Futures: sizing, history, warm-up

Position sizing (notional, not set_holdings)

To take a target notional (portfolio-fraction) exposure in a future, do NOT use set_holdings: for a future it sizes off (already-leveraged) buying power, so a target near 1.0 takes on many times 1x notional — margin calls and blow-ups — and |target| > 1 is rejected outright. Also, the continuous/canonical future symbol (e.g. /ES) is a DATA symbol and is NOT tradable — market_order or portfolio reads on it silently do nothing, so symbol below must be a real contract (the front month via future.mapped, or the specific contract the strategy selects). Size by notional with explicit contract math:

price = self.securities[symbol].price
multiplier = self.securities[symbol].symbol_properties.contract_multiplier   # read it; never hardcode (ES = 50)
target_contracts = round(self.portfolio.total_portfolio_value * target_weight / (price * multiplier))   # may be 0 on a small account — correct, not a bug
delta = target_contracts - self.portfolio[symbol].quantity; self.market_order(symbol, delta)   # trade the delta

History — use self.history(); do NOT rebuild a DataFrame from TradeBars

Seeding a raw look-back buffer (e.g. an expanding regression window) is a HISTORY REQUEST, not a warm-up (warm-up pumps data through on_data and updates indicators; a history request hands you a DataFrame to process yourself).

• Request continuous-contract history on the canonical future.symbol. Roll behavior is set by the data_mapping_mode / data_normalization_mode / contract_depth_offset passed to add_future (e.g. DataNormalizationMode.BACKWARDS_RATIO for a roll-adjusted price/return series).
• self.history(symbol, n, Resolution.DAILY) ALREADY returns a DataFrame — do NOT iterate self.history[TradeBar](...) and build one by hand. For a future it comes back MultiIndexed by (expiry, symbol, time); get a clean time-indexed series by dropping the leading levels: closes = self.history(symbol, n, Resolution.DAILY)["close"].droplevel(["expiry", "symbol"]) then group/resample by month on that Series. (Do NOT use history.loc[symbol] — that indexes the expiry level and fails.)
• A request can return fewer bars than asked when available history is limited; check the row count and widen the lookback if short.

Warm-up (for indicators, not raw buffers)

To ready indicators before the start date: set_warm_up(n[, resolution]) or set_warm_up(timedelta(...)), gated with if self.is_warming_up: return; or warm_up_indicator(symbol, indicator[, resolution]) / settings.automatic_indicator_warm_up / indicator_history(indicator, symbol, n, resolution). After warming, check readiness and widen the lookback if short.