Refine towards the moves you mark
Most people refine a strategy by hand the same way: change a condition, look at where the matches land on the chart, try again. Refine's fourth goal, Entries where you aim, runs that loop as a search. You paint the moves you wanted to be alerted in and the stretches you did not, and every variation is scored on where its matches fall.
How it differs from the other goals
The other three Refine goals score each variation with a backtest: every match opens a trade against a fixed stop and target, and the score is a composite of what those trades did. This goal simulates no trades. It only asks where the matches landed, and it judges that against areas you drew. That is the same question the entry check asks — did the entry fire at the right moments — with you saying which moments were right, instead of forward returns saying it.
Painting the areas
- Open a chart. Pick the market and a timeframe to draw on. The timeframe is only a lens: an area is stored as two instants, so it means the same whatever chart it was drawn through, and the strategy's matches are always computed on its own conditions' timeframes. You can open up to five markets as tabs, each with its own areas.
- Mark the start of each move you wanted, as Aim. Not the whole move. From just before the turn to partway in. Painting a rally end to end rewards a variation for knowing it is in a rally, which a slow moving average does throughout the leg; painting only its start rewards matching early, which is the useful thing.
- Mark the stretches you did not want, as Avoid. Paint falling legs whole, rallies inside them included. A variation that fires at the start of every rise also fires at every bounce inside a decline, and the Avoid areas are what separate a turn from a bounce.
- Leave everything else unpainted. Unpainted time is ignored: a match there counts for nothing either way. Consolidations, the late part of a move, anything you have no view on.
- Save it if it took a while. A marking is saved separately from any strategy, so the same areas can be run against a different strategy later. Swap Aim / Avoid turns a marking made for a long strategy into one for a short.
As you paint, the panel beside the chart shows where the strategy you are refining matches today: how many in Aim, how many in Avoid, how many outside any area. That is the starting point every variation is compared with. The direction of the strategy is shown at the top, because it decides what Aim means — rises for a long strategy, falls for a short.
How a variation is scored
Matches are counted the way alerts fire: a crossing that repeats within the alert re-arm window is the same alert. Then two things are measured.
- Timing. For each Aim area, how early the first match came — full marks for a match at the start of the area, half at its end, nothing for an area with no match. Averaged over the Aim areas. Extra matches inside the same area earn nothing more, so a condition that flickers on and off does not score higher than one that fires once.
- Confidence. The share of the variation's matching, per trading day, that fell in Aim rather than Avoid time — taken at its cautious end, so that a thousand matches with a few in Avoid outscores a hundred with one. Measured per day so a short Aim area and a long Avoid area are compared fairly.
The score is timing multiplied by confidence, between zero and one. A variation that matches at the start of every Aim area and never in Avoid scores one; one that fires everywhere at the same rate scores about a half; one that fires only in Avoid scores zero. Your trades-a-year minimum still applies, scaled to the Aim time you painted: twenty-four a year against one month of Aim is a floor of two matches.
Reading the result
The result leads with the picture: the chosen variation's matches drawn across the whole window, over your areas, coloured by where they landed. Unpainted time is shown too, so a variation that fits your areas and fires constantly everywhere else is visible as exactly that. The table ranks variations by score and shows, for each, its Aim alerts, its Avoid alerts, its confidence and its matches a year, beside your own strategy's.
Two things on the result keep the search honest. The overview shows how your marked time divides between Aim and Avoid, because alert counts only mean something against it — alerts that divide the same way as the time do no better than chance. And when you painted at least three Aim areas and two Avoid, the app holds one of each back from the search, chosen for you and the same every time for the same marking, and scores every variation on them afterwards. That held-out column is the one to read first: it is the only part of the result the search could not have fitted to.
What to expect
The search reaches for a trend filter first. With rallies painted as areas, a variation that adds "price above a long moving average on the 4-hour chart" matches early in every one of them, and it will usually top the table. That is not wrong — it is what your areas asked for — but it is worth painting shorter, earlier Aim areas and seeing whether it still wins, and worth reading the matches-a-year column. The Fastest timeframe setting stops the search moving a condition to a faster chart than your strategy already uses, which it would otherwise do, because faster conditions match more often.
The risk you should understand
The areas are hindsight. You chose them knowing what the market did, and a search over hundreds of variations fits a handful of areas easily. A high score on the areas you painted is evidence that a variation describes those moves, on that market, in the past. It is not evidence about the next move. Run the winner through an entry check on history outside the areas, and read the held-out column, before you give it any weight.