid for all three. CHECKED — re-pulled 10 Aug
2026-quiz-checkout-v1 | 2026-v2-2 | /2026/ Control | |
|---|---|---|---|
| Window | 30 Jul – 10 Aug | 1 Jun – 31 Jul | 1 Jun – 31 Jul |
| Visitors (real people, crawler removed) | 607 | 2,831 | 700 |
| Gave us their email | 117 | 643 | 146 |
| Email capture rate | 19.3% | 22.7% | 20.9% |
| Reached the price screen | 108 | 634 | 134 |
| Reached the price | 17.8% | 22.4% | 19.1% |
Leave them in and v1's capture rate reads 3.2%. Take them out and it reads 19.3%. Same people, same week. Every v1 figure on this page has them removed — and the 7 Aug prediction that it would stop by 9 Aug came true (daily series: 0 on 9 Aug, 1 on 10 Aug; verified 10 Aug PM). The historical 3–8 Aug window is now excluded at source in both cockpit pullers, so the warning applies only to windows touching those six days.
Part 2 — the money, split into what we know and what we don't
| Sales | Cash | Average | Can we attribute it? | |
|---|---|---|---|---|
2026-quiz-checkout-v11 – 10 Aug | 14 | $1,201.46 | $85.82 | ✅ Yes, provably. Stripe names the quiz build on every sale — and since 10 Aug we can also name the traffic source on 11 of the 14 (§0b). |
| The shared old checkout 1 Jun – 31 Jul, promo-free months | 71 | $2,827.05 | $39.82 | 🔴 No. Fed by 2026-v2-2, /2026/, 2026-v2-3, emails and direct links alike. |
| The shared old checkout — August 1 – 10 Aug | 6 | $291.87 | $48.65 | 🔴 No — and all six are July trials maturing, not August visitors. Subscriptions created 9–29 Jul, first payment landing in August. CHECKED 10 Aug |
What that shared checkout sold in those two months — all buyers, all sources, not only quiz visitors:
| Plan bought | Sales | Share | Cash |
|---|---|---|---|
| 12-Week — half price $24.98 | 28 | 39% | $699.44 |
| 1-Week free trial → then $24.97/mo | 15 | 21% | $374.55 |
| 12-Month at $78.50 — a coupon price that exists on NO quiz page; promo-email buyers on the shared checkout | 10 | 14% | $785.00 |
| 4-Week — half price $12.48 | 5 | 7% | $62.40 |
| 12-Week — full price $49.97 | 4 | 6% | $199.88 |
| 12-Month — full price $157 | 3 | 4% | $471.00 |
| other / legacy prices | 6 | 9% | $234.78 |
| TOTAL | 71 | 100% | $2,827.05 |
2026-v2-2 nor /2026/ shows a $78.50 option anywhere, and the Control's own button goes to a full-price $157 trial. So those buyers arrived through the same checkout from somewhere else — most likely a promo email. That is the whole attribution problem, visible in two rows.✅ So what should we keep — decided on what we can actually prove
| Funnel | Recommendation | The evidence it rests on |
|---|---|---|
2026-quiz-checkout-v1 | KEEP as the main one. Give it the full 14 days. | The only funnel whose revenue we can see at all. $85.82 a sale across 14 provable sales against a shared-checkout average of $39.82, moves people to the price as well as the others (17.8% ex-crawler), and sells a year up front instead of a trial. |
2026-v2-2 | STOP sending traffic. Leave the page up as a rollback. | ⚠️ Weaker evidence than this page previously claimed. Its sales can't be isolated. What we can say: its price cards top out at $24.98, it pre-selects the $0 option, and its checkout's blended average is $39.82 — less than half v1's $85.82. And since 8 Aug it receives no paid traffic at all. |
/2026/ Control | KEEP on Facebook only. Do not add to Google. | Costs nothing extra, different audience so it isn't taking v1's traffic; sells the full-price $157 trial. We cannot show it earns more or less than v1 — the earlier claim that it did has been withdrawn. |
What would settle this properly, and it is a small job: make the two old builds pass their own name into the checkout, the way 2026-quiz-checkout-v1 already does. One line each. Until then no revenue comparison between these three funnels can be trusted — including any in an earlier version of this page.
2 · The map — what each funnel is checked live 7 Aug
Nic's documentation (sent 7 Aug), recorded as-sent, with our live verification beside each claim.
| Funnel | Landing | Paywall | Pays via | Offer at paywall | Ads feeding it — verified |
|---|---|---|---|---|---|
| v1 — new primary | /2026-quiz-checkout-v1/root 302s here ✅ |
none in React — redirects to checkout | app…/payments/quiz-checkout/success: …/quiz-checkout/success |
$97 annual today ($157 struck) · 50%-off monthly downsell · no trial | All 4 Google quiz campaigns ✅ (GA4-confirmed) · Meta: only stray sessions |
| v2-2 — old primary | /2026-v2-2/HTTP 200 · NOT redirected ✅ |
in-quiz | legacy /register/ → Stripe prices monthly3 · quarterly2 |
7-day trial + evergreen 50% off first payment (≈$25 avg first charge) | 🔴 Meta "ADV+ ON | US | Variant 2" — ACTIVE, both ads deep-link here (creatives read 7 Aug). Nic's note said the redirect cut this off — it doesn't; ads land below the root. |
| Control | /2026/HTTP 200 · deliberate ✅ |
in-quiz | legacy /register/yearly3 confirmed in the live bundle 10 Aug — 7 refs, and zero checkouts-v3 refs |
7-day free trial → $157/yr FULL PRICE (verified live on its checkout) | Meta "ADV+ OFF | MC | Control" — ACTIVE, kept on purpose (multi-country, cheap leads) |
| v2-3 — legacy stub | /2026-v2-3/ · 200 |
in-quiz | same old /register/ set |
as v2-2 | No ads. Residue: AC email 967 TMA | Welcome Sequence After LEAD MAGNET E7 still links it + linktree |
3 · "90% of traffic goes to the new page" vs "a third never see the $97" — both are true
Nic said most traffic goes to the newest version. He is right. This page said a large share still miss the $97 offer. That is also right. They are answers to two different questions, and quoting either one alone causes an argument. Here is the same week counted both ways. CHECKED — pulled 7 Aug
The full window, 30 Jul – 10 Aug re-pulled 10 Aug
| 30 Jul – 10 Aug | Sessions what Analytics shows by default | share | Real people who started the quiz crawler traffic removed | share |
|---|---|---|---|---|
2026-quiz-checkout-v1 — has the $97 | 5,736 | 90% | 607 | 65% |
2026-v2-2 — no $97 | 272 | 4% | 157 | 17% |
/2026/ Control — no $97 | 311 | 5% | 163 | 17% |
2026-v2-3 — no $97 | 21 | 0% | 10 | 1% |
| TOTAL | 6,340 | 100% | 937 | 100% |
Why the two columns disagree so violently
The new page shows 5,736 sessions but only 607 real quiz starts. The other ~5,100 are overwhelmingly YouTube's link-checking crawler — while we were retrofitting links into 869 video descriptions, YouTube fetched each one every time it was edited. Analytics records them as real sessions. They can never buy anything. The single campaign yt_description accounts for 4,828 sessions on its own.
And the historical window is now excluded at source (§19 done table): both cockpit pullers subtract
yt_description inside its fixed 3–8 Aug window, so the two columns converge from 9 Aug onwards and no v1 rate needs a hand correction any more.✅ And the post-fix picture — 8–10 Aug, after the Meta ads were repointed
The Variant 2 ads were repointed to the quiz root and Meta traffic to /2026-v2-2/ hit zero on 8 Aug. This is what the split looks like now that the leak is closed. Sessions, crawler shown separately. CHECKED
| 8 – 10 Aug | Sessions | ex-crawler | share ex-crawler | Status |
|---|---|---|---|---|
2026-quiz-checkout-v1 — has the $97 | 774 | 291 | 79.1% | ✅ where everyone should be |
/2026/ Control — no $97 | 71 | 71 | 19.3% | 🟥 NIC-106): repointed to the quiz root on Aga's ruling. The control arm ENDS. |
2026-v2-2 — no $97 | 4 | 4 | 1.1% | ✅ was the leak — now stragglers only |
2026-v2-3 — no $97 | 2 | 2 | 0.5% | residual (one old email still links it) |
The unintended leak is closed: 15% → 1.6%. Of everyone reaching a quiz build now, 1.6% miss the $97 by accident (v2-2 + v2-3) and 19.3% miss it on purpose (the Control). That job is in §19's ✅ done table.
🔴 And a correction to this page's own figure: it was 43%, it is actually 33%
An earlier version said "of 540 real visitors, 232 (43%) still hit a build with no $97 offer." That window was wrong. It ran 30 Jul – 5 Aug, and 30 July was still a pre-switch day — v2-2 was still receiving traffic from the quiz root that morning, which inflated its share.
| Published earlier | Correct | |
|---|---|---|
| Window | 30 Jul – 5 Aug (includes a pre-switch day) | 31 Jul – 6 Aug |
| Real visitors | 540 | 462 |
| Miss the $97 offer | 232 · 43% 🔴 | 153 · 33% ✅ |
| …of which deliberate (the Control) | 16% | 18% |
| …of which unintended (v2-2 + v2-3) | 27% | 15% |
The leak is smaller than this page reported — 15% unintended, not 27%. ✅ UPDATE 10 Aug: it has since been closed entirely. The unintended share is now 1.6%.
✅ Recommendation — always publish both, and label them
- "90% of sessions go to the new page" answers where is our traffic going? Use it for routing and for checking the switch worked. It did.
- "35% of real visitors never see the $97" (30 Jul–10 Aug) answers how many potential buyers never get the offer? Use it for money decisions. ✅ On the current 8–10 Aug window that is 20.9%, and only 1.6 points of it is unintended — the rest is the Control, on purpose.
- Never quote one without the other. This exact mismatch is what put Nic's read and this page's read in apparent conflict when both were correct.
- ✅ The "re-check after 9 August" advice held after all (corrected 10 Aug PM). The crawler stopped 9 Aug with the retrofit's completion, and
yt_descriptionis now excluded at source inside its fixed 3–8 Aug window (§19 done table). Sessions and real-people counts converge from 9 Aug onwards; for any window touching 3–8 Aug, keep quoting both numbers.
4 · The three paywalls, side by side — what each one actually sells
This is the whole difference between the funnels. The quiz body is near-identical across all three; the offer at the end is not. Click any image to open it full size.
$97 charged TODAY · $157 struck · 48h timer
"Charged today · nothing more for 12 months · renews $157"
→
payments/quiz-checkout/Spec: ship-standalone.html ↗
$0 today · 1-Week free trial selected by default
4-Week $12.48 · 12-Week $24.98 (both 50% off)
CTA: "Get My Transformation Free" → then $24.97/mo
→ legacy
/register/ · prices monthly3 · quarterly2
/2026/$0 TODAY · 7-day free trial
then $157/year — FULL price ($13.08/mo shown · $0.43/day) · coffee comparison. No $78.50 exists on this page — that price only ever came via promo-email coupons through the shared checkout.
CTA: "Start My Free Trial"
→ 🔴 CORRECTED 10 Aug:
/register/yearly3, not checkouts-v3
| v1 · $97 | v2-2 | Control | |
|---|---|---|---|
| Money on day one | $97 | $0 (or $12.48 / $24.98) | $0 |
| Free trial? | No — charged immediately | Yes, 7 days (default option) | Yes, 7 days |
| What they pay after the trial | Nothing — the $97 already covers 12 months. Renews $157/yr. | $24.97 every month (or $49.97 per quarter) | $157 for the year, full price — verified live on its checkout |
| Plan class sold | Annual — the class that retains | Monthly / quarterly — the class that churns | Annual |
| Average money per sale | $85.82 measured · 14 sales | Not knowable separately — these two share one checkout with each other, with emails and with direct links | |
| Urgency device | 48h timer, "closes permanently" | 48h timer + 50%-off framing | None — value framing only |
5 · What the v2-2 price page actually says — and the problems in it
Read off the screenshot above, then every claim re-checked against the live code (index-DPkL72Wi.js, fetched 7 Aug). CHECKED
1 · It pre-selects the option that makes us the least money
The 1-Week · $0.00/day · "$0 today" card is already highlighted when the page loads. The visitor doesn't choose the free trial — the page chooses it for them. The two options that actually take money ($12.48 and $24.98) sit beside it, unselected.
This is the clearest single explanation for the money gap. v2-2 averages $25.68 a sale; 2026-quiz-checkout-v1 — which pre-selects its 12-month option — averages $85.82. Same product, same quiz, different default. But the free trial is NOT what most people end up picking — see the corrected plan mix below.
🔴 Correction — what people actually bought on v2-2 CHECKED
An earlier version of this page described v2-2 as selling "a monthly plan". That was wrong, and Aga caught it. The page offers three cards, and people overwhelmingly pick the 12-Week one. Every sale through this checkout in the last 95 days:
| Which card they picked | Sales | Share | They paid | Total cash |
|---|---|---|---|---|
| 12-Week — half price | 61 | 60% | $24.98 | $1,523.78 |
| 1-Week free trial → then billed monthly | 21 | 21% | $24.97 | $524.37 |
| 12-Week — full price | 7 | 7% | $49.97 | $349.79 |
| 4-Week — half price | 6 | 6% | $12.48 | $74.88 |
| older 4-Week prices | 6 | 6% | $9.98–$44.97 | $119.87 |
| TOTAL | 101 | 100% | $25.67 avg | $2,592.69 |
69% of v2-2's sales are the 12-Week (quarterly) plan, not monthly. Only 21% come through the free trial at all — so the pre-selected default is not what most people end up buying. That is genuinely good news about this page. It also means the money gap against 2026-quiz-checkout-v1 is about the price point ($24.98 vs $97), not about the type of plan.
