Bennet

The portal configuration has subscription_update.proration_behavior and the default is create_prorations, which is exactly the credit you are seeing. Set it to none and the switch takes effect immediately with no money moving.

Better shape, if you have twenty minutes: keep prorations for upgrades (people like getting charged less to move up) and handle downgrades yourself with a subscription schedule that flips the price at current_period_end. They keep plan A until the date on the receipt they already paid, then land on B. Nothing to credit, nothing to explain.

Also worth knowing that portal configurations are objects, not one global setting. You can allowlist which prices are switchable and hand different customers different configurations.

That creates a gap where the customer has no active subscription, which means your entitlement checks have to handle "paid until Tuesday but cancelled" as a state, and it depends on them coming back to resubscribe, which some fraction will not.

You also lose the continuous subscription record, so anything you compute off subscription age (churn cohorts, lifetime value) now sees two customers where there was one. A schedule phase does the same thing without any of that.

Separate thing to check while you are in there: in v4 the class strategy is not a config key any more. You want

@custom-variant dark (&:where(.dark, .dark *));

in your CSS. If you upgraded from v3 and left darkMode: 'class' in a tailwind.config.js that nothing loads, dark: quietly falls back to following the OS preference instead. On a machine set to light that looks exactly like a flash that never resolves, and on a machine set to dark it looks like everything works.

If you are on Postgres you can go one step further and set the org id as a session variable that row level security policies read. Then a handler that forgets its WHERE clause returns nothing instead of everything, which is a much better failure mode to have as your default.

Add tailwind-merge at the same time or your escape hatch will not work. A consumer passing className="p-8" on a component whose variant emits p-4 gets whichever one the compiler happened to put last in the stylesheet, which is not the one they wrote, and it will look correct in one place and wrong in another.

One thing that trips people up on the way to this question: the share price typically drops by roughly the dividend amount on the ex-date. Buying just before the ex-date to capture a payment gains you nothing — you pay for the dividend in the price and then owe tax on it. Not what you asked, but it is where this question usually leads next.

v4 detects sources automatically from the stylesheet's project root and respects .gitignore. Two ways that goes wrong:

One, your build output directory is not actually gitignored, or you added an explicit @source that reaches it. Now it scans your previous build, and every class string that has ever existed in that project comes back from the dead. This is the common one and it explains "hex colours I never typed", because they were in a build from four months ago.

Two, some directory of content - markdown, JSON fixtures, a vendored file - contains words that happen to parse as candidate utilities.

Fix is to stop relying on detection. Narrow it with explicit @source rules for the directories you actually author in, and carve out the offender with a @source not rule. Then diff the output again and it should drop back.

No, v4 still only generates what it has seen in a source file. What changed is how it decides which files to look at, which is exactly why this thread exists. If it emitted everything, the number would be the same for everybody and it very much is not.

It is a decent rough filter but it throws away whole sectors where a high payout is structural. I use it as a prompt to look harder rather than to exclude.

The maintenance versus growth split is not disclosed cleanly anywhere. Any shortcut for estimating it?

Keep seats for who is allowed to log in and meter the thing that costs you money. Two dials, both of which a finance person can read off an invoice without calling you. Seats-only breaks exactly the way you described, usage-only makes small customers nervous.

The case for the hook is when the difference is not visual. Different data fetch, different subtree, mounting a map in one and a list in the other. CSS cannot decline to render a component, and doing it in CSS by hiding one means you paid for both.

The thing that stops conversion is usually not the size of the free tier. It is that at the moment somebody hits the limit, the paid tier is not obviously the answer to the problem they are having right now.

This is the actual argument against DRIP and it almost never comes up. My twelve year old position has 48 lots in it.

Something people miss: the reported character can be reclassified after year end. What shows as return of capital on a quarterly statement can be restated as income or gain on the annual tax document, so do not plan off the quarterly line.

as is "trust me" with a keyword.

Check the currency too. If the dividend is declared in one currency and paid into an account in another, some brokers hold it a day or two for conversion and take a spread you never see itemised anywhere.