Hypothesis and Strategy are separate aggregates

The v2 design draft moved a single Candidate object through one state machine from proposal to live trading. That conflated two things with genuinely different lifecycles: a Hypothesis is a question about the world that resolves once and costs alpha, while a Strategy is a deployable artifact that costs money and has an operating life of pause, resume and retire. We split them, and made the join a hard rule: a Strategy may only become deployable by citing a resolved Hypothesis whose verdict supports it.

Considered options

• **One Candidate.** Simpler, one event log. Rejected because alpha cost becomes ad hoc exactly where it matters: instrument variants of a confirmed effect, resuming a paused strategy, and strategies resting on more than one question. Ad hoc alpha accounting is how the donor programme reached a search-size-corrected 1.70.
• Strategy only, no register. Closest to the reference architecture we are modelling. Rejected because pre-registration is the donor repository's entire value, and without it the pipeline is a machine for generating overfit strategies at industrial scale.

Consequences

The deploy gate stops being a judgement call and becomes a lookup. A Hypothesis that resolves null is still valuable and is kept; only the Strategy that rested on it is retired. Two state machines and a join must be maintained instead of one.