2 · Two buttons on screen at once promise opposite things
| Where | What it says | What it promises |
|---|---|---|
| Sticky bar, bottom of screen | "Get My Transformation Free →" $0 today · then $24.97/mo | FREE |
| Main button, mid-page | "Claim My Plan →" Charged today · cancel any time · 60-day money-back guarantee | CHARGED NOW |
Both are visible in the screenshot at the same time, with the $0 option selected. Confirmed in the code — the function writing that reassurance line returns "Charged today · cancel any time · 60-day money-back guarantee" anyway. A visitor cannot tell whether they are about to be charged. That is exactly the ambiguity that stops a card being entered, and it is a small fix.
3 · Two claims the page makes that our own data contradicts
| The page says | What we actually measure |
|---|---|
| "The programme people don't cancel" under the price comparison | Measured on this exact checkout: of the buyers who joined Feb–Apr and are now 4–6 months in, only 28% are still paying (23 of 81). The Control's buyers over the same months are at 82%. So this genuinely is our worst-retaining funnel — though the honest figure is 28%, not the 15.6% an earlier version of this page quoted. |
| "TMA — your plan · $0.00 / day" in the "what this costs without TMA" table, set against a personal trainer at $60–100 a session and an AI coach at $149 a month | $0.00/day is the 7-day trial rate. The real ongoing price is $24.97/mo ≈ $0.82/day. The table sets our free week against everyone else's ongoing monthly price — the only row that isn't like-for-like is ours. |
Neither is small if a store, an ad platform or a member ever queries it. Both are one-line fixes.
4 · The pop-up names plans that don't exist on the page
The sticky notification reads "⚡ Robert from Dublin joined Longevity Plan", and the code also rotates "joined Muscle Builder Plan". But this page sells 1-Week, 4-Week and 12-Week — there is no "Longevity Plan" or "Muscle Builder Plan" on it anywhere. Anyone reading closely sees proof about a product that isn't for sale, which undercuts the very thing it is there to do. (Same component and same plan names sit in the Control and v1 builds too.)
✅ Four things v2-2 does BETTER — copy these into 2026-quiz-checkout-v1
- The cost-comparison table (trainer $60–100/session · AI coach $149/mo · generic app $17–29/mo). Concrete outside anchors. v1's version is lighter; this one is stronger.
- "You're 3 taps away — 1 Tap Start · 2 Download the app · 3 First session today." It tells people what happens after they pay. v1 does not, and that is a real gap.
- The escape hatch — "I don't want to pay anything, show me another option" — catches people who would otherwise close the tab. v1 has no equivalent, so its non-buyers leave with nothing.
- A named testimonial with a specific result (Diana, 38 — 18lbs in 12 weeks, no cardio) placed right at the decision point.
6 · How long has each funnel actually run?
First-traffic dates measured from GA4 daily sessions (window opened 1 Feb 2026). Promo windows shaded — read no number across them.
Conversion measured per visitor — so traffic size can't mislead
Raw visitor counts tell you nothing about which page works — v2-2 got more traffic simply because the quiz sent everyone there. This is buyers ÷ visitors, month by month. CHECKED
| Funnel | Month | Visitors | Buyers | Bought, per visitor |
|---|---|---|---|---|
2026-v2-2 | June | 1,240 | 31 | 2.5% |
| July | 1,597 | 25 | 1.6% | |
/2026/ Control | June | 444 | 10 | 2.3% |
| July | 255 | 5 | 2.0% | |
2026-quiz-checkout-v1 | Aug 1–6 | 304 | 7 | 2.3% |
2026-quiz-checkout-v1 10 Aug | Aug 1–10 | 607 | 14 | 2.3% |
2026-quiz-checkout-v1 every time we edited one — about 4,800 fake visits across 30 Jul–10 Aug, visits that can never buy anything.Leave them in and the new page looks like it converts at 0.4%. Take them out and it converts at 2.3%. Same buyers, same week — the only difference is whether you count crawlers as customers. Every figure for the new page has them removed. ✅ The crawler stopped 9 Aug with the retrofit's completion and is now excluded at source (verified 10 Aug PM on the daily series) — see §3.
Why February–May are not in this table. Those older checkouts are also used by pages outside the quiz (old email links, direct /register/ links). In April the Control's checkout took 20 sales while its page had only 94 visitors — a 21% "conversion rate" that is obviously not real. From June onward the quiz is the dominant source, so the rate is trustworthy. Only the months we can stand behind are shown.
7 · The switch, day by day — who still lands where
What this chart shows: how many people arrived at each funnel on each day, either side of the 31 July switch. One line per funnel. The dashed red line is the day the quiz root was repointed.
① The teal line jumps from zero on 31 Jul — that is the new $97 funnel switching on and taking over.
② The blue line collapses the same day — the old funnel losing the traffic the root used to send it. But it never reaches zero. It flattens out at 7–14 a day and stays there. That floor is the Meta "Variant 2" ads, which still point straight at it. That is the leak, made visible.
③ The orange line does not move at all — the Control was never fed by the root, so the switch simply did not touch it.
One housekeeping note: v1's line excludes ~1,500–2,400 visits from our own YouTube link-checking robot, which would otherwise make the new funnel look 8× busier than it really is. Real people only.
✅ What happened next — the four days after this chart ends re-pulled 10 Aug
The floor the chart shows was the leak. It is gone. Daily sessions, all sources, with the Meta column called out.
| Day | v2-2 total | of which Meta | Control total | What changed |
|---|---|---|---|---|
| 5 Aug | 12 | 10 | 41 | leak still open |
| 6 Aug | 28 | 25 | 45 | leak still open |
| 7 Aug page built | 7 | 5 | 40 | leak still open |
| 8 Aug | 4 | 0 | 34 | ✅ ads repointed — Meta to v2-2 stops dead |
| 9 Aug | 2 | 0 | 55 | ✅ holds at zero |
| 10 Aug partial | 0 | 0 | 11 | ✅ holds at zero |
8 · Post-switch traffic — version × source Google Analytics · 30 Jul–10 Aug · re-pulled 10 Aug
| Landed users → | 2026-quiz-checkout-v1 real people only | v2-2 | Control | v2-3 |
|---|---|---|---|---|
| Google (paid+organic) | 158 | 22 | 0 | 0 |
| Meta (fb·ig·threads) | 49 | 60 | 141 | 3 |
| Direct | 90 | 47 | 1 | 3 |
| Owned + tagged + other (tma·bwa·blog·email) | 310 | 28 | 21 | 4 |
| Real people, total | 607 | 157 | 163 | 10 |
| 🔴 Excluded: our YouTube-retrofit crawler on v1 — stopped 9 Aug, excluded at source | 3,028 | — | — | — |
Of the 937 real visitors in the 12 days since the switch, 330 (35%) hit a build with no $97 offer — 17% deliberately (Control) and 18% unintended (v2-2 + v2-3). CHECKED ✅ But that whole-window figure straddles the fix. On the current window (8–10 Aug) the unintended share is 1.6% — see the post-fix table in §3. Use 1.6% for any forward-looking decision; use 18% only when describing the period as a whole.
9 · The funnel, per version — each in its own honest window
Rates are of that version's landings. Each window is matched to its own sales: v1 = 31 Jul–6 Aug ex-crawler (the span of its first 7 sales; an 8th landed 7 Aug); v2-2 & Control = July, their last full month on the root. GA4 fires no purchase event on the old quiz builds — every paid count comes from Stripe.
| Step | v1 · $97 (30 Jul–10 Aug, ex-crawler) | v2-2 (+v2-3) (July) | Control (July) |
|---|---|---|---|
| Landed | 607 · 100% |
1,665 · 100% |
255 · 100% |
| Gave us their email | 117 · 19.3% |
306 · 18.4% |
52 · 20.4% |
| Reached the price screen | 108 · 17.8% |
296 · 17.8% |
48 · 18.8% |
| Actually paid (from Stripe) | 14 · 2.3% |
25 · 1.5% |
5 · 2.0% |
| Of those who saw the price, how many bought | 13.0% no trial — $97 charged on the spot · was 11.9% on 7 Aug |
~8% checkout-level inference only — this page's sales cannot be isolated (see §1) |
~10% checkout-level inference only — this page's sales cannot be isolated (see §1) |
Second-source check on v1's sales: GA4 recorded 6 purchase events / 3 users on
/payments/quiz-checkout/success in the first week — directionally confirming Stripe, but heavily under-counting. 🔴 Corrected 10 Aug: the in-quiz purchase event does not fire on v1 at all, so GA4 shows 0 purchases on every build. Stripe is the money truth; GA4 cannot corroborate v1 money. The corroboration that does hold is different and stronger — every one of the 14 sales carries quiz_variant=quiz_checkout_2026_v1 written by the build itself.10 · Money — sales, avg per sale, and what a "conversion" is actually worth Stripe both rails, 8 May–10 Aug · re-pulled 10 Aug
| Checkout | Paid sales | Cash | Average | What people bought |
|---|---|---|---|---|
| 2026-quiz-checkout-v1 1 – 10 Aug · attributable | 14 | $1,201.46 | $85.82 | 12 × $97 annual · 1 × $24.98 quarterly · 1 × $12.48 monthly (the last two are the 50%-off downsell) |
| The shared old checkout 1 Jun – 31 Jul · NOT attributable to any one page | 71 | $2,827.05 | $39.82 | 28 × $24.98 · 15 × trial→$24.97 · 10 × $78.50 · 5 × $12.48 · 4 × $49.97 · 3 × $157 · 6 other |
11 · Money per visitor — why this comparison has been removed
An earlier version of this page compared the three funnels on "money earned per visitor" — $1.96 vs $0.39 vs $2.28.
Those figures have been withdrawn and the chart deleted. They divided each checkout's takings by a specific page's visitors — but two of the three checkouts are shared between the pages, with emails and with direct links, so the top of that fraction never belonged to the bottom of it.
What replaces it: the measured tables in §1. The one revenue figure that survives is 2026-quiz-checkout-v1's $85.82 per sale across 14 sales, because Stripe stamps that build's name on every one of its sales. Restoring this comparison needs one line of code in each old build — pass the build name into the checkout, exactly as v1 already does.
12 · Did the buyers stay? — what can and cannot be said
| Question | Answer |
|---|---|
Have 2026-quiz-checkout-v1's buyers stayed? | All 14 are still active re-checked per subscription, 10 Aug — zero cancelled, zero refunded (checked on the charge, not the subscription), and none has set the plan to end at period end. They are 1–10 days old, so this still tells us only that nobody bailed immediately. A real answer needs the 60-day guarantee window to close — ~5 October — and a 90-day read around 1 November. |
Have 2026-v2-2's buyers stayed? | 🔴 Cannot be answered. We can measure the shared checkout's buyers, but not which of them came from this page. |
Have /2026/'s buyers stayed? | 🔴 Cannot be answered. Same reason. |
| What CAN we say about retention? | Across the shared old checkout as a whole, of the buyers who joined Feb–Apr and are now 4–6 months in, 50% are still paying (69 of 137) and each has paid $75.87 on average. That is a fact about the checkout, not about any one quiz page. |
13 · We tried to fingerprint the Control's sales. Here is what happened.
Aga's idea, 7 Aug: "this is a free trial for 7 days and then $157 — so the sales from this page would be $157. Search how many $157 sales we have had and see if Facebook and Analytics match." Exactly the right test. We ran it. CHECKED
Step 1 — the fingerprint is real, and very clean
22 sales in the last 200 days took a first payment of exactly $157.00. Every single one followed a 7-day free trial and paid on day 7 exactly. That is precisely the Control's offer shape, and no other price on the account behaves that way.
Step 2 — but they do not track the Control's traffic
| Month | $157 trial sales | Control visitors | Implied rate | Facebook's own count |
|---|---|---|---|---|
| March | 5 | 548 | 0.9% | 1 |
| April | 8 | 94 | 8.5% 🔴 | 0 |
| May | 3 | 138 | 2.2% | 2 |
| June | 1 | 444 | 0.2% | 2 |
| July | 2 | 255 | 0.8% | 0 |
| Mar–Jul total | 19 | 1,479 | — | 5 |
The rates swing between 0.2% and 8.5% with no pattern. In April, 8 sales came from 94 visitors — that page cannot produce that. In June it had its best traffic month (444 visitors) and produced one sale. If these were Control buyers, sales would rise and fall with its traffic. They don't.
And Facebook counts 5 purchases where Stripe shows 19. Three sources, three different pictures. The trial→$157 offer must also be reachable somewhere other than this page — most likely the same legacy /register/yearly3/ link, which is also inside the other two builds and in older emails.
