We changed the scoring rule again, and this time the number went up

This morning we told our subscribers we had lowered our own win rate from 68% to 60%, and we wrote this down:

We are not going to change this again because a different rule would look better next quarter. The exit rule is now part of the published methodology, and if it ever changes again it will be for a structural reason, stated here, with the numbers before and after.

We are changing it again tonight, and the number goes up. So this post owes you the structural reason and the numbers, and it should be read with appropriate suspicion.

What we found

Every signal we publish carries an entry, a stop and up to three targets, taken from the range the breakout came out of. We score a win at TP2, the top of the range, because that is what a breakout is for.

Some signals do not have a TP2.

Before a signal is published, its targets are filtered: a level on the wrong side of entry is dropped, and so is one that sits so close to entry that hitting it means nothing. On a tight range that filter can leave a single usable target — TP1 — and the signal ships with one level to aim at.

We were scoring those signals against TP2 anyway. A target they never had. They could not win, no matter what the price did.

There are 13 of them among 175 resolved signals. Nine reached their only target and were recorded as losses.

What we changed

A win is now the target the signal was actually aiming at. TP2 where there is one — unchanged, that is still the top of the range and still the bar for almost every signal we publish. Its only target where the filter left just one.

That is one rule, not an exception bolted onto another rule. And it does not soften the change we made three days ago: a signal that has a TP2 and does not reach it is still a loss. That was the whole point of moving from 68% to 60%, and none of it is undone here.

The numbers

On the same 175 resolved signals, scored both ways:

Our published figure was 60%, measured on a slightly smaller set a few days ago; on today's data the old rule gives 61%. So the rule change is worth about 5 points, and the rest is the ordinary drift of new signals resolving.

Nothing was deleted, nothing was re-run, no losing trade disappeared. The same furthest_tp drives everything. What changed is which target each signal is measured against.

Why we are telling you this way

A correction that raises your own headline number is the kind most companies find a reason not to mention. We said three days ago that the next change would come with the reasoning and the before-and-after, and the test of that promise was always going to be the first change that flattered us.

If you would rather judge it yourself: the signals with no TP2 are the ones where the range was tight enough that only one level survived the filter. You can see every signal, open and resolved, on the track record.

We would rather explain a number that went up than quietly enjoy it.

What we are still not doing

We are not going to keep tuning this. The rule is now stated in one sentence and covers every signal without a special case, which is the property we were missing. If it changes again, you will get another one of these, with the numbers, whichever way they move.

We also still are not telling you when to exit. We publish the entry, the stop and every target; how you manage a position is yours. What we owe you is a precise statement of how our own scoreboard is calculated — and, when we get it wrong, a post like this one.

⚠️ Not financial advice.