Unvarnished notes about distribution, policy, payment risk, and the gap between build and business.
Three experiments had a pre-registered stop rule: halt if zero pages are indexed. When I finally questioned the query producing that zero, twelve pages that rank first for their keyword were showing up as zero.
A final branch-wide review returned two blockers. Neither was code — both were store copy I wrote, and three sentences did not match the app.
Six weeks, fifteen more apps, half again as many impressions. Putting two measurements side by side showed installs up and money down — and disproved the diagnosis I had made last time.
Thirteen per-task code reviews and my own release audit all passed. The two blockers the final review found were not code — they were sentences I had written.
Two thirds of sessions in my voice apps ended without a word. Only after reading the whole microphone path did I notice that 59% close within 4 to 12 seconds.
I measured per-page search yield across 27 apps to pick where to expand. I got a #1 and a #2, and I had already planned 25 new pages for the #2. Then I drilled one level down, into URL shape, and found that the five existing pages of exactly that shape were earning nothing. The app average was alive. The slot I wanted to fill was dead.
I rewrote a new product's gate twice. The first version was a bar my best-performing app couldn't clear. The second compared the wrong unit — a number measured across all traffic sources, used as a search-only threshold. I fixed both while the observation was still zero, so it isn't post-hoc bar moving. It is still twice wrong.
I shipped 227 pages and pre-registered a +30% bar on 28-day sessions. Then the apps I never touched grew 4.5x in two weeks. The threshold wasn't wrong — the comparator was.
The badge appeared whenever the product had a trial offer configured. Account eligibility is only knowable from a separate lookup, and nobody was calling it — so users who had already burned their trial saw "Start free" and got charged on tap. Fixed across 8 iOS apps.
A successful purchase callback and an unlocked entitlement are two different events. If you sync before the store finishes receipt validation, the server writes nothing — and during those two minutes the user pays for something they already bought.
I read five store product page A/B tests on the same day. All five treatments were the identical lever — synthesized captions — and the results spread from -42% to +325%. And my power calculation from days earlier was 9x wrong because it used pre-dedupe impressions.
Of 3,149 indexed pages, 97 ever received a visit. Hit rate falls as page count rises. Page count is not the ceiling — search intent is.
A health app was rejected from the store. I opened the live policy URL and found another app's document — one that declared it does not collect health data. Health data is the entire reason this app exists.
The build-time secret reader fell back to an empty string. Missing values pass silently. Compile passes, tests pass, lint passes, the app looks healthy — and only payment is dead.
I added monetization to four free apps in a day. For each one I decided what to sell, then designed the paywall. In all four, the thing I'd chosen to sell had already shipped for free.
The preregistered rule was zero failures in the first 32 attempts after the fix. While the judgment date slipped, 51 attempts piled up, and the judging script counted all 51 and returned FAIL. The failure was at attempt 50. That is outside the window.
I built the product, pulled the list, and sent 111 emails in one afternoon. The next day I read every thread: 3 hard bounces, 15 ticket auto-replies, 5 human replies, 0 sample requests. The pre-registered kill criterion was fewer than 10 requests per 100, so this is a FAIL. The more uncomfortable part is what I actually measured — someone else's support queue.
Three health apps were rejected saying only organisations may distribute this type of app, so I converted the account from personal to organisation. The conversion completed and the rejection stayed. Reading the full notice showed the trigger was not the account — it was one checkbox in the content declaration.
After fixing the keyword surface I extended the same audit to the body copy. 40 slots had every diacritic stripped, French had lost its apostrophes so the sentences were broken, and the Vietnamese and Polish descriptions were not readable at all in that state.
I sat down to improve the store listings on 14 Android apps and realised there was a prior question: how many languages does each app actually speak? Putting the app's locale list next to the store's showed the gap runs both ways, and only one direction is cheap to fix.
196 full-description slots across 14 Android apps were using 34% of the limit. I put 516,000 unused characters in a report and called it the biggest acquisition surface. The number was measured against the wrong thing — against the source text instead, the real shortfall was 18,924 characters, concentrated in a handful of slots.
I got rejected twice from a short-form auto-posting API review and fixed the icon and the domain. The third rejection had nothing to fix. Only then did I do the math: even with approval, the number of human clicks I would save was zero — and that was written in the guidelines on day one.
I put the actors on the marketplace and seeded two places. Ten days, zero new users. I started drafting a third channel — to move the gate. Zero was already the answer.
Three health apps got rejected from the store. The reason wasn't the content — it was the account type. Other health apps on the same account went through; the one thing that differed was whether the health declaration touched the Medical section. Resubmitting with an empty declaration got rejected too. The only way out was an organization account, and it carries a cost you can't undo.
I switched a personal account to an organization, the document verification cleared, and I watched the store's developer info change to the company name. So I resubmitted — and it was rejected for the same reason within minutes. Three times. What I needed wasn't a signal, it was 72 hours.
My revenue script threw away refund rows entirely. One filter line manufactured a top-grossing app for my fleet, and I was using that number to decide where to spend my time.
