Task 10

Enter a watchlist (2–15 tickers). Each name gets its long/flat signal from the Task 4 technical agent; an LLM then picks one sizing policy from a fixed menu (equal-weight / inverse-vol / risk-parity / signal-proportional, plus a single-name cap, gross cap, vol target and rebalance cadence) using as-of universe stats — NOT per-name weights, which are computed deterministically. A lookahead-free portfolio backtest then allocates across names vs an equal-weight basket and the S&P 500.