I run two small apps on two store platforms. One is a Shopify app that hides, renames and reorders payment and shipping methods at checkout. The other is a Cafe24 app. Cafe24 is a Korean hosted store builder, and that app previews and then writes search meta tags for a store's products in bulk, through the API, with an undo.
In review, both stores mostly flagged what I'd written and how screens looked, not code.
What review actually flagged
Cafe24 rejected my first submission for three things. A line of description in the header looked like a menu button and did nothing when clicked. There was no guide on first launch. And the listing's detail page wasn't good enough.
The second round flagged the detail page again, plus a real bug: product import stopped at 50 items. The reviewer also asked for an admin screen that felt more familiar and suggested selling extra volume as a paid option. I sent the third version in as a free app with nothing paid on any screen, and it passed. In-app billing came later, as an update that went through review on its own.
Shopify approved my app five days after I submitted it, with no rejection. Sixteen days after approval, a routine quality check flagged it. The app description used a plan name. A feature line said how many rules the free plan runs. A plan comparison image had no numbers on it at all. All three counted as pricing information outside the pricing section. The notice said the listing would be demoted and then removed if I didn't fix it. I fixed it the same day and it passed again about 19 hours later.
So approval isn't permanent, and a plan name counts as pricing.
Across both stores, one flag was a code bug. The rest were about what I'd written or how a screen looked.
Tokens break at the edges
On Shopify, auth broke after uninstall. Shopify's webhook auth helper refreshed the offline token first. Once the app is gone, that refresh always fails, and the webhook returned 500. On the first day of review, the reviewer's store sent an uninstall webhook and a shop-redact webhook, and Shopify retried each one nine times with exponential backoff over about four hours. Now the handler verifies the signature itself and carries on even when the token can't be refreshed.
On Cafe24, it broke under parallel load. The app runs as serverless functions, so several instances can get a 401 at the same moment. If two of them refresh with the same old refresh token, Cafe24 rejects the loser. The fix re-reads the token from the database before refreshing and uses the new one if another instance already won:
def _refresh(self, stale_access): with self._lock: # one refresh per process if self.tok["access_token"] != stale_access: return fresh = load_token(self.mall_id) # another instance may have refreshed if not fresh: # app was uninstalled: don't revive it raise AuthExpired(...) if fresh.get("access_token") != stale_access: self.tok = fresh return self.tok = _token_request(...) # we're first, so refresh The if not fresh check came later, from a security pass. Without it, a refresh still in memory could bring back a token that uninstall had just deleted.
One more Cafe24 detail I only found by measuring it. Refresh tokens last 14 days, and the expiry time comes back in Korean time, so I subtract nine hours before comparing it with UTC.
Limits and sleep
Cafe24 has two rate limits stacked. There's a short bucket of 40 calls that drains at two per second, reported in a header. Then there's a total over ten minutes. Cross that and you get a 429 and at least ten minutes blocked. My client gives up instead of waiting when the 429 asks for more than five seconds, so a request is meant to finish within a minute.
On Shopify, the limit that hurt was my own database. The app runs serverless with a free database that sleeps. The first query after a nap took longer than Prisma's default connect and pool timeouts, a webhook answered late and was retried, and two installs failed to get through auth. I raised the timeouts to 15 and 20 seconds. Then I found the Prisma session storage marks itself "not ready" for the life of the instance if its first check fails, so I wrapped it to check once more and retry.
Uninstall and data
Shopify requires three privacy webhooks, registered with one compliance_topics line in the app config. Unsigned requests get a 401. I don't store customer data, so two of the three have nothing to do. Shop redact deletes the sessions and the install record and keeps an anonymous history.
It helped that the rules never live in my database. They sit in app-owned metafields on Shopify's own object, and the checkout function reads the same metafields in its input query. There's no second copy to drift. My database holds three models: sessions, installs and install history.
Cafe24 sends me one webhook, for uninstall. The app first asks Cafe24 to revoke the token, then deletes it right away. If the revoke call fails or takes more than five seconds, the deletion goes ahead anyway. A scheduled job removes the data 30 days later. For new products I picked a daily fetch over the product webhook.
Moving a store between them
I also wrote down what moving a store from Cafe24 to Shopify would break. Shopify products take three options and up to 2,048 variants, so a Cafe24 product with four or more options, or a text-input option, needs reworking by hand. Images inside product descriptions have no import path, so they break when the old store closes. Past orders have no CSV route, only migration apps or the Order and Transaction APIs. Customer passwords can't move at all.
The same note keeps one line: the token a store gave my app is used only for what the store agreed to, and a migration isn't that.
Takeaway
Most of my fixes on both platforms sat at the edges: after uninstall, under parallel load, after the database woke up, and in the listing copy.
If you need payment or shipping methods hidden, renamed or reordered at a Shopify checkout, I take that work on here: https://www.upwork.com/services/product/development-it-shopify-checkout-payment-shipping-methods-hidden-renamed-or-reordered-2120151037133890296
If you've shipped an app to a platform store, what got flagged first in review: your code, or something you wrote about it?
This post was drafted with AI help. The numbers and facts come from my own logs and repo.

