execution-positionlisted

# Execution & Position The distance between a good signal and a good trade is sizing, limits, and costs. This skill is the boring adult at the door: it sizes with declared inputs, checks the hard limits, prices the friction, and emits **paper intents only**. ## Authority boundary (read first) The gate's output is a `paper_only: true` intent. A request for live execution FAILS the gate mechanically — live orders are a human decision made outside this skill, in the broker's own interface. This mirrors the platform rule: unimplemented broker paths return errors, never fake fills. ## Sizing doctrine - **vol_target** (default): weight = risk budget / annualized vol. Needs a real volatility estimate; the gate refuses to guess when none is supplied. - **kelly**: only from a **declared** edge (win rate + payoff ratio) — the declaration is the user's claim and ships verbatim in the intent for audit. Computing an edge from sample means is prohibited here (that is backtest-validation's jurisdiction, and even then it is a backtest, not an edge). Fraction hard-capped at 0.25: full Kelly on an estimated edge over-bets by construction. - **fixed_weight**: explicit user weight, still subject to every gate. ## Workflow 1. Assemble the spec (`references/order-schema.md`): order, portfolio state, limits, market context (vol, ADV, spread). Missing limits get conservative defaults; missing volatility fails vol-target sizing honestly. 2. Run `python scripts/position_gate.p