Trade against the crowd — when the data says so.

Re-armed October 6, 2026 — third live cycle, opponent-conditioned. The matrix now conditions every state on the strategy's own trailing record. A retroactive test over 463,802 resolved outcomes confirmed the thesis the project started with: extreme winning streaks mean-revert and deep losing streaks persist — fading a martingale on a hot streak measured −2.2 pips forward over 98,687 visits. Live execution resumed under the strictest rules yet: fade-direction only, 1h/4h horizons only (the 24-hour tier stays demo-only), no position held into a weekend, fixed 100-unit size. Qualification was seeded by re-scoring the full two-month forward record with each strategy's point-in-time trailing state — no price data was re-simulated, but the conditioning thresholds were calibrated on that same history, a bias the walk-forward and per-day gates only partly remove. The live book stays small while fresh forward data confirms or retires each cell.

Live execution paused September 16, 2026 — second live cycle closed. The overlay executor traded live September 9–16 on fade-direction cells only: 73 trades, 34% win rate, −5.3% on the account. The failure was specific and measurable: every 1-minute / 24-hour-horizon cell realized 25–50 pips below its measured edge, while 4-hour-horizon cells tracked their measurements. The 24-hour tier counts overlapping visits as independent evidence — the same flaw that ended the August cycle — and it is being removed from live eligibility before any re-arm. Demo signals and the measurement matrix continue uninterrupted. Every number on this page remains forward-tested only: it accrues after the strategies go live, on live data. No backtests are published here, and none ever will be.