A rule you cannot evaluate is not a rule that passes. It is a rule that refuses.
This is the quietest correctness bug in discount and limit logic. A promo code carries a per-customer cap. The request arrives with no customer attached. The cap query comes back with zero uses. Zero sits below the cap, so the code redeems, and it redeems every time, because an anonymous caller keeps a clean record.
Nothing threw. Nothing logged. The limit just stopped existing for the one caller who could abuse it most.
In our open-source project promocodes, the validation path is built so that cannot happen. The check answers with the first rejection that stops a redemption, and before it evaluates anything it asks whether it holds the facts the rules need.
If there is no customer, and the code is either audience-restricted or carries a per-customer limit, the answer is CUSTOMER_REQUIRED, returned ahead of the assignment check, ahead of the validity window, ahead of every usage count. The order minimum mirrors it at the other end of the list: if a minimum total is configured and no total was supplied, the answer is TOTAL_REQUIRED, and the below-minimum comparison runs only once a total is in hand.
Two details make that hold under pressure.
First, the storage layer enforces the same rule instead of trusting the caller. The usage lookup requires the per-customer count to be zero when there is no customer, and the in-memory implementation returns only a global count for an anonymous caller. There is no zero to wave through, because there is no per-customer number at all.
Second, every entry point (redeem, quote, capture) reaches the rules through the same validation function. A refusal you can route around is not a refusal.
The general shape: a precondition failure and a rule failure are different answers, and the precondition has to be its own named rejection, evaluated first. Collapsing them into one boolean is what turns a missing input into a silent approval.
There is one deliberate inconsistency worth naming. A total supplied in the wrong currency raises instead of returning a rejection. Missing data is a normal outcome the caller can fix by sending more, while a mismatched currency is a caller bug, and quietly handing it back as a business rejection would hide a defect in the integration.
When you audit limit logic, a happy path that enforces the cap tells you very little. What matters is what the cap does when the field it depends on is null.
We would ask where your validation layer draws the line between "refuse, data missing" and "raise, caller is wrong".