Move from paper trading to controlled live execution

Define what the paper environment actually simulates

Record the data provider, brokerage model, account type, starting state, fees, slippage, fills, buying power, latency and market hours used in paper trading. QuantConnect documents that its paper brokerage uses fictional capital and simulated fills, with behaviour depending on the brokerage model. Other paper environments differ. The word paper does not establish which conditions are real and which are modelled.

Create a configuration record beside every result. Include engine and strategy version, parameters, instruments, permissions, data subscriptions and the period observed. Confirm whether the algorithm runs on live real-time data, delayed data or a replay. Without this record, a difference after going live cannot be traced to strategy logic, data, model, account state or deployment configuration.

Expect data and fill differences

Live data can arrive at different times and resolutions than historical or simulated input. Corrections, corporate actions, missing values, auctions and provider coverage can change the sequence seen by the algorithm. A paper fill model may complete an order immediately or without market impact while a real order waits, partially fills, rejects or receives a different price. QuantConnect documents several of these reconciliation gaps.

Test decisions around stale data, spreads, order type, liquidity, session boundaries and partial execution. Do not judge only the final equity curve. Compare each decision timestamp, submitted order, broker acknowledgement, fill and position. A paper result may reveal deterministic code errors and operational gaps. It cannot reproduce every queue position, counterparty event, market impact or live account constraint.

Carry state and controls across environments

Match starting cash, holdings, open orders, brokerage model, time zone and warm-up behaviour. Verify restart state, scheduled events and persisted data. A live account may contain positions or restrictions absent from the paper account. If the strategy assumes an empty portfolio or instantaneous state restoration, the first live action can differ even when the source code is identical.

Set hard limits outside the strategy where possible: instruments, maximum order, position and exposure, frequency, trading window and loss or anomaly stop conditions. Decide which actions require human approval and who can disable execution independently. Paper trading should test rejection and stop paths, not just successful orders. A limit that has never been triggered is configuration, not demonstrated protection.

Prepare monitoring, reconciliation and response

Monitor inputs, decisions, orders, broker events, positions, cash and risk state as one chain. Alert on silence, stale input, repeated rejection, inconsistent position and lost connectivity. Assign an owner and runbook to each alert. Email or dashboard visibility is not enough if no one can investigate, stop and reconcile the system during the relevant market window.

Reconciliation compares intended, platform and broker state. Practise it after a simulated disconnect, restart and partial fill. Define whether the safe response is stop, cancel, reduce, hold or manual review. Never replay blindly when external state is unclear. IMRYN treats monitoring and human oversight as control architecture, not as evidence that a strategy will perform or that every incident can be prevented.

Use a gated and reversible live step

Before real capital, require a review of strategy assumptions, code, access, data, controls, provider configuration, legal or regulatory duties and operational coverage. QuantConnect publishes separate strategy, portfolio, market, counterparty, operational and error risks for live trading. The responsible parties must decide whether those risks are acceptable. This page cannot make that decision and is not investment advice.

If approved, begin with the smallest live authority that can test the operational path, not the amount hoped to earn. Define maximum duration, capital, instruments, order size, expected events, stop conditions and review time. Compare live and paper traces, document every divergence and expand only through another explicit decision. A live test can verify parts of operation; it still cannot guarantee future performance.

Decision criteria

Use the same questions for every option before choosing.