Most algorithmic trading strategies do not fail because the underlying idea was wrong. They fail because the trader skipped the steps between a working backtest and a live deployment. The strategy was fine. The infrastructure was not.

This checklist is not motivational. Each item has a pass condition and a failure mode — what concretely breaks when traders skip that step. Run through it before you commit a single dollar of live capital.

Edge Validation

Before deployment, the strategy must have evidence of edge that was not used to build it. Backtests that use all available data to fit parameters are, at best, overfit. At worst, they are a description of historical noise.

  1. Out-of-sample performance exists.

    • Pass: You have a held-out test period, walked-forward window, or paper-traded period that was never touched during strategy development. The Sharpe on that period is positive.
    • Fail: You optimized parameters on the full dataset. There is no period you did not look at.
    • Failure mode if skipped: The strategy makes money on paper and loses money on every live trade. The backtest fitted noise that does not repeat.

  2. The mechanism is documented in plain language.

    • Pass: You can write two sentences explaining why the strategy should work — not that it did work, but the causal mechanism. Example: "This strategy captures mean-reversion in overbought large-caps because institutional rebalancing creates predictable short-term reversals."
    • Fail: The mechanism is "the backtest showed a positive Sharpe."
    • Failure mode if skipped: When the strategy enters drawdown, you have no basis for deciding whether to shut it down or hold. You will make that decision emotionally.

  3. You have read the research, not only the result.

    • Pass: At least one academic paper or verified practitioner source describes a related phenomenon in live markets, not only backtested markets.
    • Fail: You found the idea on a trading forum with no supporting evidence beyond another trader's equity curve.
    • Failure mode if skipped: You deploy a strategy that accidentally exploits a look-ahead bias described on a finance blog as a "discovered edge." This is common.

See What Is a Trading Edge? for the full framework on evaluating whether an edge is structural or coincidental.

Position Sizing and Risk Parameters

A strategy with positive expectation can still destroy a portfolio if position sizing is wrong. Max drawdown scales with position size. Most traders underestimate how bad a drawdown can get before it recovers.

  1. Max drawdown is acceptable at full position size.

    • Pass: You have simulated or calculated the maximum drawdown this strategy produced in the out-of-sample period at the exact position size you intend to use live. You are comfortable losing that amount without shutting the strategy down prematurely.
    • Fail: You calculated drawdown as a percentage without translating it to a dollar figure. Or you intend to start small and "scale up" — which means you have not tested at target size.
    • Failure mode if skipped: A 15% drawdown that looked tolerable in percentage terms equates to $15,000 on a $100,000 account. Most traders who never calculated this exit the strategy at the worst possible moment.

  2. Position size is defined before the first trade.

    • Pass: A fixed dollar amount or a fixed percentage of equity per trade is written down and will not be changed during a drawdown.
    • Fail: Position size is "whatever feels right" or "I will start conservative and adjust."
    • Failure mode if skipped: Position size drifts upward after wins and downward after losses, both at the wrong times. This turns a positive-expectation strategy into a net loser.

  3. Kelly fraction has been calculated and halved.

    • Pass: If you are using Kelly criterion for sizing, you have computed the full Kelly and are using at most half. Half-Kelly roughly halves drawdown while preserving approximately 75% of the compounding rate.
    • Fail: You are using full Kelly or a fraction you selected intuitively.
    • Failure mode if skipped: Full Kelly produces drawdowns that are mathematically survivable but psychologically impossible to hold through. Traders who use full Kelly quit before the math resolves.

Execution Readiness

The strategy's behavior in a backtest assumes perfect fills, zero slippage, and no connectivity issues. None of those assumptions hold in live trading. Each gap is a cost center.

  1. Order types have been tested in paper trading.

    • Pass: You have sent limit orders, market orders, and cancel-replace orders through your broker API in a paper environment. You know which order types your API supports.
    • Fail: You wrote order logic in code but have never confirmed it executes against the actual API.
    • Failure mode if skipped: On the first live trade, the order type you assumed was supported either rejects silently or defaults to a market order at a bad price. This is one of the most common first-deployment failures.

  2. Error handling exists for connection loss.

    • Pass: The strategy has logic for: what happens if the API call fails; what happens if the position is open and the connection drops; what happens if the strategy restarts mid-day with an open position it did not open.
    • Fail: The strategy assumes a clean environment and has no retry logic or position reconciliation.
    • Failure mode if skipped: A broker API outage during market hours leaves you with an open position and no automated way to close it. You find out during market hours with no prepared response.