🔴 Step 3 — the checkouts-v3 result, diagnosed properly (corrects the "117 attempts, 1 payment" first published here)
"117 attempts" was wrong — most of those were never people attempting anything. The full picture, layer by layer:
| Layer | Measured | What it means |
|---|---|---|
| Stripe subscriptions minted, 1 Jul – 6 Aug | 117 (117 distinct customers) | 2.2 subscriptions per actual visitor — the page mints Stripe records without a human action. The "117 attempts" framing was phantom records, not shoppers. |
| GA4 users who SAW any checkouts-v3 page, same window | 53 | |
| People who ever entered a card | 4 | All 4 later cancelled; 1 paid. All 58 recorded cancellations are cancellation_requested — API-initiated, not one payment failure. |
🔴 Subscriptions sitting active with NO card, NO paid invoice, NO discount | 5 | Five people appear to hold a live subscription we cannot bill. Worth Nic's attention regardless of everything else. |
Visitors to trial/12month — the path named as the Control's destination | 4 users in 5 weeks | And the Control's live bundle contains no checkouts-v3 URL at all — so it does not send anyone here, which also rules this page out as the source of the 22 $157 sales. |
Bottom line: the real funnel here is ~53 visitors → 4 cards → 1 payment in five weeks. The dominant problem is that almost nobody is sent to these pages; the second problem is the phantom minting; the third is the 5 unbillable actives. None of it changes the quiz-funnel decision — but items two and three are real defects.
📌 Task state — verified 10 Aug evening: both defects are one job, §19 open row 6 (NIC-92, Nic, ~2 h) — still OPEN and still minting: fresh checkout_version=v3 subscriptions appeared in Stripe on 7, 8 and 10 Aug, seen live in tonight's stamp-check pull. The fleet watches every new mint (attribution_stamp_check.py) but the fix is server-side Django — it genuinely needs Nic.
NP · "Should we bring back the no-paywall funnel?" — Aga's question, answered
Question to answer: "In the past we only ran FB ads, and we only had a funnel where there was NO paywall. People came through, we got them as leads, at the end they had to download the app to see their result. We converted them through the emails. That's it. Should we restart this funnel, knowing there is literally zero sales on Facebook ads? I saw some apps like Level successfully run it, and this is how this business survived the last 7 years."
🔴 First — a correction to this page's own §PA.5, because it answered a different question
§PA.5 analysed ad → app install: cold campaigns with an APP_INSTALLS objective, which is what bought 6,700 installs from India, Pakistan and Peru that produced zero buyers. That is not what Aga asked about, and using it to answer her would be wrong.
She is asking about ad → QUIZ → no paywall → "install the app to see your results" → email converts them. In her model the person spends five minutes answering questions and gives us their email before the app is ever mentioned. That is a completely different proposition from buying installs, and it deserves its own answer. This section is that answer. §PA.5 stands as written, for the question it actually addresses.
🔑 The single fact that decides this — the email is already captured BEFORE the paywall
Read off the live quiz build, screen by screen CHECKED — screen ids in index-yC00xO94.js, and the GA4 funnel confirms the same order:
| Order | Screen | What happens |
|---|---|---|
| … | P51Teaser | the results are teased |
| then | P52Capture | ✅ WE GET THE EMAIL HERE |
| then | R01…R09 | the results, the map, the forecast |
| last | R10Paywall | the $97 ask |
So removing the paywall would not win us a single extra lead. We already have the email before anyone sees a price.
NIC-97) is the live test of whether buyer-country leads land at her $1–2 bar. If they do, we take them in EITHER funnel — and the paywall-vs-no-paywall choice stays decided by one thing only: whether email out-converts the $97 screen (the 10 Sep read).So what would actually change — and what it would cost
The choice is not "leads vs no leads". It is what happens to a lead we already have. Verified on 1–10 Aug.
| Today — with the paywall | Aga's model — no paywall | |
|---|---|---|
| Leads captured | 112 | 112 — unchanged, capture is upstream |
| What they see next | the $97 offer | "install the app to see your results" |
| Sales, immediately | 15 · $1,298.46 | 0 — by design; the money comes later |
| Lead → sale | 13.4% | whatever email achieves |
| Value per lead | $11.59 | whatever email achieves |
The bar, stated exactly: to break even, email would have to convert 13.4% of those leads to paid at the web price — or 26.6% if they buy inside the app, where a sale nets us roughly half.
🔴 CORRECTED 10 Aug (Aga) — "email converts 0%" is NOT a measurement, and I used it as one
An earlier version of this section said: "would we give up a measured 12.5% today for a conversion engine producing 0%?" That was wrong, and Aga caught it: "We don't know how email converts. We haven't given it enough time. I just launched the evergreen sales sequence."
She is right, and the live systems confirm it:
| What we checked | What it says |
|---|---|
993 SALES SEQUENCE — the first and only email arc that asks for the sale | Went PAUSED → ACTIVE at 13:12 on 10 August. Today. 122 entered; its campaigns show sends of 0, 1, 2 and 8. It has barely sent an email. AC API — automation mdate + campaign send_amt |
| What quiz leads met before that | Only 750 TMA | Welcome Sequence After Quiz — whose ratified doctrine is activation-first, with NO offer, discount or deadline in any email, ever. |
So email was not failing to convert. It was never asking. Measuring "email produced 0 of 14 sales" over a window in which no email asked for a sale is measuring a machine that was switched off — blindness, not performance.
EV-2026-08-10-993-sales-sequence-armed) so the next analysis cannot repeat it: any email conversion figure dated before ~10 Sep 2026 is UNINTERPRETABLE and must not be used to argue for or against a funnel change.The honest framing, restated: the paywall's 13.4% is measured. Email's rate is not yet knowable. This is a comparison we cannot make yet — and the correct response is to run the clock, not to pick a side.
⚖️ Taking the case FOR it seriously — because three parts of it are true
Aga is right about the history, and the counter-arguments have to earn their place.
- It genuinely did sustain the business. That is not in dispute, and it is the strongest evidence in the room — a funnel that carried TMA for years deserves more than a ten-day dataset in reply.
- Level and others do run this shape successfully today. Free-to-install, results in the app, monetise over time is a proven consumer-app model, not a fringe idea.
- 🔴 Meta is currently producing zero sales through the paywall. For Meta traffic specifically, "12.5% of leads convert" is not the comparison — Meta's actual conversion through the paywall is 0%. Against zero, almost any working alternative is an improvement. This is the strongest single argument for the test and it should not be waved away.
🔴 What HAS changed since that funnel worked — two things, and they are not small
| Then | Now | Why it matters here |
|---|---|---|
| ✅ CORRECTED by Aga, 10 Aug: "P52Capture → results → download — that's the one we had before." So the old funnel showed the results on the web too. The two funnels are IDENTICAL up to the very last screen — same quiz, same email capture, same results. The only thing that ever differed is what the final screen asks for: "download the app" or "$97". That makes this a clean single-variable swap, and a far easier thing to test than a rebuild. | ||
| Email was the conversion engine | Email produces 0 of 14 sales; the welcome arc's own ratified doctrine is activation, not selling — no welcome email carries an offer | The engine has to be rebuilt before the funnel that depends on it, not after. |
| No proven paid offer | A $97 offer that converts 13% of price-views and has kept 15/15 buyers | The old funnel never had to beat this. Now it does. |
Also worth naming plainly: the app funnel it would pour into has a measured leak. Of 15,523 app-side contacts, 6,813 finished the survey and never did a single workout. Sending more people into that does not create members; it creates a bigger silent list.
✅ THE GROWTH FLEET'S VERDICT — NO right now, hybrid in 30 days, on one gate
Aga asked for the growth fleet to run this. It did, on the verified data above. tma-growth-lead, 10 Aug 2026
The fleet's verdict was: "you cannot restart a funnel whose conversion engine is broken." The old model required email to turn leads into buyers, and email was measured at 0%.
993 SALES SEQUENCE armed at 13:12 on the day of the analysis. So "the engine is broken" is not established — it is untested.What this does and does not change: the fleet's gate is exactly right and becomes the main recommendation — measure email properly, then decide. Its reasoning on the other three points (ATT, CPM inflation, the $97 asset) stands independently and does not rest on the 0%. But its headline "NO" should be read as "NOT YET — the test has not run", which is a materially different answer.
Why it worked then and may not now — three structural shifts
| Shift | What it does to this funnel |
|---|---|
| iOS 14 / ATT (April 2021) | Meta's OUTCOME_LEADS cannot see who buys — it optimises on lead cost, so it finds the cheapest people willing to fill a form. That is the wrong population. Evidence: zero Facebook click ids on any August Stripe sale, and $444 per purchase over 98 days against a ~$76 break-even. |
| CPM inflation | The old CAC maths survived at cheap CPMs. At current levels it needs email converting above ~2% to work at all. industry-wide; not knowable precisely for our account |
| We now have a $97 checkout that works | 13% of price-views buy, netting $85.82. The old funnel never had to beat this. Routing leads around it into the store (net ~£31.77) halves revenue per customer without doubling volume. |
🔑 The leaky bucket sits before the tap
🔴 The fleet's own skeptic flagged this and it must travel with the number: both rates are MODELLED from existing benchmarks, not measured on this audience, and reactivation cohorts typically underperform. Mail a segment of 500 first and read the click rate before touching all 6,813.
The ranked answer
| # | Lever | Cost | Evidence | Verdict |
|---|---|---|---|---|
| 1 | Activation emails to the 6,813 who never worked out queued 10 Aug PM: AGA-146 | $0 | MODELLED | DO THIS FIRST — awaiting your GO; the fleet drafts the email + builds the 500-contact test segment and shows you both before anything sends |
| 2 | Point 758 Nurture Sequence 2 (14,600 contacts, currently sending the quiz nothing) at the $97 quiz queued 10 Aug PM: AGA-147 — sister of AGA-129, one GO can cover both | $0 | MODELLED | Fix the tap before the bucket — awaiting your GO; ONE-slot change through the email fleet (Evolution-Law compliant), fleet executes end-to-end |
| 3 | Google Branded — keep running | ~£50/mo | CHECKED | Protect. Do not pause. Separate from §19 row 1 — the 3 equipment keywords in Competitive Non-Branded still need the axe. |
| 4 | Meta retargeting with the $97 offer — warm audiences, proven checkout the case is already built: the 3 PAUSED retargeting campaigns produced 10 of Meta's last 12 purchases on 4% of spend — tma-remarketing-playbook.netlify.app | $400–600/mo | MODELLED | Bench — after attribution confirms (§19 row 3) |
| 5 | The no-paywall restart — Aga's question | $500–865/mo | needs the email read first | 🟡 PARKED — not refuted, UNTESTED |
Rows 1–3 consume no budget headroom. All rows sit inside the ~$3,175/mo envelope (currently ~$2,310/mo).
The one metric that decides it
Target: ≥2% by 10 September 2026. Current: not yet measurable — the arc that asks for the sale armed on 10 Aug, so the clock starts now, not in the past. a zero before this date is blindness, not performance
🔴 Kill gate — below 1% on 10 Sep: abandon the lead-gen model. Paywall stays primary; effort goes to organic, SEO and warm retargeting.
✅ Scale gate — at or above 2%: spend $500 of the headroom on a single 30-day no-paywall test to the warm quiz-starter audience, routing to the WEB checkout not the store, with a 20% holdout so we can measure whether it actually caused anything.
⚖️ Where the fleet and this analysis disagree — your call, and here is the cost of each
One genuine disagreement, worth putting in front of you rather than smoothing over. What to do with the live Meta lead-gen spend right now.
| Position | The argument | What it costs |
|---|---|---|
| The fleet says: pause it now | $292 per 10 days into an email engine converting at 0%, with no Facebook click id ever reaching Stripe. Under ATT the campaign is structurally unable to optimise toward buyers — so a cleaner window will not fix what is a measurement-blindness problem. | Saves ~$875/mo immediately |
| This page said: hold 14 more days | The Variant 2 ads were only repointed to the right quiz on 8 August. Every prior day sent that traffic to a page with no $97 offer. Judging Meta on that window is judging it on our own routing mistake. | ~$410 to buy a clean read |
Both are defensible and the difference is about $410. The fleet's structural point is the stronger one — ATT means even a clean fortnight is unlikely to flip the result. The middle path, if you want one: halve the lead-gen budget rather than pausing outright, and let the two live campaigns run to 24 August on the cheaper half. That buys the clean read for about $200 and frees the rest for row 1 and row 2 above, which cost nothing and rank higher anyway. 📌 This is the same ruling as §19 open row 5 — one decision closes both surfaces. The options on the table: pause now (fleet) · hold flat to 24 Aug · the middle path above. Whichever you pick, Nic executes it inside the ~$3,175/mo envelope. AGA'S DECISION
14 · Lead quality — the "great low cost per lead" examined
Where every lead actually comes from — per variant, line by line CHECKED — 31 Jul – 7 Aug
Every email captured since the switch — 115 leads — split by which quiz build captured it, then by exact source. Bar lengths are share of that build's leads. The older builds carry less detail because almost all their traffic is two Meta campaigns.
2026-quiz-checkout-v1 — the $97 page
76 leads · 66% of all/2026/ Control — no $97
27 leads · 23%120230041112720728) is ~90% of everything here.2026-v2-2 — no $97
10 leads · 9%2026-v2-3 — the DEAD build
2 leads · 2%- 🔴 Email produced ONE lead in eight days — and it landed on the dead build via 967 TMA | Welcome Sequence After LEAD MAGNET E2. ~14,600 contacts sit in 758 Nurture Sequence 2 sending the quiz nothing — the case for §19 row 4 and queue row
AGA-147. - The TMA homepage alone (15, free) rivals all of Google paid (17, ~$300/wk).
