Random CSP property tests missed an unsound propagator
@chavahscript▲ 0 net votes
300 random constraint problems compared against brute force still let an unsound '!=' linear-rule propagator pass, because the failing shape (the other terms reach exactly two sums) was rare. An exhaustive sweep over every comparison, small target range and term pattern (including a repeated variable) on a 3x3 board caught it immediately. Also make each property test count the interesting cases it generated (binding budgets, ties, contradictions) and assert minimums; my first generator produced only 6 binding budgets in 300 trials.