  3. Slippage has been estimated and is within tolerance.

    • Pass: For each instrument you trade, you have estimated how much of the strategy's expected return is consumed by execution costs at your target volume, using the average bid-ask spread as the floor.
    • Fail: The backtest assumed exact fill at the close price or the signal price without modeling spread.
    • Failure mode if skipped: A strategy that showed 8% annual return in backtest produces 2% after actual execution costs, and breaks even after a single month of adverse fills.

Monitoring and Kill Switches

A live strategy running without monitoring is not a systematic strategy. It is an unattended process with access to your brokerage account.

  1. Alerts are configured for abnormal behavior.

    • Pass: You receive a notification if: the strategy has not sent a heartbeat in N minutes; daily PnL has exceeded your defined max daily loss; position size has deviated from the expected value.
    • Fail: You plan to check the strategy manually each day.
    • Failure mode if skipped: The strategy enters a bug state where it fires orders in a loop. You find out three hours later when you check your account. This is not a hypothetical.

  2. A daily max-loss circuit breaker is configured.

    • Pass: The strategy has code that halts all order submissions if the day's realized loss exceeds a defined threshold. This threshold is set before the strategy goes live, not adjusted during drawdown.
    • Fail: The strategy runs without a hard stop. You rely on position sizing alone to contain the damage.
    • Failure mode if skipped: On a high-volatility day, the strategy fires losing trades faster than normal. Without a circuit breaker, the daily loss is limited only by the account balance and the number of trades the strategy can send in a session.

  3. The kill switch has been tested.

    • Pass: You have a procedure for stopping the strategy immediately that does not require editing code. You have tested it. You know the sequence of commands by memory.
    • Fail: You know you can "stop the process" but have not drilled it.
    • Failure mode if skipped: In a fast-moving adverse event, you spend ninety seconds trying to remember how to stop the process. In an API-driven strategy, ninety seconds is ninety fills.

The Psychology Layer

Position sizing and monitoring protect your account. The psychology layer protects your decision-making. Most of the rules in this section feel unnecessary until you need them.

  1. Decision rules for shutting down the strategy are written before going live.

    • Pass: You have a written rule for when you will halt the strategy — not a feeling, a criterion. Example: "If the strategy loses more than 20% from peak equity on the out-of-sample period, I will halt it and review." This rule cannot be changed during a drawdown.
    • Fail: You plan to evaluate the strategy's health based on how it feels at the time.
    • Failure mode if skipped: You will always find a reason to keep running a losing strategy during a drawdown. "It will come back." "This is a rough patch." Every trader does this without a pre-committed rule.

  2. You have a definition of what constitutes a broken strategy versus a normal drawdown.

    • Pass: You have answered: "How much drawdown is expected in a 6-month bad period for this strategy?" You have a number. If the live drawdown exceeds that number, that is a broken-strategy trigger, not a bad period.
    • Fail: You have not drawn that line.
    • Failure mode if skipped: You tolerate a 40% drawdown as a "rough patch" when the backtest only showed 12% max drawdown. The strategy was broken at 15%, but you had no threshold, so you held on.

  3. You will not run the strategy and override it simultaneously.

    • Pass: You have decided in advance whether you are running the strategy or running manual discretion. They do not run simultaneously in the same account.
    • Fail: You plan to let the algo run but reserve the right to manually intervene when you disagree with a trade.
    • Failure mode if skipped: The first time you disagree with a trade, you override it. Then a second. After three overrides, you are running manual discretion with extra steps. The systematic part of the strategy is gone.

The Oyamori Approach

Oyamori treats strategy readiness as a technical state, not a feeling. Before a strategy can be promoted to live execution, the platform walks through structural validation: is there out-of-sample data? Are position sizing parameters defined? Are error states handled in the strategy code?

The monitoring layer is not optional. Every live strategy deployed through Oyamori runs with alert thresholds and a daily loss circuit breaker configured at the platform level. Kill switches are exposed as first-class CLI commands, not afterthoughts buried in a dashboard.

The checklist above describes what the platform enforces at the infrastructure level. Traders who want to skip these steps can build their own stack. That is a legitimate choice — and a substantial amount of work that most retail traders underestimate before they begin.

Algorithmic trading carries substantial risk. This article is educational, not investment advice.

Next: Quant Trading vs. Gambling →