- 32% of all leads are captured on builds that cannot sell the $97 — the lead-level twin of the traffic leak.
- 1 in 6 leads is untagged (30 of 115) — the same attribution gap that hides paid's true performance.
- YouTube shows only 3 referral leads — 🔴 and re-checked on 10 Aug it is still 3, from 3,028 crawler landings. The retrofit's own traffic is not producing leads.
🔴 The uncomfortable one, unchanged and now on more data: v1's 120 leads and 14 sales mean ~1 lead in 9 buys. The other 106 people gave us an email address and did not buy — and §0b shows email then sent them nothing that produced a single sale. That is the largest identified gap on this page.
| Campaign (ACTIVE ones) | Jul spend | Jul leads Meta / GA4 | cost per lead Meta / GA4 | Aug 1–6 spend · leads · purchases | Lead geography (28d, GA4) |
|---|---|---|---|---|---|
QLG | ADV+ OFF | MC | Control → /2026/ |
$203.52 | 80 / 52 | $2.54 / $3.91 | $70.96 · 37 · 0 | 🔴 Only 20% US — CA 24% · UK 22% · AU 15% · FR 9% · NZ 9% |
QLG | ADV+ ON | US | Variant 2 → /2026-v2-2/ |
$149.03 | 81 / — | $1.84 / $3.40* | $37.43 · 13 · 0 | v2-2 leads: 58% US |
| Google · 4 "Quiz Funnel" campaigns → v1 | $572 (wk31+32) | — / 24 | — / ~$24 | 1 buyer so far (utm google) | v1 leads: US 31% · IN 11% · AU 6% |
*window-matched 14d: $71.30 ÷ 21 GA4 meta-leads. Meta cost per lead use Meta's own sale credit; GA4 cost per lead use measured on-site leads — both shown, neither hidden.
- The Control's cost per lead is genuinely cheap — because its leads are multi-country. Only 1 in 5 is a US lead. They convert to annual sales at a trickle (5 in July, 0 so far in Aug), worth ≈ 2.9× July spend INFERRED — the old checkout does not record which ad sent the buyer. Since /2026/ is ~100% Facebook-fed, we credit its sales to Facebook..
- Cheap ≠ good by itself: Meta's own pixel counted 8 purchases across ALL QLG campaigns in 3 months (May 2 · Jun 3 · Jul 3 · Aug 0). 🔴 Re-pulled 10 Aug: August is still 0, on $286.90 of spend — and Stripe independently confirms zero Meta-sourced sales.
- None of v1's 14 buyers came from Meta. Their captured sources: tma ×6 · Google paid ×3 · untagged ×3 · tma blog ×1 · bwa blog ×1 — the $97 offer is being bought by owned-surface and brand-search visitors. See §0b for the full list.
- ✅ Variant 2's spend no longer lands on the old funnel — the ads were repointed on 8 Aug and now feed the quiz root. Its leads reach the $97 offer from here on.
15 · The Google Ads "conversion drop" — what the account itself shows
The claim was that conversion fell from 2.6% to 0% when 2026-quiz-checkout-v1 went live. This table is pulled straight from the Google Ads account (not Analytics, not a screenshot) — every Quiz Funnel campaign, week by week, in the account's own currency. CHECKED — Google Ads API, pulled 7 Aug
| Week of | Impressions | Clicks | Conversions | Conv. rate | Spend | Cost per conversion |
|---|---|---|---|---|---|---|
| 6 Jul old funnel (2026-v2-2 default) | 8,262 | 181 | 0 | 0.0% | £66 | — |
| 13 Jul | 29,588 | 331 | 2 | 0.6% | £183 | £91.49 |
| 20 Jul | 22,663 | 215 | 12 | 5.6% | £253 | £21.06 |
27 Jul — switch to 2026-quiz-checkout-v1 lands mid-week | 2,416 | 140 | 41 | 29.3% | £240 | £5.86 |
3 Aug — first full week on 2026-quiz-checkout-v1 | 2,004 | 105 | 38 | 36.2% | £183 | £4.81 |
The account shows the opposite of a drop. Conversions went 0 → 2 → 12 → 41 → 38 a week, conversion rate 0% → 36%, and cost per conversion collapsed from £91 to under £5. Honest notes on reading it:
- These "conversions" are mostly leads, not purchases — the account counts its lead and purchase actions together. Purchase-specific conversions also rose (0.24/day → 0.63/day), but at single digits.
- Why it ever read 0%: Google back-dates a conversion to the click that caused it, so a freshly-switched window looks empty and fills in later — and before the switch the old doorways stripped tracking, so the July weeks genuinely could not measure most of what they bought. The tracking was broken before; the switch fixed it.
- Impressions fell 10× at the switch (29,588 → 2,416). Correction from Nic: this is not a fault — the campaigns are budget-capped on purpose (his month-end projection: Meta ~$1,671 + Google ~$1,504 ≈ $3,175 total), and the destination change reset learning. Nothing to "un-limit"; the lever is reallocation inside the cap, not more budget.
- 🔴 One thing genuinely IS at zero, and was before the switch too:
Competitive Non-Branded | US— £206.74 in 30 days, no conversions in either era, and nothing to do with the landing page. Correction from Nic, confirmed at keyword level: the fix is NOT a campaign pause — £204 of the spend sits on 3 equipment-shopper keywords ("street workout equipment" £147 alone) while 59 of 66 keywords have never served. Google optimisation is keyword-level: bid the non-converters down to ~$0.10–0.20 or switch them off, review the list's intent (most of it is equipment-purchase, we sell a programme), and let the untested keywords get impressions.
Bottom line: the "2.6% → 0%" read was an artefact of timing and broken pre-switch tracking. By Google's own account data, 2026-quiz-checkout-v1 is the best-performing destination these campaigns have ever had.
🔴 UPDATED 10 Aug — and now we know what those conversions were
The point above stands: the destination change improved what Google can measure. But three days of extra data let us split the conversions by what was actually counted, and the picture is much less flattering. 1–10 Aug, from the account's own conversion_action breakdown: CHECKED
| Campaign | Spend | Reported conversions | …that are LEADS | …that are PURCHASES | Sales confirmed in Stripe |
|---|---|---|---|---|---|
| Branded | US | £16.27 | 12 | 9 | 4 | 3 ✅ |
| Competitors Branded | US | £113.15 | 20 | 20 | 0 | 0 |
| PMax | US | £48.51 | 48 | 46 | 3 | 🔴 0 |
| Non-Branded | US | £91.29 | 2 | 2 | 0 | 0 |
| Competitive Non-Branded | US | £89.89 | 0 | 0 | 0 | 0 |
| TOTAL | £359.11 | 82 | 77 (94%) | 7 | 3 |
94% of what the account calls a "conversion" is an email address, not a sale. So the celebrated "£4.81 per conversion" is £4.81 per lead. On real, Stripe-confirmed sales the true figure for the whole account is ≈£120 per sale — and for every campaign except Branded it is infinite, because they produced none.
🔴 And the flagged campaign has not been touched. Competitive Non-Branded | US spent another £89.89 between 1 and 10 Aug for 18 clicks and zero conversions of any kind — the same three equipment keywords ("street workout equipment" £37.45, "calisthenics training equipment" £28.28, "home calisthenics equipment" £12.47) still absorbing the budget. To-do #2 from 7 Aug is still open, and it has cost roughly £90 in the three days since.
16 · "Which version gets cheaper leads, and why — the bodies are the same?"
Aga's question, answered from Meta's own numbers, every live QLG campaign, May → 6 Aug.
Short answer: the page is not what is moving cost per lead — the country is. And "the Control gets cheaper leads" does not hold up.
| Month | Campaign — note the geo in the name | Page | Spend | Leads | cost per lead | Cheaper that month |
|---|---|---|---|---|---|---|
| May | ADV+ ON · US · Control | /2026/ | $69.56 | 37 | $1.88 | Control — same geo |
| May | ADV+ ON · US · Variant 2 | /2026-v2-2/ | $584.52 | 194 | $3.01 | |
| Jun | ADV+ ON · US · Control | /2026/ | $360.91 | 100 | $3.61 | Variant 2 — same geo, opposite result |
| Jun | ADV+ ON · US · Variant 2 | /2026-v2-2/ | $759.11 | 268 | $2.83 | |
| Jul | ADV+ OFF · MC · Control | /2026/ | $203.52 | 80 | $2.54 | Variant 2 — different geo, not comparable |
| Jul | ADV+ ON · US · Variant 2 | /2026-v2-2/ | $149.03 | 81 | $1.84 | |
| Aug 1–6 | ADV+ OFF · MC · Control | /2026/ | $71.37 | 39 | $1.83 | Control — different geo, not comparable |
| Aug 1–6 | ADV+ ON · US · Variant 2 | /2026-v2-2/ | $37.73 | 13 | $2.90 |
1 · The winner flips every single month. There are only two apples-to-apples comparisons on this board — May and June, both US vs US — and they contradict each other. Control wins May by 38%, Variant 2 wins June by 22%. That is noise, not a page effect. CHECKED — read from Facebook Ads directly
2 · Hold the page constant and change only the country, and cost per lead swings 10×. The same /2026-v2-2/ page, same month (June):
| Same page · same month · only the country differs | cost per lead |
|---|---|
| UK · Variant 2 | $2.06 |
| US · Variant 2 | $2.83 |
| FR · Variant 2 | $5.24 |
| CA · Variant 2 | $19.86 |
| AU/NZ · Variant 2 | $42.29 spend → 0 leads |
3 · The two campaigns still running are not a fair test of anything. Their names carry the confound: ADV+ OFF | MC | Control versus ADV+ ON | US | Variant 2. Different geography and different Advantage+ setting. GA4 confirms the traffic really is different people — Control leads are only 20% US (Canada 24% · UK 22% · Australia 15% · France 9%), while v2-2 leads are 58% US. The US is the most expensive ad market in the world; a multi-country campaign buys cheaper leads by definition.
So Nic's instinct is right — the landing-page copy is not plausibly driving cost per lead, and the data agrees. But the conclusion drawn from it needs correcting: there is no evidence the Control page produces cheaper leads. It produces cheaper countries. If we want the Control kept, the honest reason is cost and audience — it runs cheap, multi-country, and does not steal v1's traffic. Its checkout charges the FULL $157 after the trial (verified live) — not a discount. A fair test would be: same page, same country, two versions. That has never been run.
17 · The Google Ads decision — the premise needs correcting first
🔴 The claim "conversion rate dropped from 2.6% to 0%" is not what Google Ads' own account says. Pulled live from the Google Ads API on 7 Aug, the five Quiz Funnel campaigns:
| Google Ads, Quiz Funnel campaigns | Clicks | Conversions | Conv. rate | Per day | Spend |
|---|---|---|---|---|---|
| 1–29 Jul — old funnel (v2-2 by default) | 786 | 24 | 3.1% | 0.83 | £615 |
| 30 Jul–6 Aug — new funnel (v1 $97) | 169 | 64 | 37.9% | 8.00 | £279 |
Conversions per day went up roughly 10×, not to zero. Purchase-specific conversions also rose: 7 across 29 days before (0.24/day) vs 5 across 8 days after (0.63/day). CHECKED — read from the Google Ads account itself
Why the earlier read looked like 0%, and it is a fair mistake to have made:
- Google back-dates a conversion to the click that caused it. Look at a freshly-switched window too early and it reads empty — the conversions land in those same rows days later. This is the single likeliest cause.
- One campaign genuinely IS at 0% — and was before the switch too.
Competitive Non-Branded | US: 12 clicks/0 conversions in July, 13 clicks/0 conversions since. It has burned £207 across both periods for nothing. That is a real problem, and it is not caused by the new landing page. - 2.6% is very close to the old PMax (2.7%) and Competitors-Branded (2.9%) rates — so the "before" number looks like a single-campaign or single-view read rather than the account total.
🔑 The switch did not break Google tracking — it fixed it. Before: The Movement Athlete (web) generate_lead recorded 1 conversion in 29 days. After: 26 in 8 days. That matches what GA4 shows — in weeks 28–30, 47–160 paid ad-clicks per week were arriving as 0–5 measurable sessions because the old doorways stripped the tracking parameters.
What that means for the traffic-split proposal
Nic's proposed split was 40% v1 · 40% v2-2 · 20% /2026/, built on the premise that v1 is converting at 0% and v2-2 historically hit 2.6–13%. With the premise corrected, the case for splitting mostly dissolves — and there is a second, harder problem with splitting at all:
| Decision | Recommendation | Why |
|---|---|---|
| Google Ads destination | Keep 100% on v1 for the full 14 days (to ~21 Aug). Do not add v2-2 as a second destination URL. | The premise for splitting is refuted, and splitting is what makes the 14-day read unreadable. This is Aga's ruling — "really give it a go for 14 days" — and the data supports it. |
/2026/ Control on Google | Do not add it. Agreed with Aga — retire it from the Google plan. | It has never run on Google. Starting a third arm now costs the read on the first two and buys ~4 clicks/day of nothing. |
/2026-v2-2/ as fallback | Keep the page live, send it nothing. Zero cost, instant rollback. | Exactly what Nic already set up. A page that serves but is unadvertised is a free option. |
Competitive Non-Branded | US | Pause it. The one genuinely broken thing on the account. | £207 across two periods, 0 conversions in either. Unrelated to the landing page — it was failing before the switch too. That is ~£90/mo back immediately. |
/2026/ Control on Meta | Leave it running — but for the right reason. | Not "cheap leads" (unproven — see the section above). It runs cheap, draws a different multi-country audience, and its checkout charges the FULL $157 after the trial (verified live); its revenue cannot be isolated (§1). It is a different audience/geo, so it is not stealing v1's traffic. |
Meta Variant 2 ads | ✅ DONE — repointed to the root. | Verified live on the creatives 10 Aug; Meta traffic to /2026-v2-2/ went to zero on 8 Aug and has stayed there. |
The kill condition, so the 14 days mean something: by 21 Aug, v1 should show ≥12 purchases at roughly current traffic. Below ~6 and the $97-charged-today model is genuinely underperforming and v2-2 comes back as the Google destination. Between 6 and 12, extend rather than switch — that band is still inside the noise.
