Traders often worry that a bot's backtest and its live account behave differently. The agreement score checks that, every day.
What it claims, exactly
Each live session is run again, separately, as a backtest of the rules your account was running, over the same prices. The score is how closely the two agree, within stated allowances for fills.
That is the whole claim. It does not measure whether you made money, and it does not rate your strategy. It runs through the day too, about twenty minutes after each trade closes.
What is counted
Every trade that should have matched is counted. There are three kinds:
- a pair: live and the backtest both took the trade
- a backtest-only trade: the backtest took it and live did not
- a live-only trade: live took it and the backtest did not
The score is the pairs that agreed, out of all three kinds together. So a trade on only one side counts as a miss, the same as a pair that disagreed. One matching pair plus eight backtest-only trades reads as 1 of 9, not as perfect.
A pair agrees when it matches exactly, or when the difference sits inside a written allowance: measured fees, measured slippage (the gap between the price you wanted and the price you got), and the broker's measured fill delay. Pairs that only agreed within an allowance are counted separately, so you can see how many matched exactly.
What is left out, and still shown
Four kinds of trade are left out of the score. Each is shown on its own instead:
- A pair that differs only in the number of contracts. Your risk per trade is never exceeded. If a fill lands further from the stop, the size comes down to fit. Each side does that from its own fill, so one side can hold one contract fewer and still agree on everything else. This is neither a match nor a miss.
- A setup live skipped because of a safety rule. A news window, an exchange close, a daily cap, or the rule of one trade at a time can only make live do less. Counting those would lower the score for protecting your money. The same goes for a session you had paused or switched off.
- A trade that could not be measured. A gap in the price data is not a verdict, whichever side took the trade.
- A trade made on a different version of Wildflower from the one checking it. Comparing two versions proves nothing. Pairs that agreed leave along with pairs that disagreed, so the score is never pushed up.
Today, and yesterday
The card on your Dashboard starts with the session in progress: how many of today's trades matched, out of how many. The check runs through the day on finished stretches of time. The newest trades show as awaiting comparison, never as matched, because that check has not been made yet.
Under it is the confirmed figure for the last full session, from the run after the market closes. These are two different claims and they are never mixed.
Switching an account off does not remove that day from the check. The after-close run covers every account that was on, or trading, at any point in that session.
What is never scored
A day that could not be measured shows as unmeasured, not 100%. If Wildflower was down, or the re-run did not finish, that day drops out rather than showing green. "We could not look" and "we looked and it was fine" are different, and only the second is good news.
Your broker connection only signs in and takes no trades, so it is never scored.
The score always shows its coverage
You always see how many days were measured and how many trades were checked.
What can go wrong
On a demo account, the trades come from Wildflower's own simulator. So its score checks that the checking itself is consistent, not that broker fills agree. The panel says so.