Provides deterministic and random exposure argument generation to enforce min/max ownership percentages across a set of lineups. The algorithms are independent of sport-specific logic and could be repurposed for any scenario requiring controlled repetition of selections across multiple optimization runs (e.g., portfolio construction, team selection). It imports from internal ORM (Roster, Player) but the core logic for generating exposure limits is separable.

Defines LineupConstraints class that encapsulates banned/locked players and positions. The builder-style API (add_banned_player, add_locked_position) is self-contained and can be reused across different optimization setups. It depends on the Player class from orm.py for type hints, but the structure is a clean pattern for declaring discrete constraints in a combinatorial solver context.

Contains OptimizerSettings, PlayerPoolSettings, CustomRule, and Stack classes. CustomRule allows users to define arbitrary constraints via a (callback, rule_type, position_group) tuple, enabling complex rules like 'do not pair a QB with opposing defense' or 'force at least 2 players from one team'. The Stack class models stacking constraints (e.g., QB+WR from same team). This pattern provides a flexible DSL for extending OR-Tools constraints without modifying the optimizer core, and could be adapted to other optimization problems requiring user-defined constraints.