Every figure here is measured against this table — six bot archetypes with known habits — and not against a solver. A line that beats a calling station grades well here and would not in a GTO tool.