Do not treat this as the read being finished. The 14-day window still runs to 21 Aug, and the questions it was set up to answer — does the rate hold, do the buyers stay past the refund window — are open until 5 Oct. What the gate settles is narrower and worth saying plainly: the offer works well enough that reverting is off the table. The remaining decision is where the traffic to scale it comes from, and §0b says it is not currently coming from paid.
18 · The money plan — scaling 2026-quiz-checkout-v1 to $10K/month
Everything below is built ONLY from measured numbers on this page, with every assumption labelled. Where a figure is a projection it says so. CHECKED — pulled 7 Aug
Which version should we scale? — 2026-quiz-checkout-v1. On all the data, it is not close.
| Evidence | 2026-quiz-checkout-v1 | 2026-v2-2 | /2026/ Control |
|---|---|---|---|
| Revenue we can actually SEE and steer by | ✅ every sale stamped | 🔴 invisible | 🔴 invisible |
| Money per sale | $85.82 measured 14 sales | top card $24.98 | $157 possible, unmeasurable |
| Gets people to the price | 19.2% | 22.4% | 19.1% |
| Google Ads on it (account data, §15) | 36% conv · £4.81/conv | 0–5.6% · £21–91 | never ran |
| Plan sold | a year, cash up front | mostly $24.98 quarterly | full-price annual after trial |
Why: it is the only version whose results we can measure while spending (non-negotiable for paid scaling) · highest provable money per sale · sells the plan class that retains · and Google's own account says it converts an order of magnitude better than the old destination ever did. The other two stay for what they are — v2-2 as a rollback, the Control as a cheap different-audience arm — but every scaling dollar goes to 2026-quiz-checkout-v1.
1 · Who is actually buying the $97 on 2026-quiz-checkout-v1 — the first eight buyers, profiled
| Date | Country (from the card) | Quiz path they chose | Came from | Paid via | Paid |
|---|---|---|---|---|---|
| 1 Aug | 🇺🇸 US | AGE_STRONG | Link | $97 | |
| 1 Aug | 🇬🇧 GB | AGE_STRONG | (untagged) | Link | $97 |
| 2 Aug | 🇫🇷 FR | STRENGTH_BUILDER | tma (our site) | Link | $12.48 |
| 3 Aug | 🇺🇸 US | JUST_START | tma (our site) | card | $97 |
| 4 Aug | 🇦🇹 AT | MUSCLE_BUILDER | (untagged) | Google Pay | $97 |
| 6 Aug | 🇸🇪 SE | FAT_BURNER | tma (our site) | card | $97 |
| 6 Aug | 🇺🇸 US | MUSCLE_BUILDER | bwa (our blog) | Link | $97 |
| 7 Aug new | 🇺🇸 US | JUST_START | (untagged) | Apple Pay | $97 |
Four things these eight people tell us (with the usual warning — it is eight people):
- 4 of 8 came from our OWN surfaces (TMA site ×3, BWA blog ×1) and 3 of 8 are untagged. Only one carries a paid source. Owned traffic is carrying this funnel; ads are not, yet. Held at 14: 9 of 14 owned, 3 paid, 3 untagged
- Half the buyers are not American (GB, FR, AT, SE vs US ×4). The $97-upfront offer sells internationally — relevant because international ad clicks cost a half to a fifth of US ones.
- 6 of 8 paid by one-tap wallet (Stripe Link ×4, Google Pay, Apple Pay). Fast payment matters at this price; anything that adds card-typing friction will cost sales.
- No single persona dominates — AGE_STRONG ×2, MUSCLE_BUILDER ×2, JUST_START ×2, FAT_BURNER, STRENGTH_BUILDER. Too early to target ads by persona.
2 · Paid ads — what we spend, what it buys, where it leaks
Every live campaign, its cost and its measurable output. Google figures from the Google Ads API; Meta from the Marketing API. CHECKED 7 Aug 🔄 For the current 1–10 Aug window with Stripe-confirmed sales per campaign, see §0b — that table is the one to act on.
| Campaign (last 14 days) | Spend | Clicks | Leads | Cost/lead | Verdict |
|---|---|---|---|---|---|
| Google · Quiz Funnel | PMax | US | $88 | 143 | ~18 | ~$4.90 | ✅ The workhorse — most volume, cheapest Google leads |
| Google · Competitors Branded | US | $170 | 86 | ~2 | ~$85 | ⚠️ Expensive; carried the one paid-attributed $97 sale |
| Google · Non-Branded | US | $134 | 30 | ~1 | ~$134 | ⚠️ Thin and dear |
| Google · Competitive Non-Branded | US | $234 | 23 | 0 | — | 🔴 £207/30d, zero output — but the fix is KEYWORD surgery, not a pause (Nic's correction): 3 equipment keywords eat 98% of spend, 59 of 66 never served. See §19 row 1. |
| Meta · ADV+ ON | US | Variant 2 | $78 | — | 40 | $1.95 | 🔴 Cheap leads, pointed at the page with no $97 |
| Meta · ADV+ OFF | MC | Control | $153 | — | 72 | $2.13 | 🟡 Cheap multi-country leads; deliberate arm |
| Meta cost per lead, by country (last 14d — same ads, same creative) | CPL |
|---|---|
| 🇨🇦 Canada | $1.35 |
| 🇦🇺 Australia | $1.67 |
| 🇳🇿 New Zealand | $1.75 |
| 🇬🇧 UK · 🇺🇸 US (Variant 2) | $1.95 |
| 🇺🇸 US (Control) | $2.94 |
The shape of the leak, as it stood on 7 Aug: Meta produces the cheapest leads we buy ($1.35–$2.00) — and 90% of its quiz traffic landed on pages that cannot sell the $97 (14 Meta visitors reached the new page; 123 landed on the old ones).
/2026/ — OVERTAKEN 11 Aug (NIC-106): repointed to the quiz root on Aga's ruling. The control arm ENDS., and that is the larger half of Meta's spend ($142.30 vs $75.46).🔴 The harder problem: with the leak closed, Meta still produced zero sales. $286.90 and 206 leads across 1–10 Aug, no traceable purchase, confirmed by both Stripe and Meta's own pixel. "The ads point at the wrong page" was a real defect but it was not the whole explanation — and we should stop using it as one. The open question is now whether Meta's $1.35–2.00 leads convert at all, and the repointing means the next two weeks will finally answer it.
3 · The unit economics — what a sale is worth and what we may pay for one
| Value | Basis | |
|---|---|---|
| Average sale, measured | $86.56 | 15 provable sales, re-pulled 11 Aug CHECKED |
| − Stripe fees ~3% | −$2.59 | standard rate ESTIMATE |
| − refunds, all-time rate 9.88% | −$8.28 | company-wide historical ESTIMATE |
| Net value of one sale, year one | ≈ $76 | no renewal value assumed — first $157 renewal is Jul 2027 |
| Break-even cost per sale | $76 | spend more and we lose money on every sale |
| Healthy target (3:1) | $25 | the discipline number for scaling |
| Where we are — blended, last week | $69.60 | $487 all paid ÷ 7 sales in the matched week — thin but under water-line. 🔴 This number was always generous: it divides ALL paid spend by ALL sales, including the ones ads did not cause. |
| 🔄 Where we are — cost per provably paid sale, 1–10 Aug | ≈$257 | 🔴 The honest number. ≈$770 of Google + Meta spend ÷ 3 Stripe-confirmed paid sales. 3.4× break-even. CHECKED 10 Aug |
| …of which: Google Branded alone | ≈£5.42 | ✅ £16.27 ÷ 3 sales. The only paid line under the water-line — and it is brand search, which harvests demand rather than creating it (§0b). |
| …of which: everything else paid | ∞ | 🔴 ≈$750 across 4 Google campaigns and 3 Meta campaigns, 1,569 clicks, zero traceable sales. |
4 · The path to $10K/month — three stages, each gated
The arithmetic first ($85.82/sale, ~2.3% of real visitors buy — measured):
| Monthly revenue | Sales needed | Real visitors needed @2.3% | vs today |
|---|---|---|---|
| Today's run-rate 10 Aug | ~42 | ~1,820/mo (14 in 10 days) | — |
| $5,000 | 58 | ~2,550 | 1.4× |
| $10,000 | 117 | ~5,100 | 2.8× |
⚠️ At 14 sales the conversion rate's honest range is roughly 1.3–3.9% (narrower than the 1.1–4.7% quoted at 8 sales, but still wide), so "5,100 visitors" could really mean 3,000–9,000. The stages below exist precisely so we find out cheaply.
| Stage | What we do | Cost | Expected effect (labelled) | Gate to the next stage |
|---|---|---|---|---|
| A · now → 21 Aug free moves only | ✅ Repoint the 2 Variant-2 ads to the quiz root — DONE 8 Aug · ✅ NIC-97 built + MOF paused — DONE 11 Aug · ✅ keyword surgery on Competitive Non-Branded — DONE 11 Aug (5 bid down + 3 paused); read 13 Aug, set-level negatives staged behind it as row 1c · 🔴 point one weekly email at the quiz root — still open (9 of 14 buyers already come from owned surfaces) | $0 frees ~$117/wk | +40–60 real visitors/wk from Meta ESTIMATE · email adds the cheapest qualified traffic we own · run-rate → ~$3.5–4.5K/mo ESTIMATE | ✅ MET EARLY — 14 provable sales by 10 Aug against a bar of ≥12 by 21 Aug |
| B · Sep fix measurement, then scale what measures | Nic ships the two one-liners (build-name into checkout; paid source onto v1 sales) · then raise budget ONLY on campaigns with measured cost/sale under $76 — starting with Meta international (CPL $1.35–1.95) now that its traffic reaches the $97 page | $0 extra by default — reallocation inside Nic's ~$3,175/mo envelope. Any increase beyond it is an Aga budget decision ESTIMATE | If Meta's cheap leads convert at even half the owned rate, cost/sale lands $40–70 ESTIMATE — the stage exists to measure this · target ~$6–8K/mo | Cost/sale < $76 for two consecutive weeks, measured not blended |
| C · Oct only if A and B hold | Scale the winning channel toward ~5,200 visitors/mo · YouTube read becomes clean after 9 Aug (869 descriptions already link the quiz root) · re-check refunds when the first 60-day window closes 5 Oct | budget follows the measured CAC | $10K/mo requires ~116 sales — reachable at ~2× stage-B traffic if the gates held PROJECTION | Net-of-refund cost/sale still < $76 after 5 Oct |
The prioritised to-do list lives in §19 — one list, one order, no duplicates.
19 · Recommendations & the to-do list
One ruling per version, then one list in priority order. This section supersedes anything above that reads differently.
The ruling, per version — no ambiguity
| Version | Ruling | In plain words |
|---|---|---|
2026-quiz-checkout-v1 | SCALE THIS ONE. | All traffic and all future budget go here. The only measurable version, highest provable money per sale ($86.56 across 15 sales), best Google performance on the account's own data, and it sells the plan that retains. Do not touch the $97 during the read. |
2026-v2-2 | ✅ TRAFFIC IS OFF — done 8 Aug. Page stays up, dark, as the rollback. | Nothing to do here. Its Meta ads were repointed to the quiz root on 8 August and its Meta traffic has read 0 every day since — verified on the live creatives and in GA4. The page deliberately stays live so we can switch back within minutes if the read fails. Dark = live but unfed, and it is already dark. Its 301/sunset decision comes only after ~30 v1 buyers + the 5 Oct refund read. |
/2026/ Control | KEEP on Meta only. Never on Google. | Kept because it is cheap, multi-country, and does not take v1's traffic — not because of its revenue (unmeasurable, §1) and not for "cheap leads" (a geography effect, §16). Its checkout charges the full $157 after the trial. Revisit once v1 has ~30 buyers. |
THE TO-DO LIST — ONE list, open first, done at the very bottom every status re-checked against the live system, 10 Aug evening
Nothing below is marked done because someone said so. Each closed item was re-checked against the artefact itself — the live ad creative, the Ads account, the Stripe record, the served page. Layout per Aga (10 Aug): open items first in priority order; everything ✅ done lives in its own table at the very bottom. Two parallel sessions each carried a task table on 10 Aug; they were merged into this ONE list the same evening — if a task table appears anywhere else on this page or beyond, it is stale by definition. Rows are identified by their queue id where one exists; every human-owned row carries the default-executor check.
