You finish it at 1am. You tap through the whole thing on your laptop: sign up, add an item, edit it, delete it, log out, log back in. Everything works. You feel a small, stupid amount of joy, then you send the link to your mum.
Ten minutes later: "It says my email is wrong."
You look. It is not wrong.
"I didn't get the code." It went to her spam folder, in a mail app she never opens.
"I clicked the link and got a blank page." She clicked it on her phone, the link had already been used on her laptop, and the page did not know what to say.
"It took my card and then nothing happened." It did not take her card. It showed a spinner and forgot about her.
Nothing was broken. That is the part people underestimate. Everything worked exactly as well as it did for the person who built it - who was already signed in, whose email was already confirmed, whose card was already saved, and whose browser already had every cookie that makes the app feel simple.
The bug was the tester.
The five doors a stranger walks into
Every app people pay for has five doors that are invisible to the person who built it.
The first door is the one you have never walked through since the first day: a brand new account, a confirmation email that lands somewhere you did not expect, a link that opens in a different browser than the one that asked for it.
The second door is the one you never use at all: "I forgot my password." You have never once clicked it, so you have never learned that the reset email arrives with a broken link at 3am.
The third door is a phone held in one hand, in a car, with bad signal. Your app was tested on a laptop, on wifi, with the patience of its author.
The fourth door is money. Taking a payment is not a button. It needs a business identity, a webhook that actually answers, and a screen for the case where the payment succeeds and the app crashes before it says so.
The fifth door opens a week later, when two people do the same thing at the same time, or the server restarts at 3am, or someone's subscription silently fails and they keep using the app for free.
None of these are features. Not one of them will ever appear on your to-do list, because you cannot see them from inside the builder's seat.
You can find them in ten minutes, tonight
Hand your phone to a person who has never seen the app. Do not explain anything. Do not help. Just watch where their thumb stops.
Then check three emails on a mail app you have never used: the sign-up one, the password-reset one, and the receipt. If any of the three looks like an error message, that is your next two days of work - not the extra feature you were planning.
Why this is where the last 20% goes
I finish apps that got to "almost ready" and stopped. The shape is almost always the same: the visible part - the screens, the idea, the thing that made you start - is done and looks good. What is left is this door work. Ugly, invisible, and two to four days of it.
It is also the only work that turns a demo into something you can charge for.
I do this as a one-person studio in Seoul: I take the app you already have, keep what works, and get it to the point where a stranger can install it, pay for it, and come back tomorrow. Three to five days for a focused web or mobile app. You talk to the person writing the code, the first conversation costs nothing, and you pay the rest only after you have seen it working.
- Studio (English): https://cpcai-studio.netlify.app/en/
- Email me directly: [email protected]
If you have an app sitting at 80%, the useful question is not "what else should I add". It is: which of the five doors has a stranger not walked through yet?
Top comments (4)
Sign in with Apple has one of those doors. I never hide my email, so I never walk through it.
Pick "Hide My Email" and Apple gives the app a relay address. Then Apple started using a second relay domain, my code only knew the first one, and those people got a username made out of the random relay string š
That is the exact door I meant, and you found the nastier version of it: the relay domain itself changed under you.
privaterelay.appleid.comwas the only one anyone coded against, then Apple started issuing@icloud.comrelay addresses, and any check that compared the domain or built a username from the local part starts producing garbage.The lesson I took from that class of bug is to stop treating the email address as identity at all. The Apple
subclaim is the only stable identifier; the address is a delivery detail that can change, can be a relay, and is sometimes missing entirely when the person picks "Hide My Email" and never forwards. Any screen that shows the address then needs a readable fallback, because2f8c9dā¦@privaterelay.appleid.comas a username is the bug, not a cosmetic issue.In practice that cost me a settings field (let people set their own display name, never derive it from the local part) and an allowlist that has to accept both relay domains, including the ones that do not exist yet. That last part is the uncomfortable one: that list is not something you can fully enumerate in advance.
Did the second relay domain break sign-in for those users outright, or only the generated username? I have only seen it surface as ugly names so far, and I would rather learn the failure mode from someone who actually hit it.
Only the username. Sign-in matches on Apple's own user id and treats the relay address as a normal email, so people with the new domain got in fine. The address just failed my "is this a relay" check, so the username came from the random part.
The new domain is
private.icloud.com, by the way, so a check for@icloud.comalone won't match it.Thanks for coming back with the domain.
private.icloud.comis the detail I would not have found on my own, and it fits what I saw here: sign-in kept working, only the derived name broke.That pushes me off the domain allowlist entirely. The ID token carries an
is_private_emailclaim that Apple sets to"true"for relay addresses, which is the provider's own answer to "is this a relay" and does not need editing when Apple adds a domain. I keep the domain check only where a wrong answer is cosmetic, never where it decides a stored value.Two things I would add, given your failure was contained:
nameclaim at first authorization is the only human-readable thing Apple gives you, and it arrives exactly once. I store it then and let people edit it later; the local part never becomes a name.The
sub-as-identity part is the one I would repeat to anyone building this: the address is a delivery channel, not a key. The migration on my side is slowly keying accounts on(provider, subject)and merging the duplicates that predate it - dull work, but it is the only thing that turns a relay-domain change into a non-event.