From backtest to live: what changes

The moment you press Save

Right now your strategy is a draft, and a draft is a sandbox. You can rewire the filter tree daily, re-run the backtest, tear it all down and start over, and nothing leaves the room. Press Save and that changes. The configuration locks, live detection switches on, and from that point the platform watches every upcoming Pinnacle matchup on your sport, waiting for one that satisfies your tree. Saving requires an active plan: drafts and backtests are open to everyone, but a strategy that actually sends you picks is a plan feature. If you are still deciding which one, the plans page has the details.

The lock matters more than it sounds. In a backtest you are free to nudge a threshold every time a result annoys you. Live, you cannot, and that is the point. The record you build from here on is the record of one fixed set of rules meeting an unknown future, which is the only kind of record that means anything.

How a live pick is born

Detection is mechanical. A pre-match matchup appears, the platform evaluates your tree against it at Pinnacle's opening odds, and if every condition passes, a pick fires and goes out on your notification channels. Two details are worth engraving:

• Never in play. Picks never fire on in-play matches. If a game has kicked off, it is out of scope, full stop.
• Once, by default. The check happens a single time, at the opening price. If you enable odds drift, the platform also re-evaluates the matchup when Pinnacle's price moves, so a match that missed your odds range at open can qualify later if the line drifts into it.

That second point is a genuine trade-off, and we will come back to it.

The price is not the one you tested

Here is the honest gap this article exists for. Your backtest priced every simulated bet at the closing odd by default, the last price Pinnacle posted before kickoff (opening odds are an option in the backtest settings, which narrows this gap). Your live picks record the odd at the moment of detection, which can be hours or days before the close. Those two numbers are rarely equal.

So a live pick can get a better price than the backtest assumed, or a worse one. The difference is not noise to shrug at. It is precisely your closing line value. Detect at 2.10, watch it close at 1.95, and you beat the close: the market moved toward your side after you got in. If your picks systematically close shorter than you took them, your live results should, over time, outrun what the backtest promised. If the market keeps moving against you after detection, expect the opposite. This is the single most useful lens on early live performance, and Value betting and CLV makes the full case for it.

One bookkeeping rule keeps the record clean: a strategy holds at most one bet per matchup. If a match was already in your history as a simulated bet and detection then fires on it live, the live bet wins and replaces the simulation. No double counting, and the live price, the one you could actually have taken, is the one that stands.

Your record splits in two

After going live, every stat on your strategy exists in two scopes. Overall keeps the backtest: simulated history plus live picks, blended. Since shared starts from zero the moment you go live and counts real detections only.

You will be tempted to stare at Overall, because it is big and it is the number you fell in love with. But Since shared is the one the leaderboard ranks on and the one prospective followers judge, for a blunt reason: a backtest can be tuned to fit the past, while live picks called the future or they did not. Your Overall yield is your hypothesis. Your Since shared yield is the experiment. Expect the second to look worse than the first for a while, as almost everyone's does, and read Backtesting and overfitting if you want to know why in advance rather than the hard way.

The first weeks will lie to you

A strategy that averages thirty bets a month has, after three weeks, about twenty live results. You already know what twenty bets are worth as evidence: roughly nothing. A 65% winner over that stretch proves little; a 30% stretch refutes little. Early live weeks are variance wearing your strategy's clothes.

The discipline is to judge the live sample by the same standard you applied to the backtest: sample size first, everything else after. You would not have trusted a 20-bet backtest; do not panic over a 20-bet live start, and do not take a victory lap over one either. Watch the CLV on each pick while the results accumulate. It tells you whether the machine is finding good prices long before the profit column can. When the live sample finally is large enough to read, read it with the backtest metrics guide open: same metrics, same thresholds, no special pleading because this time it is your money.

The knobs you still control

Locked rules do not mean a locked life. Three controls remain in your hands:

• Pause. The Enabled toggle in the strategy header stops live detection without touching the strategy or its record. Useful for an off-season or while you rethink your bankroll. One restriction: public strategies cannot be paused, because followers are relying on the feed.
• Odds drift. On, you get more picks, often closer to kickoff, because matches can qualify as prices move. Off, you keep strict opening-price discipline: fewer picks, taken earlier, at the price shape your backtest most resembles. Neither is free; drift trades entry-price purity for volume.
• Archive. When a strategy has run its course, archiving retires it cleanly: detection stops, the strategy leaves your listings, and its slot is freed for a new build. The record survives. Archiving is a retirement, not a deletion.

Where to go next

Before going live, make sure the numbers you validated were the right ones: How to read your backtest. To understand the price gap that now defines your results, read Value betting and CLV. And when the live record starts looking respectable, Publishing your strategy covers taking it to the marketplace.