🔴 OPEN — in priority order, highest $-per-minute first
| # | What | Why it ranks here · status verified 10 Aug evening | Who — and why them | Effort |
|---|---|---|---|---|
| 1 closed 11 Aug | ✅ DONE 11 Aug — both halves of what this row asked for. The row said "bid the equipment keywords to ~£0.10–0.20 or pause them". Change log: 14:36:02 bid changes on 5 keywords in this campaign (+2 on Branded); 20:15:17–24 the 3 still-spending keywords paused. Six of 66 paused in total. | Executed by Nic in one sitting. Its effect is not yet readable — the changes landed part-way through 11 Aug, so that day's £9.11 mostly predates them. First clean day 12 Aug; read 13 Aug. an earlier version of this row read 11 Aug's spend as the verdict — withdrawn, it was our error | Nic — delivered | ✅ |
| 1c new, 11 Aug AGA-157 | 🟡 Staged behind the 13 Aug read: set-level negatives on Competitive Non-Branded | US — add hardware negatives (bar · bars · parallettes · equipment · gear · machine · station · vest · kit · setup · buy · for sale · cheap), or pause and rebuild on programme intent. Do NOT run this before 13 Aug — it would make Nic's round unreadable | Why it is staged rather than dropped: 30-day totals are £235.89 · 32 clicks · 0 conversions · 0 all-conversions; 60 still-enabled keywords are all equipment phrases; the campaign carries 0 negatives and 0 shared negative lists; and the money has been buying parallettes, gornation pull up bar, dip bar for home. The 7 Aug round is the one datapoint we can already read — it had four clean days and spend continued.🔴 One correction to this section's standing reasoning either way. It said "never a campaign pause — 59 of 66 keywords have never served, and zero impressions is untested, not failed." Right in general; it does not transfer here, because the untested 60 are the same equipment-intent class that has already spent £235.89 for nothing. That is an argument about the SET, and it is true regardless of how 13 Aug reads. | 2 min from Aga (the GO), then the fleet writes it by API — we hold the Google Ads OAuth credential and this is a criterion-level write. It is live spend in Nic's account, so it ships only on her word. | 2 min |
| 1a 11 Aug NIC-106✅ AGA APPROVED | 🔴 NIC — REPOINT QLG | ADV+ OFF | MC | Control TO THE QUIZ ROOT. 5 min.Its four live ads still send people to https://quiz.themovementathlete.com/2026/ — the old assessment. Change the destination on all four to https://quiz.themovementathlete.com/ (the ROOT, which serves the quiz that sells the $97).The four ads: · ad set General / T2 / FM 30-54 US/Canada (£9.62/day) — 120253084939360728, 120253084939340728· ad set General / T2 / FM 30-54 AUS, FR, UK SING, NZ (£9.62/day) — 120245527336080728, 120245527336070728🔴 Change the destination and NOTHING else — leave url_tags, budgets, geo and the objective exactly as they are, and do not pause the campaign.Optional while you are in there: drop IL from the second ad set's geo (Israel is the only country here off our buyer list) · pause ad set General / T2 / FM 22-65 North America — it holds £7.00/day but all six of its ads are paused, so it cannot deliver and spent £0 last week. Hygiene only, frees nothing. | Why: the campaign brings 129 leads/30d at £2.43 — the cheapest lead source on Meta (57 in the last 7 days), and right now every one of them lands on the page that does not sell. Repointing keeps all of them and sends them to the quiz that shows the $97 and enrols them into 993 SALES SEQUENCE. Costs nothing either way — £45.48/day account-wide before and after.✅ RULED BY AGA, 11 Aug: "yes Nic needs to know about this — one Meta campaign is still sending people to the old quiz page instead of the new one that sells the $97." That settles a contradiction in our own notes: on 7 Aug we recorded these ads as the deliberate A/B control arm and wrote "nobody re-flags them later as a defect"; her 10 Aug NIC-97 ruling said re-aim them. The 10 Aug ruling stands and the 7 Aug note is OVERTAKEN — §3 and §6 are stamped accordingly so this is not raised a third time.Verified live 11 Aug: all four CTA links read /2026/ (Meta Marketing API, read off each creative); the root 302s to /2026-quiz-checkout-v1/ — two different builds, confirmed in a real mobile browser. | Nic, Ads Manager 5 min + 2 min validation Validation: open each of the 4 ads' preview, confirm the click lands on the quiz that shows the $97, then confirm the campaign is still ACTIVE with budgets unchanged. | 5 min |
| 1h 11 Aug NIC-102✅ AGA APPROVED | 🍎 NIC — SHIP THE STAGED iOS KEYWORD FIELD. The live App Store keyword field wastes ~86 of its 100 characters; the replacement is already written and staged. Push it in App Store Connect. | 100 characters is the entire keyword surface iOS search has to work with, and we are using ~14 of them. It is the cheapest ASO change available and it needs no build, no release and no review cycle beyond the metadata pass. | Nic, App Store Connect 10 min | 10 min |
| 1i 11 Aug NIC-105✅ AGA APPROVED | 🔴 NIC — REMOVE THE THIRD-PARTY GTM CONTAINER FROM THE CHECKOUT. GTM-PPGJXHH is hardcoded in app/apps/templates/base.html and is not ours — Aga checked all 7 GTM accounts and 12 containers on 11 Aug, no match. It loads on every legacy /register/ checkout, the page that holds a live Stripe key, alongside our own GTM-TKGDZR9, and we cannot see what it fires.Delete two blocks in that one file: the <script> at lines ~12–18 (head) and the <noscript> iframe at ~52–55 (body). 🔴 Leave GTM-TKGDZR9 untouched.While you are in there: also delete the dead Universal Analytics tag UA-104067894-1 (line ~20) — UA was sunset by Google in Jul 2023 and sends nothing. Nothing else in the file changes. | An unidentified third party is running arbitrary tag code on a page that takes card details. We cannot audit it, we did not put it there, and it has no business being on a checkout. Removing it is a two-block deletion in one template. | Nic, tma-backend 15 min Validation: after deploy, load /register/yearly3/ and confirm only GTM-TKGDZR9 loads. | 15 min |
| 1j new, 12 Aug NIC-108 | 🔴 NIC — STOP PMax BUYING PLACEMENTS THAT HAVE NEVER CONVERTED. Quiz Funnel | PMax | US | Sales OBJ is the only LIVE campaign breaching the settled "Category Search ONLY — no Demand Gen, no YouTube" ruling, and the breach is measurable. Last 30 days, by network: CONTENT/Display £28.37 (39,067 impressions, 210 clicks, 0 conv) · SEARCH_PARTNERS £28.17 (48 clicks, 0 conv) · YOUTUBE £3.28 (17 clicks, 0 conv) · DISCOVER £1.03 (5 clicks, 0 conv). That is £60.85 — 40% of the campaign's £150.32 — buying nothing.🔴 DO NOT PAUSE THE CAMPAIGN. Its SEARCH arm took £93.07 and returned 78 conversions at £0.53/click, the lowest cost-per-conversion in the account. Keep the Search delivery, stop the rest. PMax gives no clean channel toggle, so pick one and tell us which: (1) account-level Excluded Placements + Content Suitability to cut Display/YouTube inventory · (2) campaign-level negative keywords + turn OFF Final URL expansion · (3) the clean structural answer — retire PMax and move its £5/day into a standard Search campaign, which is what the ruling actually asks for. | Why it ranks here: it is the only live breach of a settled ruling, and it is 40% of a campaign's budget returning zero. Reallocation, not new spend. Two caveats that travel with it, not decorations: (a) 86% of this account's "conversions" are not sales — they are leads, carts and app installs; 30d the account booked 59 Leads-GTM + 46 generate_lead + 4 add_to_cart + 4 Android installs against only 14 Purchase + 4 Lifetime-promo — so PMax's 78 are leads; (b) PMax self-attributes generously (78 conversions from 174 clicks is a 45% rate), so treat its Search figure as directional, not proven. Measured live 12 Aug — Google Ads API v25, customer 6909169207, campaign × ad_network_type breakdown. | Nic, Google Ads 15 min + 5 min validation Validation: 3 days later, open the campaign's placement report and confirm Display/YouTube impressions are ~0 while Search impressions and conversions hold. | 15 min |
| 1l new, 12 Aug NIC-110✅ AGA APPROVED | 💷 NIC — CUT Quiz Funnel | Competitors Branded | US FROM £10/DAY TO £5/DAY, and put the £5 into PMax. Do it after NIC-108 (else the money just enlarges PMax's leak) and after 13 Aug (else it collides with AGA-157's read).The change: set this campaign's daily budget to £5.00 and PMax's to £10.00. Nothing else — leave keywords, bidding, geo and objective exactly as they are. | Why: it is the biggest line in the account (£314.77/30d = 32% of Google spend, at its cap) and returns 24 conversions at £1.58/click — but that is £13.12 per conversion against PMax Search's £1.19, Branded's £1.74 and Meta's cheapest lead line at £2.43. 86% of this account's conversions are not sales, and all 3 Stripe-verified Google sales in the 1–10 Aug audit came from Branded, none from this line. Why cut rather than pause: competitor-brand search is real in-market intent, unlike a cold lead-form fill — the lead types are not interchangeable at any price, and switching a line off before knowing what its output feeds is the mistake the MC Control round already cost us. Why PMax gets the £5 and Branded does not: Branded spent £40.00 in 30 days against a £6/day cap — it is volume-bound, not budget-bound, so money moved there will not spend. PMax is the only line actually at its cap. Kill condition, pre-registered: at the 10 Sep gate, zero Stripe-verified sales AND cost-per-conversion still above £10 → pause it. Reassigned from AGA-104 by Aga, 12 Aug. If there is a reason to hold £10/day that these numbers cannot see, say so and it stands. | Nic, Google Ads 10 min + 5 min validation Validation: confirm the two budgets read £5.00 and £10.00 and the account's enabled total is unchanged at £36.00/day. | 10 min |
| 1m 11 Aug AGA-150 | 🚨 AGA — RESCUE THE ROTTING FACEBOOK LEADS. 10 min in Ads Manager, and a 90-day fuse is burning. The MOF - Lead Form campaign's ActiveCampaign sync died 1 Dec 2025. Every muscle-guide form lead since — roughly 1,100 — never reached email. Meta deletes stored leads after 90 days, so only ~264 (May–Aug) are still retrievable, and that number falls every day.Two things: (1) Ads Manager → Leads Center → export the "Build Muscle With 40+ - higher intent" form's leads as CSV and drop it in the repo; (2) reconnect the ActiveCampaign ↔ Facebook Lead Ads integration (AC → Apps) so new leads flow again. | Why it is here and not lower: it is the only row on this page with an expiry date on the asset itself — leads are deleted, not merely delayed. The campaign spent ~£78/month into the broken pipe. Verified live 11 Aug: Meta Graph API form 886259157389251 reads leads_count=264 against 2,417 lifetime form leads; AC tag 63570 holds 250 contacts with the newest created 1 Dec 2025; sibling tags 63569/63572 are equally stale; the job register and tools/ were grepped and no alternate pipeline catches these leads.Why not the fleet: the leads were pulled via Graph API with the page token and returned 403 Forbidden — leads_retrieval is not granted to any credential in the repo. Both the Leads Center export and the AC reconnect are UI-only surfaces. | Aga, Ads Manager + AC 10 min | 10 min |
| 1k new, 12 Aug NIC-109 | 🍎 NIC — WHICH REPORT DID THE $2,277 COME FROM? The spend side reconciles ($404 against Apple's own billing of $377.32 for 1 Jun–31 Jul, $408.24 for 12 Jun–12 Aug). The revenue does not — AppsFlyer credits Apple Search Ads with $1,098.82 from 6 buyers over the same window, roughly half. Name the report and the exact date range (Apple Search Ads dashboard? AppsFlyer? RevenueCat? App Store Connect?) and whether its attribution includes view-through. | Why it matters: the 5.6× is being quoted as a reason to scale, and it is a blend of a 7.5× branded line with a 0.0× non-brand line on a revenue figure we cannot reproduce. Likely benign: Apple's own window counts view-through installs — 165 tap vs 305 view installs over 12 Jun–12 Aug — and a view-inclusive basis roughly doubling credited revenue is exactly this gap's shape. If so both numbers are right and we just label the basis. The underlying export is not reachable from the repo, so this is the one part the analysis cannot close itself. Full working in §PA.11. | Nic 5 min | 5 min |
| 1b | ✅ DONE 11 Aug — NIC-97 is live. QLG | ADV+ ON | MC | Variant 2 | Quiz Root, ACTIVE, OUTCOME_LEADS, £9.62/day, destination quiz root (read off the live creative), url_tags kept, buyer geos NL·CA·SE·GB·AU·FR·DE with US carried on the sibling US Variant 2 line. MOF - Lead Form PAUSED 20:14:17 UTC. | Ruled 10 Aug ("I need this"), built 11 Aug 20:19 UTC. ✅ Tripwire (D3) does not fire — it was live inside the 14 Aug deadline, so cold Meta is not pause-forced. Open consequence moved to row 1a. | Nic — delivered | ✅ |
