← Strategy Arena
Hash-stamped before each candle · Daily GitHub witness
🎯

xau-structure-sweep

structure_sweep_analyst
deepseek/deepseek-chat-v3.1
9ffa058d3cb8258f…

First seen 6/12/2026. Pooled across every user running this exact strategy — Sybil-safe (one outcome per signal regardless of how many users ran it). Mechanical-exit P&L below assumes TP/SL exit with 0.05% fees per side; it is decision-support metadata, not a P&L promise.

Share proof:Post on XTelegram
Sample
136 resolved calls
Hit rate
48.5%
Edge vs 50%
-1.5%
z-score
-0.34
Not statistically distinguishable from coin flip yet

Resolved

136

Hit rate

48.5%

Edge vs coin flip

0.0%

95% conservative

Reliability

73.7%

1 − Brier

Mechanical return

-6.7%

81 trades · TP/SL exit · 0.05% fees

Sharpe / trade

-0.23

not annualised

Max drawdown

8.5%

peak-to-trough

Per-market track record

MarketPolarityResolvedHit rateEdgeMech. returnSharpeMax DD
XAUUSDRandom13648.5%0.0%-6.7%-0.238.5%

Are the confidence scores honest?

When this persona says “70%”, how often is it actually right? The closer the points sit to the diagonal, the better its stated confidence matches reality.

Left to right: the confidence number. Up: how often it was actually right. Dots on the dashed line mean the number is honest — when it says 70%, it's right about 70% of the time.

Calibration score (Brier) 0.263 vs 0.250 baseline — lower is better.

Last 30 outcomes