Monday, September 14 2026 recap

BTC closes 17 of 18 clean while the desk rewrites its own stop-loss math

Bitcoin had its best session in weeks. Underneath it, two real engineering fixes went live mid-day — one of them because a single GBP trade closed at a loss with a label that said it shouldn't have.

Net P&L
+$890.23
49 trades closed
Win Rate
87.8%
43W · 6L
Symbols Active
5 / 5
full coverage
Biggest Win
+$150.00
EURUSDm

7-Day Equity

SEP 7 — SEP 14
$11,104.43

For most of today the desk was running on math that quietly worked against it: take-profit targets capped at a flat 50 pips no matter how wide the stop-loss had to be. On gold specifically, where volatility has stops running 200–290 pips wide right now, that meant every win banked around $10 while every loss ran the full stop for $50–70. A real reward-to-risk ratio of roughly 1:5, dressed up in the trade log as "+2.0 RR." That got fixed this afternoon — the target now scales with the actual stop distance instead of a fixed pip count.

The second fix came from a trade that shouldn't have needed one. A GBPUSDm position peaked at $22.50, correctly armed its profit-lock at a $14.62 floor, and then just kept running for another 80 minutes before closing at a $28.25 loss — with the trade log still reading "Profit Lock Hit" the whole time. The close command had been queued correctly; it just never survived an engine restart that happened in between. The retry logic now re-sends that close every tick until the broker actually confirms it, not just once.

“Profit Lock Hit — banked 65% of $22.50 peak (floor $14.62)” — the exact label still attached to a trade that actually closed at a $28.25 loss. Left in the log, not edited out, because that's the bug that got fixed today.

TRADE LOG · GBPUSDM · 13:19 → 15:05

None of that got in Bitcoin's way. BTCUSDm closed 18 trades today and lost exactly one, netting $102.50 on a 94.4% win rate — its cleanest session since we started tracking it here.

The model also retrained twice today, pooling everyone's trades the way it always does. Held-out validation accuracy moved from 69.1% to 67.2% — a real number, published either direction, and this time it went down. We're not going to pretend a two-point dip is nothing or make it sound worse than it is; it's one retrain's noise until it shows a pattern.

By symbol

SymbolTrendWin RateNet P&LNote
BTCUSDm 94.4% +$102.50 18 trades, 1 loss — cleanest session yet
ETHUSDm 80.0% −$0.03 Thin day, roughly flat
EURUSDm 100.0% +$150.00 Single trade, largest win of the day
GBPUSDm 66.7% +$59.75 Includes the mislabeled profit-lock loss, see above
XAUUSDm 86.4% +$578.01 New TP/SL ratio fix ships from today's trades

What changed since yesterday

Both fixes are live now, verified against the exact conditions of the real trades that exposed them before either one shipped. Tomorrow's recap will say plainly if the new ratio changes anything — good or bad.

Recent issues

View all →
SEP 13, 2026

BTC's win streak reaches nine as gold cools off

+$286.90
SEP 12, 2026

A late stop on EUR ends the week's cleanest run

−$34.20
SEP 11, 2026

Profit lock saves four trades from round-tripping

+$198.05
SEP 10, 2026

Retrain day: validation accuracy holds at 69%

+$122.40