In 16 Claude Code runs, every setup built a correct query, but only the setups with an explicit rule paginated: 0 of 4 for no rules and for Boost, 4 of 4 for a one-line rule.
Boost cost about 22 percent more per run than no rules and a Boost rule file about 40 percent more, mostly from extra turns spent reading guidelines.
A rule is a request and a test is a check, so back important rules with a Pest test; mine passed on a paginated run and failed on an unpaginated one.