Preview Deploy

POST /v1/accounts/{id}/deploy — deploy a validated bulb onto a paper account.

Provide exactly ONE source for the spec:

• strategy_spec: an inline strategy spec dict (validated server-side), or
• strategy_id: id of a strategy previously saved by the SAME principal.

name is an optional human label for the deployment (e.g. "oversold-bounce").

Deploys are FORWARD-ONLY (founder rule, 2026-07-17): the desk starts trading at the moment of deployment and only ever shows trades that happened while it was deployed. Historical performance lives in the strategy's BACKTEST report (stored per deployment, shown separately) — it is never replayed onto the desk as if the desk had traded it. The old backfill_days field (default 60!) silently pre-filled two months of simulated trades on every deploy and was removed for exactly that reason; unknown fields sent by old clients are ignored by pydantic.

Response to POST /v1/me/accounts/{id}/deploy/preview — what a deploy of the SAME body would do, computed on the deploy's own code path, with NOTHING written (beyond one agent-audit row).

valid/errors mirror exactly what the deploy would refuse with (same plain words). blocked_by lists the desk/account gates (maintenance, email verification, agent freeze, the key's capability, the desk's deploy cap, the plan's bot cap) that would refuse the deploy right now — each with the status and words the deploy would answer. would_deploy is true only when the spec is valid AND nothing blocks. preview_id is sha256(account_id + canonical JSON of normalized_spec)[:16]; pass it to the deploy to get preview_match back. Money: estimated_trade_cost (a number only where the sizing makes one knowable before a fill, else "depends on the fill") and desk (equity + available buying power).