- Some requests are refused outright, for every customer, on every plan.
- A refusal always names the rule and offers the lane that is open.
- Automation makes a bad instruction repeatable — which is exactly why the limit lives in code.
Fair housing is the clearest example of a rule that cannot be a preference. An agent who says something careless on one call has made one mistake. A campaign that encodes the same carelessness makes it four thousand times before lunch, in writing, with timestamps.
01What gets refused
Targeting or excluding by protected characteristic, steering language about who a neighbourhood is for, and any instruction that asks the assistant to describe an area in terms of the people in it rather than the property. These are refused at the point of instruction, not caught later in review.
02There is no override
No enterprise tier unlocks it. No support conversation ends with us making an exception. If a limit could be lifted by asking the right person, it was never a limit — it was a default, and defaults get changed by whoever is in the most hurry.
If a rule looks like a limitation, that's because it is one — the law's, not ours.
03Always name the open lane
A refusal that stops at no is a support ticket waiting to happen. Every blocked action comes back with the rule that stopped it and something concrete that is allowed: the property-led way to describe the same listing, the channel that still has consent, the time the queued dial will go out.
That is the whole posture. The system is not trying to be difficult, and it is not pretending the constraint away. It says what it cannot do, says why, and immediately says what it can.