At 12:49pm a BTCUSDm short hit its profit-lock trigger: floating profit had reached $38.91, so the system armed a guaranteed floor at $27.24 — 70% of that peak, banked, no matter what happened next. That's the whole point of the mechanism: it doesn't cap the trade, it just makes sure a reversal can't erase the gain. What happened next was the reversal never came. Price kept falling for another 24 minutes and the position closed at $103.00 — nearly 3x the number the lock had promised. The floor did its job by not being needed.
Forty minutes earlier, a XAUUSDm short had the opposite session. It opened at 4284.088, touched a real peak of +$5.10 in the first few minutes, and then just sat there — underwater, recovering, underwater again — for close to three hours (10,871 seconds) before the stop finally caught it at −$59.08. No lock ever armed because profit never got close to the trigger. This is what a normal stop-loss looks like when the trade genuinely doesn't work: slow, patient, and it still costs the full amount the risk model set aside for it.
Zoomed out, BTCUSDm carried the day almost by itself: 5 wins, 1 loss, +$104.37 — more than the entire desk's net total. Gold went the other way: 4 wins against 2 losses, but net −$32.16, because one bad stop (the one above) outweighed four small locked gains. Same shape we keep seeing on this symbol — wins arrive in singles, losses arrive in one bill.
The model retrained twice today. First pass: 314 samples, 69.8% held-out accuracy. Second pass a few hours later: 318 samples, 65.6%. We publish both numbers as they landed — no averaging them into something friendlier.