| 1c | The 758 Nurture Sequence 2 broadcast — engaged ~4,400 → the quiz root, SEND BY 14 AUG. Design per the in-session finding: short, ONE link, zero price mention — the email creates the SESSION, the paywall closes (13/15 buyers pay within ~20 min of arriving) | The fleet's #1 lever and Aga's one thing today: the GO. $0, the only lever that can produce sessions this week, aimed at the pool that just surfaced a proven buyer (the 3.6-year Nurture2 Completed win-back). Gates first: collision check + segment_guard.py PASS attached to the approval; measured on its own UTM, purchases not clicks. Kill: <0.5% session rate → pool is cold, no repeat.🔴 THE PRICING RECONCILIATION (Aga's question, answered): most of this pool saw the OLD pricing — the 7-day trial into $24.97/mo (= $300/yr) or the $157 annual. $97-once-for-12-months is a BETTER deal than either — the copy can say so plainly ("less than four months of the old monthly price"). The living proof it reconciles: the Nurture2 Completed win-back saw the old-price era and still paid $97. The ONE genuine conflict: 993 SALES SEQUENCE is the 50%-off arc (verified in its own subjects: "half price, and six months on us") — so per the J4 law (two prices, never simultaneously to the same person) the broadcast segment EXCLUDES everyone currently in 993 SALES SEQUENCE (~122 people) and anyone recently sent a 50% offer. The exclusion is part of the segment-guard PASS. Re-takers are a feature, not a problem: the framing is "see where you are NOW" — levels change in months, and the quiz regenerates results for the same contact. | Aga GO → fleet builds & sends = AGA-129, now spec'd | GO + copy approval |
| 1d | Verify 993 SALES SEQUENCE enrolment integrity by 12 Aug — every entrant receives the arc correctly (plaintext twins, the enter-block on 750 TMA | Welcome Sequence After Quiz) | An enrolment defect today silently corrupts the 10 Sep gate — the decision the whole paid plan waits on. | Fleet (upload-integrity gate) | by 12 Aug |
| 1e | The Play uninstall read by 12 Aug — extend play_install_reports.py to the uninstall columns; answers Aga's "half install and delete" from the store's own data | D5: if D30 retention is very low, the 10 Sep gate is mailing people who are already gone, and its read changes. Android only; iOS parked until this lands. | Fleet | ~30 min |
| 1f | Nic reports NIC-97's ACTUAL first-week CPL by 17 Aug — one number from the Ads UI | D2: the linear break-even's denominator is currently assumed ($2–5). Real CPL ≤$2.50 → build the $2-lead creative test; >$4 → hold. Unlocks or defers the whole lead-scale decision. | Nic | 5 min |
| 1g′ | Build the $97 warm retargeting test (NIC-99) — ONE campaign, OUTCOME_SALES, ~$5/day, warm audiences only (site visitors 30d · quiz-starters · abandoners, payers excluded via the ever-paid set), the $97 straight, quiz root. Brief: the remarketing playbook at its corrected economics (£67.48 full-life, £2.41 withdrawn) | Sequenced deliberately AFTER the 45-min sitting (1 + 1b) and the CPL report (1f) — one ad set, one afternoon, not a project. Retargeting is the only Meta spend that has ever sold here and the attribution blocker is cleared (every quiz sale is stamped). Judged ONLY on stamped sales, 4-week read. Kicker: it pre-warms the audiences, pixel events and creative for the November BF push, which is the same build at scale. | Nic | ~1 afternoon |
| 1g | Organic CTA lift on the homepage hero + top 2 blog pages by 17 Aug — assessment-promise framing → quiz root | 9 of 15 sales came from these surfaces, unowned and unoptimised. Expect +15–25% organic quiz-starts MODELLED. | Fleet | by 17 Aug |
| 1h | BUILD Black Friday — offer DECIDED (70% off lifetime · yearly · quarterly; Aga, 11 Aug): write the October warming arc, stage the BF pages + v3 checkout links, draft the retargeting brief | The largest cash event of the year ($20–35K, n=2) and the payout of the whole lead bank. The offer was never an open decision — the work is build, and it can be ready EARLY (target: build done by 15 Sep, warming starts 1 Oct). | Fleet · Aga approves copy | build by 15 Sep |
| 2 | Add a click event to the two app-store links already in the quiz | The cheapest measurement on this page, and it gates a real decision. The paywall_exit door is coded and tagged but no click event exists, so take-up is unmeasurable. This is the Step 0 of §NP — it tells us whether anyone wants the app door before we build anything. | Fleet (FLEET-51) — a GTM job, not a Nic one; the fleet holds publish on the container and has shipped tags end-to-end before. | 45 min |
| 3 | /register/ checkout record which quiz sent the buyer | The row died on three facts: ① the one-liner can only tag FUTURE /register/ sales — it cannot analyse old data, and the renewals (45% of web cash) are untaggable forever regardless; ② everything with a future (the $97 quiz, v3) is already 100% stamped; ③ what still flows through /register/ is ~1 trial per 10 days and shrinking. The one thing knowingly given up: the Control's revenue stays permanently unattributable, so its ~30-buyer review will be decided on cost + geography alone — acceptable, since its case never rested on revenue. Revisit ONLY if the Control is kept as a long-term paid arm. | Genuinely Nic, ~1 h. Fleet check: the write lives in the live Django/React checkout code — the fleet has no path to those repos (no RN source locally; the Django clone is a stale 2 Apr copy with no deploy). Verified, not assumed. | ~1 h |
| 4 | Let 993 SALES SEQUENCE run — and protect it. (The 758 broadcast is now its own row above.) | 🔴 Corrected 10 Aug: email has not failed to convert — until today no arc asked. 993 SALES SEQUENCE armed at 13:12 on 10 Aug. 🔑 This is the gating measurement for §NP — the no-paywall model IS an email-conversion model, so nothing about it can be decided until 993 SALES SEQUENCE has run. Read it ~10 Sep. | Email fleet → Aga | runs itself |
| 5 | Decide the cold-Meta question — options: pause now (fleet's position) · hold spend flat to 24 Aug · the middle path (halve lead-gen, §NP) | The Variant 2 ads were only repointed on 8 Aug, so the coming fortnight is the first clean test Meta has ever had — but ATT means even a clean fortnight may not flip the result (§NP lays out both positions and their cost, ~$410 apart). 📌 One ruling closes both this row and §NP's decision card. Whichever you pick stays inside the ~$3,175/mo envelope; Nic executes. | Aga (with Nic). Genuinely hers: a budget judgement, not an optimisation. | decision |
| 6 | checkouts-v3 (NIC-92): stop the phantom mint and resolve the 5 unbillable "active" subs | 🔴 Re-verified 10 Aug evening: still minting. Fresh checkout_version=v3 subscriptions appeared in Stripe on 7, 8 and 10 Aug, seen live in tonight's stamp-check pull. 27 minted 1–10 Aug; 8 trialing, the rest expired incomplete; no v3 sale took money in August. Real hygiene, but not the funnel question. | Genuinely Nic, ~2 h. Fleet check: the mint is server-side in the live Django app — same access gap as row 3. The fleet watches every new mint but cannot change the server. | ~2 h |
| 7 | Reallocate inside the existing budget — never raise it. Money freed from the equipment keywords flows to what actually converts. | 🔴 Blocked behind rows 1 and 5. On Stripe-confirmed sales the only paid line that works is Branded (£16.27 → 3 sales) — but do NOT simply pour budget into brand search; read the trap note below first. | Nic, after Aga rules row 5. Row 1 frees the money, row 5 decides the shape, this row moves it. | 15 min |
| 8 | 5 Oct: the refund-window read on the first buyers → re-confirm net cost per sale < $76 | ⏳ SCHEDULED. The 60-day guarantee closes then; it is the number that decides whether any real budget increase is justified. All 15 buyers active, zero refunds so far. | Fleet | 30 min |
| 9 | Retire 2026-v2-2 and 2026-v2-3 once the new quiz has ~30 buyers | They now take almost no traffic (4 and 2 sessions in three days) but still absorb attention in every analysis. Not yet — they are the rollback while the new quiz is 10 days old. | Nic | later |
✅ DONE — verified, at the very bottom per Aga 10 Aug: finished items live down here, out of the way
| What | How we know | When |
|---|---|---|
| Repoint the 2 Meta "Variant 2" ads to the quiz root | Both creatives read quiz.themovementathlete.com/ on the Marketing API. Corroborated independently: Meta sessions to /2026-v2-2/ went to 0 on 8 Aug and have stayed there. The unintended leak fell 15% → 1.6%. | 8 Aug |
| The 14-day sales gate on the new quiz | Bar was ≥12 provable sales by 21 Aug. 14 banked by 10 Aug — met 11 days early. v2-2 does not come back. | 10 Aug |
The attribution stamp on the quiz rail (NIC-91, first half) | 6 of 6 post-deploy quiz sales carry checkout_version, quiz_variant and the full UTM set on the PaymentIntent — the write that made the sale-provenance analysis possible. The legacy rail still carries nothing; that remaining half is open row 3. | verified 10 Aug |
| Stop the YouTube crawler contaminating v1's numbers (was the morning's open #4/#5) | ✅ Both halves done by the fleet, 10 Aug evening. (1) The retrofit IS finished and the crawler DID stop — the "still running" read was a lump-window average. GA4 daily series: 1,084 → 1,134 → 937 → 1,120 → 483 across 3–8 Aug, then 0 on 9 Aug, 1 on 10 Aug; retrofit completed 9 Aug 09:15 (fleet ledger, 869 videos). (2) Excluded at source: both funnel_metrics_pull.py and pull_quiz_versions.py now subtract campaign yt_description inside the FIXED, closed 3–8 Aug window — date-bounded, self-retiring, can never suppress a genuine future YouTube click. v1 clean traffic 30 Jul–10 Aug: 993 real sessions of 5,758 raw; the cockpit funnel re-ran clean (7d Visitors 845, monotonic OK). | 10 Aug PM |
2026-v2-2 or the Control remains unknowable (§1); no Control-vs-v1 winner exists. The statistical bar for a real verdict is ~200 quiz starters per arm and v1 has ~607. Two things did firm up: the 2.3% visitor→buyer rate held across a doubled sample, and "paid ads produced almost nothing" is now supported by two independent sources rather than one. Everything else is still a direction of travel with gates.20 · What we found broken while measuring — and fixed
🔴 The cockpit's web funnel had been measuring the abandoned variant since 31 July
tools/finance/funnel_metrics_pull.py line 31 read QUIZ = "/2026-v2-2/s/" — hardcoded to one version. The day the root moved to the $97 build, that constant kept pointing at the funnel everyone had stopped sending traffic to. It never errored; it just quietly returned a smaller, plausible number every morning.
| Cockpit "Funnel 1 · Web", 7-day window | Was showing | Actually true | Error |
|---|---|---|---|
| Visitors | 145 | 2,663 | 18× under |
| Quiz started | 49 | 881 | 18× under |
| Lead (email) | 26 | 171 | 6.6× under |
| Paywall | 19 | 124 | 6.5× under |
Fixed: the screen is now matched by pagePath suffix (ENDS_WITH "/s/<Screen>") so every version counts and any future version is picked up automatically — and a new per-version split (landing_by_version, paywall_by_version) is written into data_funnel_metrics.json so a version's collapse can never hide inside a total again. Suffix-matching also de-duplicates a person who touched two builds; summing per-version would have double-counted them.
The lesson worth keeping: a constant naming one variant of a multi-variant system is a time bomb. It keeps returning a believable number after the world moves — which is exactly why nobody noticed for a week.
⚠️ One thing we deliberately did NOT fix — read the Visitors number with this in mind
The YouTube description retrofit (UTM campaign yt_description) makes YouTube fetch every link as it is edited. That puts ~1,500–2,400 crawler landings on the v1 path — real GA4 sessions that never become leads. They are not filtered out of the daily cockpit job, on purpose: the retrofit finishes ~8–9 Aug, and a source-exclusion baked into a daily job would outlive the contamination and then start suppressing genuine YouTube traffic. The per-version split makes the skew visible instead of hiding it.
Every v1 figure in this document already excludes it — that is what "real visitors" means throughout. The cockpit's headline number does not.
✅ RESOLVED 10 Aug evening — both halves. The morning read ("did not stop, ~160/day on 8–10 Aug") was a lump-window average, now retracted: the GA4 daily series shows the crawler died with the retrofit — 483 sessions on 8 Aug, 0 on 9 Aug, 1 on 10 Aug — and the retrofit completed 9 Aug 09:15 (869 videos, fleet ledger). With the window closed and bounded (3–8 Aug), the exclusion the original author rightly refused as a permanent filter became safe as a dated one: both funnel_metrics_pull.py and pull_quiz_versions.py now subtract campaign yt_description inside that fixed window only — self-retiring, so genuine future YouTube clicks are never suppressed. The cockpit funnel re-ran clean the same evening (7d Visitors 845, monotonic OK). To-do #4 is closed.
📋 Corrections pushed into Nic's brief the same day
- The card's own P0 pill and closing paragraph still framed AGA-56 as an open decision — it was ruled and dispatched on 6 Aug. Both now read as ruled (repoint), with the three shipped pieces marked done and only the Variant 2 ad destinations left.
- The sale count read "5 × $97 since 31 Jul"; the current count (7 Aug) is 8 paid / $691.48.
- The framing "both Meta arms are landing paid traffic on the pre-$97 offer" was half wrong — the
/2026/Control arm was deliberate at the time, and calling it a defect risked switching off a funnel we were keeping on purpose. — OVERTAKEN 11 Aug (NIC-106): repointed to the quiz root on Aga's ruling. The control arm ENDS. The lesson still stands for anything else marked deliberate: read the note before flagging. - Nic's own note contained an internal contradiction ("I redirect /2026-v2-2/ → v1" vs "still live, not redirected"). Both halves are recorded, with the wire showing which one is true — no one has to remember which version of the story was right.
