A week ago the book was flat after nine stopped shorts. On the weekend of 20 and 21 September the rule turned it long. Five of those longs were cut inside three days. The rest are still open. And in two days at the desk, five ideas for doing better were written down, tested and rejected. Here is all of it.
Quiet, and leaning up. On the measure our staged volatility filter uses, 23 of the 27 markets we track are in their low-volatility regime this weekend; only BCH, LTC, UNI and XRP are loud. Eight markets sit within half an ATR of their 55-bar high (AAVE is on it), and none is near its 55-bar low. Bitcoin is 2.2 ATR below its own high and 8.2 above its low; ether 1.8 and 7.1. The tape is in the upper half of its channels with most of the volatility drained out of it. That is the shape this week's five stops came out of: breakouts that fired one at a time and did not follow through.
The news feed carried between 400 and 590 headlines a day, more than two thirds of them labelled neutral by the classifier, and no single story dominated. We do not trade on it, and this week gave no reason to start.
Issue 03 ended with nine shorts stopped and the book flat. Over the weekend that followed, a run of long signals fired as markets pushed toward their highs, and the rule took them. By Wednesday five of those longs had been stopped, all within three days of entry: the longest held 3.0 days, the shortest 0.8. The public record shows 20 open positions this weekend, the most since the book went live. As always, the count is published and the names are not.
Five longs, five stops, −5.31R in total, about one R each. Each was cut at its initial stop when the breakout did not hold; the largest loss, −1.13R, was a stop that filled through its level on a gap. Average hold 2.1 days.
That makes fourteen closes in a row that were losses: nine shorts, then five longs. The record's longest such run is eighteen. The book is built to survive runs like this; each loss is capped at about one R while the winners are not capped at all, and 22 of the 91 closes so far have paid for the other 69.
The record stands at +12.11R after 91 closed trades (70 days live). Against the probability envelope fixed on 19 July 2026, before the first trade closed, that is inside the band but a band lower than last week: between the 25th and 50th percentiles, just under the median. The full chart is on the Record page.
A losing fortnight makes every fix look obvious, so this week we tested five of them, each the way the method requires: the rule written down exactly, registered with locked pass bars and a prediction, the script reviewed for look-ahead by someone who did not write it, then one run across every trade since 2021.
All five were rejected. Waiting one bar to confirm a breakout removed a quarter of the signals, all of them poor, and still lost a third of the five-year profit: the bar of delay cost more on the winners than the fake-outs cost on the losers. Taking breakouts only in the direction of the long trend removed the regime turns, which are where the best trades begin. Refusing to re-enter after a failed breakout saved nothing, because the repeat attempts were close to break-even. Skipping the most extended breakout bars removed the best entries in the whole record, not the worst. And the one idea from the other side of the coin, selling the first touch of a prior high after an extended run, won 29% of the time and did worse than a coin flip on its own setups: on this tape a touch of an old high is a breakout in progress, not a ceiling.
Those three are the rules staged for the review on Thursday 1 October, and their shadow ledgers kept counting this week: the concurrency cap logged thirteen entries it would have refused that went on to stop out, the volatility filter four, and the early-exit rule flagged one position it would have closed sooner. The ledgers also carry the winners those rules would have refused; weighing both sides is what the review is for, on bars that were written before any of this was seen. Issue 05 will carry the verdicts. The full list of what has been tried is in Tested & Rejected, now twenty-two entries long.
Not investment advice. Past performance does not predict future results. We publish outcomes, not recommendations. Closed trades appear after a 48-hour delay; figures are R-multiples on the same basis as the public record.