A machine formoving worlds.
Hivemind treats strategy as a live system: not a prompt, not a framework, not a confidence score, but a world of actors and pressure.
The state.
A session starts by refusing the obvious frame. Hivemind builds a StructuredSituation: actors, incentives, constraints, irreversibilities, uncertainties, and causal links.
The map comes before the framework. If the map is thin, every answer downstream is treated as suspect.
The transition.
Where the structure allows it, Hivemind asks for a runnable model: causal, strategic, or system-dynamic. The point is not to pretend the model is the world. The point is to expose what the world would have to do for a recommendation to survive.
A model that cannot be run remains a metaphor. Hivemind prefers machinery when machinery can be earned.
The game.
Some situations are games. Some only pretend to be. Hivemind infers players, moves, payoffs, information, and the places where the game is too malformed to solve.
Best responses are not decoration. They are what happen after your answer leaves the page.
The validator.
Hivemind separates generation from evaluation. Surviving models are pressed by perturbation, adversarial falsification, structural fit, and constraint review.
The validator is allowed to dissent. A validator that never dissents is not a validator.
The trace.
Output is a verdict with provenance: the situation map, the frames considered, the models that failed, and the uncertainty that remains.
The recommendation is never allowed to travel alone.
The boundary.
Hivemind can run as a hosted web app or inside a private deployment. The corpus, trace, and decision material can stay behind the boundary the situation requires.