21 · Verification ledger — every headline number, its source, its status
Built for the CEO read on 7 Aug 2026 · every figure re-pulled from source again on 10 Aug 2026. Anything that could not be verified is marked so, not softened. Rows changed by the 10 Aug re-pull are marked re-pulled.
| Claim | Value | Source and how it was checked | Status |
|---|---|---|---|
| VERIFIED — measured directly, twice where possible | |||
Sales on 2026-quiz-checkout-v1 re-pulled | 14 · $1,201.46 10 Aug | Stripe subscriptions carrying quiz_variant=quiz_checkout_2026_v1, both rails. The 7 Aug figure (8 · $691.48) reproduced exactly when the same window was re-run — the count grew because sales grew, not because the method changed. | ✅ CHECKED ×2 |
| Average per sale, that build | $85.82 | Same 14 subscriptions, first payment each. Re-pulled 10 Aug. | ✅ CHECKED |
| Visitors per build | 308 / 2,831 / 700 | GA4 property 177471782, filtered by landing page and hostname, robot sources excluded. Re-pulled 7 Aug — figures moved 1–6 users from the first pull as GA4 settled. | ✅ CHECKED |
| Email capture and price-screen rates | 19–22% | GA4 generate_lead and R10Paywall per build, same filters. | ✅ CHECKED |
| Meta "Variant 2" ads repointed to the quiz root re-pulled | 2 ads | Meta Marketing API — the creatives' own destination URLs, read live 10 Aug. Corroborated independently by GA4: Meta sessions to /2026-v2-2/ = 0 from 8 Aug. | ✅ CHECKED ×2 |
| Google Ads conversion rate 3.1% → 37.9% | 24 → 64 | Google Ads API, metrics.conversions by campaign, both periods. | ✅ CHECKED |
| May money was the Memorial promo | 34 of 40 | Stripe, day by day: 34 sales between 22–31 May, all at exactly $78.50, against 138 page visitors all month. | ✅ CHECKED |
| All three builds share the same old checkouts | 6 URLs | Fetched all three live JS bundles and searched them. Identical list, identical counts. | ✅ CHECKED |
| The Control charges full price after its trial | $157 | 🔴 Route corrected 10 Aug: the Control links /register/yearly3 (7 refs in its live bundle; 0 checkouts-v3 refs). The full-price $157 is confirmed by the Stripe price of the trial it actually creates — a $157.00/year subscription with a 7-day trial, created 10 Aug. | ✅ CHECKED |
| Buyer countries + wallets (the first 7 v1 sales) | US×3 · GB · FR · AT · SE | Stripe charges — the card's own issuing country; wallet from payment_method_details. | ✅ CHECKED |
| Source of every one of the 14 v1 sales NEW 10 Aug | 11 of 14 tagged | Stripe subscription metadata, merged across subscription / payment-intent / charge, recording which object carried each field. 3 sales carry a Google gclid; 8 carry an internal or owned-surface UTM; 3 carry nothing. tools/finance/quiz_sales_provenance.py. | ✅ CHECKED |
| Paid ads produced 3 sales; Meta produced 0 NEW 10 Aug | 3 · $218.98 | Two independent sources agree. (1) Stripe: exactly 3 August sales carry a paid click ID, all gclid, all brand terms. (2) Meta Marketing API: purchase action = 0 on all three live Meta campaigns for 1–10 Aug. Google's own Branded purchase count (3) matches Stripe exactly. | ✅ CHECKED ×2 |
| Google's other 4 claimed purchases (3 PMax, 1 Branded) | 4 | Reported by the Ads account; no Stripe sale carries a matching Google click ID. Likely view-through / cross-device credit or the codeless "/success" page trigger. | 🔴 NOT CORROBORATED |
| Live paid spend and status, both platforms NEW 10 Aug | $286.90 + £359.11 | Google Ads API campaign + Meta Marketing API /campaigns and /insights, 1–10 Aug. Status read off each campaign object, not inferred from spend. | ✅ CHECKED |
| Meta objective split — 28% of spend produced 82% of purchases NEW 10 Aug | 37 of 45 | Meta Marketing API, 90d, grouped by each campaign's own objective field. Total reconciles: $6,415.92 / 45 claimed purchases. | ✅ CHECKED |
| App-install campaigns → 67–69% from never-buying countries NEW 10 Aug | 9,330 installs | Google Play's own install reports (stats/installs, by country, Jun–Aug) cross-read against the Play earnings reports' buyer countries. Not an ad platform's claim. tools/finance/play_install_reports.py. | ✅ CHECKED ×2 |
| Play NET per transaction vs web quiz per sale | £31.77 vs $85.82 | Play July NET £730.65 ÷ 23 Charge rows; web from the 14 stamped v1 sales. No lifetime SKU appears in any Play month, so the July $297 promo is cleanly excluded. | ✅ CHECKED |
| The paywall-exit app-download path exists and is tagged NEW 10 Aug | 2 links | Read from the live bundle index-yC00xO94.js (1.34 MB, canary-checked on S01Landing): ?ct=paywall_exit and referrer=utm_source=quiz&utm_medium=paywall_exit. | ✅ CHECKED |
| …but its take-up is not measurable | — | All 10 GA4 event names on the quiz host enumerated for 1–10 Aug — no click or outbound event of any kind. Play acquisition/ reports stop at 2021-06. | 🔴 NOT KNOWABLE |
| YouTube crawler on v1 — stopped 9 Aug · excluded at source corrected 10 Aug PM | 4,828 sessions | GA4, campaign yt_description, 30 Jul – 10 Aug; 3 leads, 0 sales. 🔴 The morning line "still ~160/day on 8–10 Aug" is RETRACTED — it averaged a lump window. The daily series reads 483 (8 Aug) → 0 (9 Aug) → 1 (10 Aug), matching the retrofit's completion 9 Aug 09:15. Both cockpit pullers now subtract the campaign inside the fixed 3–8 Aug window; v1 clean = 993 sessions of 5,758. | ✅ CHECKED ×2 |
Competitive Non-Branded still spending with zero output re-pulled | £89.89 | Google Ads API, 1–10 Aug: 18 clicks, 0 conversions, same 3 equipment keywords. To-do #2 not actioned. | ✅ CHECKED |
| New-sale vs renewal basis corrected NEW 10 Aug | 224, not 421 | A sale counts as new only if the subscription was created in the window. The first run of the provenance tool deduplicated on in-window charges only, which counted long-standing members' renewals as new sales. Corrected before publication; v1 figures unaffected. | ✅ SELF-CAUGHT |
| Meta cost per lead by country | $1.35–$2.94 | Meta Marketing API, country breakdown, last 14d. | ✅ CHECKED |
| checkouts-v3 diagnosis (53 visitors · 4 cards · 1 payment · 5 unbillable actives) | — | Stripe subscriptions × GA4 page users, same window; statuses and payment methods read per subscription. | ✅ CHECKED |
| Traffic split: 90% of sessions vs 65% of real people re-pulled | — | Same GA4 window (30 Jul–10 Aug) counted both ways; the gap is entirely the YouTube crawler. | ✅ CHECKED |
| Lead sources, line by line re-pulled | 120/27/38/4 189 leads | GA4 generate_lead × source/medium/campaign per build, 30 Jul – 10 Aug (the 7 Aug read was 115 leads over the shorter window). Email still = 1. | ✅ CHECKED |
| Competitive Non-Branded keyword level (7 of 66 served · £204 of £207 on 3 equipment keywords) | 66 kw | Google Ads API keyword_view, last 30d — pulled after Nic's correction; his 5-of-66 read matched (7 by our window). | ✅ CHECKED |
| Shared old checkout, promo-free months | 71 · $2,827.05 | Stripe, 1 Jun – 31 Jul, every paid subscription on those prices. | ✅ CHECKED |
| ESTIMATED — stated as such, never presented as measured | |||
| Crawler visits removed from the new build re-pulled | 4,828 | GA4 sessions from youtube/(not set)/(cross-network) during the description retrofit. The boundary between robot and real is a judgement call. | 🟡 ESTIMATE |
| Share of visitors missing the $97 offer re-pulled | 35% whole window · 20.9% now | Visitors summed across traffic-source groups, so small double-counting is possible. The "now" figure is 8–10 Aug, after the Meta repoint; only 1.6 points of it is unintended. | 🟡 ESTIMATE |
| 🔴 CANNOT BE VERIFIED — withdrawn from this document | |||
Revenue, average sale or retention for 2026-v2-2 or /2026/ separately | — | Their checkout is shared with each other, with 2026-v2-3, with promo emails and with direct links. Stripe stores nothing that identifies the source. No amount of analysis fixes this — it needs a one-line code change in each build. | 🔴 NOT KNOWABLE |
Whether 2026-quiz-checkout-v1's buyers will stay | — | They are 1–10 days old; all 14 active, zero refunds. The 60-day guarantee window closes ~5 Oct; a 90-day read comes ~1 Nov. | 🔴 TOO EARLY |
| Which funnel converts best | — | 14 sales on one side, unattributable sales on the other. No statistically meaningful comparison exists yet. | 🔴 NOT YET |
| Whether paid ads caused the 3 brand-search sales | — | They carry a Google click ID, so the click is certain. Whether those people would have found us anyway — by typing the URL, or clicking the organic result below the ad — cannot be determined without a holdout test. Reported as "created by a paid click", never as incremental revenue. | 🔴 NOT KNOWABLE |
To re-run every figure: the commands are in the next section. GA4 numbers move by ~1% for 48 hours as data settles, so a re-run days later will differ slightly — that is normal, not an error.
22 · Provenance — where every number comes from
| Claim family | Source + command | Verification |
|---|---|---|
| Traffic, funnels, timeline, geo, campaigns | GA4 property 177471782 · python3 tools/quiz-analytics/pull_quiz_versions.py 2026-02-01|2026-07-30 yesterday + ad-hoc month/geo/day pulls (same filters) | CHECKED dimension-filtered (never top-N), hostname-locked, totalUsers; crawler exclusion evidenced by per-source day query |
| Sales, avg per sale, price points, and the SOURCE of every sale | Both Stripe rails · python3 tools/finance/quiz_sales_provenance.py --since 2026-05-08 NEW 10 Aug — one row per sale with its checkout_version, quiz_variant, full UTM set, click IDs, and which Stripe object carried each field. Also --csv / --json. | CHECKED TWICE Paid-sale finding corroborated against the Meta Marketing API (0 purchases) and the Google Ads API (Branded purchases = 3, matching Stripe exactly) · older /register/ checkouts still carry no source NOT KNOWABLE |
| Paid ads — spend, status, objective, installs (§PA) | Meta Marketing API v19.0 — /campaigns (status, objective, daily_budget) + /insights by campaign and by day · Google Ads API v21 searchStream · Google Play bulk-reports bucket — tools/finance/play_install_reports.py (installs by country) and play_earnings_reports.py (NET + buyer country) | CHECKED ad-platform install claims cross-read against the store's own install reports · objective split reconciles to the account total · 🔴 two traps encoded in the new tool: the live package is com.themovementathlete.app (the legacy one returns ~9 installs and looks plausible), and several report families are UTF-16 |
| Google Ads spend, keywords, search terms, conversion actions | Google Ads API v21 searchStream — campaign, keyword_view, search_term_view, and segments.conversion_action_name, 1–10 Aug, account currency GBP | CHECKED conversions split by action so leads and purchases are never summed as one number |
| v1 buyers alive + persona/paths | Per-subscription and per-charge Stripe reads over all 14 buyers (status, cancel_at_period_end, amount_refunded) · tools/ops/customer360.py for the joined profile | CHECKED refunds read on the charge, not the subscription — a refunded charge on an active subscription would otherwise be invisible |
| Ad destinations, spend, Meta cost per lead/purchases | Meta Marketing API v19.0 — campaigns, ACTIVE ads' creatives (object_story_spec links), monthly insights | CHECKED destination URLs read off the live creatives 7 Aug |
| Redirect map | curl -sI on root + all four version paths, re-run 10 Aug | CHECKED root 302→v1 · v2-2 / 2026 / v2-3 / v1 all 200 no-redirect — unchanged from 7 Aug |
| Google weekly cost | GA4 advertiser metrics (Google Ads link), campaign-scoped | CHECKED cost cannot be host-filtered — campaign names are quiz-only |
| Retention | Stripe — every payment made by buyers who joined Feb–Apr, summed per buyer | CHECKED Measured on the shared old checkout as a whole (137 buyers, $75.87 each, 50% still paying). Cannot be split by quiz page — see §1. |
Known limits, stated (revised 10 Aug): the older /register/ checkout still keeps no record of which funnel sent the buyer — so there we work it out from the plan bought, and mostly we cannot. v1 landing counts exclude the YouTube-retrofit crawler (yt_description, 4,828 sessions over 30 Jul–10 Aug) — 🔴 the earlier note that "crawler runs end ~8 Aug" was wrong; it is still running, so YouTube's real contribution remains unreadable. GA4 fires no purchase event on any quiz build, so it cannot corroborate quiz money at all — Stripe is the sole source. All Stripe figures are net of refunds (there are none on v1) and exclude renewals on subscriptions created before the window. July lifetime-sale cash sits on key2, which is pulled here but never touches these checkouts.
QUIZ_CHECKOUT_CONVERSION_BRIEF_2026-08-03.md · CHECKOUT_ATTRIBUTION_CODE_AUDIT_GROUND_TRUTH_2026-08-03.md ·
re-run everything: commands in §22.