Turn a strategy into a testable object.

A Strategy Case is a strategy written down precisely enough to be tested: the current state, the proposed action, the assumptions it rests on, and the constraints it must respect. Once a plan is a case, the engine has a stable object to operate on — and every interface we are building, from the web workspace to file inputs and agent tools, is designed to read and write the same versioned case.

Six steps, run in order, every time. The loop is the product; the verticals change the vocabulary.

Most plans fail at an assumption, not at the average. A Shear Test stresses one assumption at a time — and the combinations that plausibly move together — to produce a Shear Profile: a map of how the plan degrades, which assumption carries the risk, and where the cliff sits. Robustness claims are always scoped to the declared branch family, assumptions, constraints, and evidence tier — never to the world at large.

When outcomes arrive, each case records what was predicted, what was observed, and the evidence tier that supports any comparison. Valid comparative evidence contributes to Decision Fidelity; every resolution can contribute provenance to a Decision Graph. The intended graph is a private, versioned record of state, action, prediction, and outcome. The outcome-resolution system is in development.

Great Bear recommends, compares, and flags. It does not execute. Every public prototype ends at a human decision, and every recommendation carries its trigger and its downside, so disagreement can be specific.