At 01:33 local time, gold's 15-minute RSI dropped to 11.5 — deep into territory the model has logged only once before. That earlier instance ended in the session's worst loss on record, a $200.12 stop-out. So when the same shape reappeared last night, the failure-pattern filter recognized it immediately: a 100% match against the recorded loss, and the trade was skipped before it opened.
Ninety minutes later, the same setup showed up again. Same filter, same result. This is by design — the model doesn't re-enter a pattern it has already lost money on without deliberately testing whether the market has actually moved past it. After being blocked twice in a row, the next occurrence gets let through on purpose, specifically to check whether the danger is still real.
"Skipped: 100% similarity to a past loss pattern." — logged twice, six candles apart, before gold's RSI finally lifted out of the oversold zone on its own.
MODEL LOG · XAUUSDM · 01:33 & 03:07
While gold sat out, Bitcoin did the opposite. BTCUSDm closed its tenth consecutive winning trade of the week, most via the profit-lock mechanism — which banks a portion of a trade's peak floating profit rather than letting it round-trip back toward breakeven. Three of last night's BTC closes locked in 65% of that trade's high point instead of giving it back.
No parameters were adjusted overnight. The model retrained on schedule after its 25th new closed trade, and held its held-out validation accuracy at 69% — a number we'll keep publishing here every time it moves, up or down. Nothing about last night was a surprise the model wasn't already prepared to sit out.