Six of fifteen signals were gone by the close

For most of July our scanner ran one hour before each market's close. A setup that fired at that moment went straight into the subscriber email. Getting it out before the day ended was the whole point.

Then we measured what happened to those setups during the hour that followed.

What nine days of telemetry showed

From 16 to 24 July we ran a second scan at the actual close and compared the two lists.

The first version of that comparison was wrong, and it is worth saying how. It counted every setup the scanner detected, including the ones our own filters drop before anything is sent to anyone. Corrected to the signals that were actually published:

• 15 signals published in that window.
• 6 of them no longer met the setup at the closing bell. The breakout had been pulled back inside its range by the time the day was scored.

There is a mirror image to that, and we can only describe it rather than count it. Setups that were absent an hour early and present at the close were logged too — but only at the detection level, before the filters that decide what actually gets published. So we know that bucket exists and we do not know how many of them would have reached anyone. We are not going to attach a number to it, for exactly the reason this post exists.

Fifteen signals is a thin sample and we will say so plainly rather than lead with a percentage. But the direction is not subtle: an hour before the close is not when the day is decided.

Why an early scan produces phantoms

A range breakout is defined by where price settles, not by where it trades. Between our scan and the closing auction an instrument can cross the level, get sold back inside the range, and close as though nothing happened. The setup is real at 15:00 and gone at 16:00.

Published early, that signal enters the track record as a trade nobody could have taken on the terms we described. For a product whose only real asset is a verifiable record, that is the expensive kind of wrong.

The mirror image costs just as much. An instrument that spends the day inside its range and breaks out into the close is exactly the setup the strategy is built for — and a scan that finished an hour earlier cannot see it.

What we changed

Every market is now scanned 20 minutes after it closes. Not before.

A setup reaches the email only if it is still there on the closing bar. There is no longer a window in which a signal can be published and then quietly stop existing.

For subscribers the trade-off is explicit, and we would rather state it than bury it:

You get the signals later in the day. You stop getting the ones that evaporate. The first moment any of them can actually be traded is the next session's open, which was true under both models.

Crypto, foreign exchange and commodities never had this problem. They were already scanned after their daily candle closed.

What did not change

Nothing was removed from the track record. No past signal was rescored, deleted or quietly reclassified because of this — the fix governs what gets published from here, not what was published before.

We also did not loosen any filter to make up the lost volume. The same rules apply. There are simply fewer phantoms in the output and a few more real setups.

What we are still measuring

Nine days and fifteen published signals justify a fix. They do not close the question. The scan-at-the-close model now produces the only data we will judge it on, and if it turns out to carry costs we did not anticipate, that gets written up here the same way this did.

We would rather publish fewer signals that were real than more that were not.

Every signal, open and resolved, is on the track record.

⚠️ Not financial advice.