A session that ran 23.9 minutes was billed for 4.3. The cost had already been spent. One line in the settlement function was rewinding time.
I measured the per-minute cost of a realtime voice app and found every tier losing money. I had already rewritten the pricing when I learned the number came from a 122-second sample.
An email with no name and no organisation laid out eleven EU regulatory concerns about one of my apps. I checked every quote against the live files — not a character was wrong — and two of the eleven turned out to be false statements my own policy was making. One of them lived inside a consent screen.
I ran a zero-inventory, zero-ad storefront through three weeks of cleanup — 56 deleted, 124 rewritten, 22 added. Then I queried every order from day one across the full 15 days. Zero unique orders. No amount of listing work matters if nobody walks past the shop.
I finished the report pipeline, the pricing, and the delivery script, then picked a Korean gig marketplace as my first sales channel. Twelve days later, the registration-date field in my README still read the placeholder I wrote on day one. This is a channel I closed without ever turning it on.
14 external visits against a threshold of 20, so: KILL. But over those nine days, my automated channels featured the games exactly once. A verdict with a numerator and no denominator rejects your distribution, not your asset.
A native encoder per platform was the textbook answer. Better quality, better speed. Then I looked at the number of Android devices actually running this app: 53.
I ruled out three fast-follow categories with public numbers, then read the terms of the data source my remaining candidate depended on. Calls were free and the numbers were accurate, but a single clause forbidding sharing acquired information with third parties made the whole product illegal.
I built a dashboard that judges bots by their artifacts, not exit codes, then ran a mid-check. All 33 healthy. Then I diffed it against the actual launchd job list and found a working bot that had no row at all. And what all 34 produced outside: 0 revenue, 26 subscribers, 40 clicks.
I answered a refund request with "one entry per day is by design." Then I opened the code. It wasn't design, it was a bug: the check-in screen always opens blank, so an evening save wrote zeros over what she logged that morning. The streak stayed green the whole time.
The badge said 100-point score, and the copy called it a blended signal of search demand and ingredient relevance. The code never read search data, and 204 of 474 products scored exactly the same.
42 iOS apps and 2 Android apps in 113 days. Across both stores, every charge that ever landed adds up to 14 transactions and $69.82. Here is the full ledger and the full app list.
Over 84 days new installs went 451 → 685. Over the same window revenue went from $30.10 to $30.10. All of the growth was landing on apps with nothing to sell — and those were the apps converting best.
Form: 5 minutes. Landing: 5 minutes. Analytics: 5 minutes. I finished the funnel in 25 minutes, and only then asked where the link was supposed to go.
I went to use ratings as a metric and found the App Store Connect API doesn't expose ratings at all. Pulling 43 apps from the public lookup, all 77 ratings were Korean — the US and Japan were zero across every app. The cause wasn't the prompt. It was retention.
AdSense rejected my site for low-value content. I was sure my CDN had blocked the reviewer's crawler — it had actually happened once before. When I checked, the crawlers were getting through fine, and my app pages had 472 characters of body text.
I audited search traffic from first principles and caught the same error twice in one day. The long tail wasn't missing — my collector only held 10 rows. One app didn't own 73% of traffic — only 6 apps were instrumented. Neither flipped by digging deeper into the data, but by doubting the collection limit.
I shipped a SaaS in a day and shelved it in a day. Four research runs all pointed at the same wall — and then dogfooding killed the feature I built.
I poured six months into YouTube Shorts. When I finally measured app traffic by source, YouTube was 13% and Threads — which I'd barely touched — was 63%. And the measurement tool already existed; my grep silently lied and I nearly rebuilt the whole thing.
The plan was to attach new creative assets to my top-revenue apps and compare the funnel before and after. I ran the power calculation before executing — and these apps couldn't detect the effect in any realistic window. The app that measures the effect and the app that gets the effect had to be different apps.
I tuned ASO across 40+ apps for months, on the assumption that more impressions mean more revenue. The first time I pulled real money and cross-checked it, impression rank and revenue rank were inversely correlated. The lever wasn't traffic — it was revenue per install.
A first-principles audit of my social and YouTube bots. The dashboards were full of view counts, but the one number that mattered — does any of this produce an app visit? — was never being measured. The receiver was already live; one wire was missing.
Threads told me I'd crossed 50K views. I opened Insights to find the hit. My top original post: 1.5K views. The other 94% wasn't my content — it was impressions borrowed from someone else's viral thread.
Three overseas Merchant-of-Record processors rejected fortune-telling and astrology as a banned category. Dropping paid, flipping every app to free + AdSense — and the trap hiding in the deploy env.
I read a month of Search Console from first principles. 2,735 impressions is vanity; the real traffic was 8 clicks. The bottleneck wasn't app count or on-page tuning — it was one thing: domain authority.
I built paywalls before checking whether my category was even allowed. Lemon Squeezy, Polar, and Paddle all said no. Then my pivot to ads hit a second wall I'd built myself.
Not a '$10k/month' post. Three language channels, five auto-generated Shorts, and the honest numbers four hours after upload — plus the one signal that actually mattered.