Demo-to-Live: A Proper Soak-Test Protocol Before You Trust Automation With Real Money

"It worked on demo for a week" is not a test. It's an anecdote. A proper soak test is structured, has a defined pass condition at each stage, and takes longer than most traders want it to — which is exactly why most traders skip it, and exactly why most blown accounts trace back to skipping it.

A single week of demo testing samples one market regime. If that week happened to be trending and low-volatility, you've learned nothing about how the tool behaves in a choppy range or during a high-impact news spike — the two conditions most likely to expose a badly-configured stop loss or a sizing bug. The protocol below is three stages, and none of them are optional.

Stage 1 — Demo, minimum 3–4 weeks, across at least one high-volatility event

Run the automation exactly as you intend to run it live: same lot-sizing rules, same risk limits, same symbols, no manual intervention — the point is to observe the tool's own behavior, not your corrections to it. Make sure the window includes at least one NFP, CPI, or central-bank rate decision, since that's where execution quality and slippage handling actually get tested.

Pass condition: every position got a stop loss, sizing stayed proportional to account equity as balance moved, and nothing unexpected happened during the volatility spike.

Stage 2 — Live, smallest possible size, minimum 2 weeks

Demo execution and live execution are not the same thing — spreads widen differently, requotes happen, and slippage on a live server under real liquidity doesn't always match a demo server's simulated fills. Move to a live account at the smallest size your broker allows. You're not trying to make money in this stage. You're confirming that what you observed on demo holds up with real execution.

Here's a concrete way to judge it: if your model assumed 1.5 pips of average slippage on EURUSD and Stage 2 shows you closer to 4 pips, that's not a rounding error — on a 30-pip stop, that's more than a 10% difference in your realized risk-per-trade versus what you planned for. A gap that size means either your risk model needs adjusting or your broker's execution quality isn't a fit for this strategy, and either way you want to find that out at minimum size, not at target size.

Pass condition: live slippage and fill quality are close enough to your Stage 1 model that the effective risk-per-trade hasn't materially changed.

Stage 3 — Live, target size, ongoing monitoring for the first month

Only after Stages 1 and 2 pass cleanly do you move to your intended position size. "Passed cleanly" means no unexplained closes, no orphaned positions, no sizing anomalies — not "it made money," which is a separate question from "it worked correctly." A tool can be profitable and still be misconfigured in a way that hasn't cost you yet.

Check logs at least every few days during this stage rather than only glancing at the equity curve. This is where a rare edge case — a specific broker error code, a symbol with unusual tick behavior, a news window your filter didn't catch — is most likely to surface for the first time, simply because target size finally makes it visible.

TIP

Keep a simple log of every stage-transition date and what you were watching for. If something goes wrong three weeks into live trading, being able to say "this behavior is new — it didn't happen in 4 weeks of demo or 2 weeks of small-size live" is the difference between debugging a real issue and chasing a phantom one.

Key takeaways

• Three stages, each with its own pass condition — not "it worked" as a single vague bar.
• Stage 2 exists specifically to catch the gap between modeled and real execution quality before it's expensive.
• A profitable Stage 3 that skipped Stage 1 or 2 hasn't actually been tested — it's been lucky so far.

FAQ

Can I shorten Stage 1 if the strategy has a long backtest history? No — a backtest tells you about historical price action, not about how *this specific deployment* (your broker, your VPS, your exact configuration) behaves live. Stage 1 exists to test the deployment, not the strategy.

What counts as "smallest possible size" in Stage 2? Your broker's minimum lot size — typically 0.01 lots — regardless of what your eventual target size will be. The goal is measuring execution quality with real money at stake, not managing meaningful risk yet.

How do I know if a Stage 3 issue is a bug versus normal variance? Compare it against what you logged in Stages 1 and 2. If the behavior is genuinely new — not something you saw during 4+ weeks of demo or 2+ weeks of small-size live — treat it as a signal worth investigating, not noise to average out.