METHODOLOGY & BENCHMARKS
Research Library
Technical specifications, structural alpha genesis, and capacity scaling dynamics. Every benchmark traces to 127 million tick-by-tick broker records.
Which win rate is the real one
Both figures below are honest readings of the same ledger under different populations. Until the record names its population and as-of date in one place, read every figure with its window attached.
The published record is currently a frozen reading, not a rolling one, and its figures answer to two different questions. Measured from inception (24 Jul 2026) to the published as-of (14 Sep 2026), the ledger carries 50 closed deals, 28 of them wins: 56.0%. Measured cumulatively - every deal closed at or before the same as-of - it carries 64 deals, 29 wins: 45.3%. At the terminal sample of the published file the same two populations read 53.7% and 44.1%. None of these is wrong; they are different windows, and a figure without its window is how three different win rates appear on one visit.
The same holds for the return figure: +41.14% is the equity index at 14 Sep 2026 (index 141.14), not a high-water mark (the index peaked at 146.35 on 24 Aug 2026), and the file's terminal reading is +39.84%. The max drawdown of -11.9% is carried from the published record with its provenance rather than recomputed, because the public deal array carries no per-deal P&L to recompute it from.
This reconciliation is open work, tracked as G2 in the remaining-gaps plan. When it lands, every route reads one named record with its population and as-of date, and this paragraph is replaced by the record itself.
Production Architecture & Empirical Benchmarks
The deployed engine's full specifications: 127M real ticks, 1.52 out-of-sample profit factor, +0.297R gated expectancy, 0.00% ruin across 10,000 paths.
- 01 Verified specifications
- 02 Walk-forward out-of-sample verification
- 03 The 4-layer risk stack
- 04 Monte Carlo ruin probability
Further records
Three more audits of the same engine: where the edge comes from, what keeps fills honest as capital scales, and the losing trades published raw.
- System Genesis: Exploiting Structural Volatility Transitions +237% Why retail indicators decay, how session transitions create predictable coiling, and how volatility compression gates entries for +3.0R targets.
- Capacity & Execution Reliability at Scale 50 Slippage and market impact, why the 50-unit cap protects fill quality, and the 0.80 pt chase limit.
- Loss Forensics: The Ledger Keeps Its Losers 44.0% The losing trades, published raw: four failure modes, -1.0R bounded outcomes, and the machinery that prices them.
- Training window
- Nine months rolling, refit on the first weekend of each calendar month.
- Stress testing
- 10,000 randomized bootstrap paths per record, published with the ruin probability they imply.
- Execution parity
- Live fills reconciled against tick-by-tick broker records, so the published model is the deployed model.
Every number on this site has a paper behind it.
The four records below publish their own benchmarks, their own failure modes and their own capacity limits. Nothing is quoted from a private source, and nothing is rounded up.
Where the record is used
The papers describe the engine. These pages show what it does with your capital.
- The quantitative thesis 5 filters How a raw breakout setup becomes a trade, and why the gate blocks more than it passes.
- The public track record +41.14% Every closed deal on the audited ledger, released under the 24-hour embargo.
- Capacity and scaling 50 seats Why the seat count is a liquidity constraint and not a marketing device.
- The execution playbook 4 sessions Session windows, bracket targets, and the slippage threshold that voids a chase.
- Pricing and access 1 open channel Telegram Smart Signals are available now; the planned Stream API remains non-purchasable.
- Planned Stream API Coming soon The intended authenticated contract and the launch gates before public access is allowed.