Kristian checked out with an address in no group and was offered both free
delivery methods set up for the Netic group. Measured on the shop before
touching anything: one group, four rules, two of them "show" on the free
delivery methods he saw.
It was not a regression. A test asserted the behaviour as intended — "an
unmatched customer sees every method" — so a "show" rule was a whitelist FOR
THE GROUP and nothing more. The plugin did half of what its own description
promises: it filtered what the group sees and never reserved anything from
anyone else. On a shop that takes money the effect is that free delivery
arranged for one company is available to every customer.
A method named in any group's show rules is now reserved for the groups that
name it.
Answering "does it work for other discounts too", which found the hole the
first fix left: it did not. A group with NO rules of its own returned every
rate untouched, and so did a group carrying only "hide" rules, because neither
engages the whitelist — so their members saw another group's reserved methods.
Both now have other groups' reserved methods removed.
The "never leave the customer without shipping" guard is kept and asserted: a
shop that reserved everything it offers must still let an ordinary customer
check out, because that failure is worse than the leak.
Four assertions, each proven red. Two of them caught my own mistakes rather
than the product's: the hide-only fixture originally left nothing for the
customer, so the guard fired and returned every rate — which reads exactly like
the leak it was written to detect. A third rate makes the behaviour observable.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
