9 Edge-Gated Engines · Out-of-Sample Proof · Enforced Risk Limits
Every other tool sells you a pretty win-rate. This one runs a statistical out-of-sample gate on each engine and, when the edge isn't real, says so — plainly. See it for yourself below: the same edge-gate, run on real historical ES bars, showing most engines with no edge and flagging only what actually clears significance. Then connect your own TopstepX or WealthCharts session and the engines re-prove on the bars your feed captures. Signals only — execution is disarmed by default, paper-first, and never leaves your Mac.
A single indicator on a single timeframe will signal all day long. Then the hedge fees, the revenge entries, and the dead strategy nobody turned off quietly erase the account. The fix isn't a better indicator — it's an architecture that refuses to act until many independent checks agree, and that turns its own losing logic off.
Retail bots fire on a single indicator. This engine won't surface a signal for an instrument until that engine has cleared a proven out-of-sample edge gate (binomial test + false-discovery control) — no edge, no signal.
Opposing entries that net to nothing still pay commissions both ways. Direction-lock blocks any conflicting entry while a position is open — one direction, full conviction.
A signal that worked last quarter can decay silently. Your graded results feed live per-engine expectancy analytics, and the Alpha Monitor factor weighs a family down as its measured edge decays — the engine routes around its own decay.
Most vendors show you a painted equity curve. This product keeps candidate research, historical replays, and any user-graded session results separate, and it does not publish a verified-live edge claim.
No emotional decisions, no single point of failure. Here is what each layer of the engine actually does — and what it blocks.
Nine signal engines run on your captured bars — mean-reversion, breakout, momentum, structure, regime and more. An engine's signal is only surfaced for an instrument once that (engine, instrument) pair has cleared an out-of-sample edge gate (binomial test with false-discovery control). No proven edge, no signal — noise never reaches you.
While any position is open, conflicting entries are blocked at the gate level. This eliminates the self-canceling commissions that quietly destroy retail bot returns — one direction, full conviction, until the trade is resolved.
A regime factor scores trend-vs-chop efficiency on every bar, and dedicated context engines classify the session — expansion, trend-day, range. Engine families are filtered by regime so the stack adapts to the tape instead of forcing one playbook all day.
When your captured feed carries real bid/ask prints, the composite scores order-flow toxicity (VPIN — Easley, López de Prado, O'Hara) and cumulative delta. When an input doesn't exist in your capture, that factor renders "no live data" instead of a guess. The engine never pretends to read data it doesn't have.
Every graded fire feeds per-engine expectancy analytics, and the Alpha Monitor factor weighs a signal family down as its measured edge decays. The engine routes around its own decay instead of riding a dead signal.
As price advances from entry to target in six equal steps, the suggested stop trails one step behind — tier one locks breakeven, and each further tier locks progressively more of the move. Winners convert to a locked plan instead of being given back.
The engine shows you why a signal fired — which factors scored and which risk limits passed — while the public proof surface stays empty until an ES candidate clears the current gate. The ledger separates research, historical replays, and user-graded sessions so nothing is dressed up as a track record.
Nine signal engines read your captured ES bars — and each one must clear a statistical edge gate on out-of-sample data before its signals count. An engine that hasn't proven an edge stays dark.
After an engine fires, enforced risk limits validate the signal. Any one red light blocks it before it reaches you.
A 16-factor composite scores every setup — and it is honest by construction: a factor contributes only when its real input exists in your capture. Anything without a source renders "no live data" instead of a guess.
Trend (higher-timeframe bias), Momentum (rate of change), StepGMA (stepped moving-average alignment), Kalman Trend (filtered trend velocity), Breakout, Key Levels, Market Structure, Regime (trend-vs-chop efficiency), and Session (time-of-day weighting).
Volume (participation vs average) and VWAP (distance from / reclaim of the volume-weighted price) score only when your capture carries real volume.
CVD Flow (net aggressive pressure), CVD Divergence (price vs delta disagreement), and VPIN (order-flow toxicity — Easley, López de Prado, O'Hara). Present only when your feed carries real bid/ask prints; dark otherwise.
Alpha Monitor reads your real fire journal to weigh edge decay. SMT (correlated-asset divergence) ships dark — it needs a correlated instrument on your feed before it will ever score.
Black Label keeps candidate research, historical replays, paper execution, and user-graded session results separate. Until an ES candidate clears the current gate on eligible ES data, the public proof state stays empty. Results to date are paper/research. No win-rate, expectancy, or P&L figure on this page is a promise of future returns — trading futures carries substantial risk of loss.
Below is the app's actual out-of-sample edge-gate — the same statistical test your own engines use — run on real, WealthCharts-captured historical ES bars. Most engines show no edge: the gate refuses to call an edge that doesn't clear one-sided binomial significance out-of-sample. Only an engine that genuinely earns it is flagged an OOS candidate. This is the whole point — a tool that tells you when it has nothing.
| Engine | Verdict | OOS trades | Win rate | Net (pts) | Significance | ES contract · window |
|---|---|---|---|---|---|---|
| Mean Reversion | NO EDGE | 314 | 83.4% | +38.9 pts | p=0.720863 | CM.ESU6 · 2026-06-19 → 2026-07-03 |
| Breakout | NO EDGE | 3 | 66.7% | +28.5 pts | p=1.0 | CM.ESU6 · 2026-06-19 → 2026-07-03 |
| Research | NO EDGE | 3 | 66.7% | +28.5 pts | p=1.0 | CM.ESU6 · 2026-06-19 → 2026-07-03 |
| Momentum | NO EDGE | 494 | 33.8% | +2.8 pts | p=0.428541 | CM.ESU6 · 2026-06-19 → 2026-07-03 |
| Structure | NO EDGE | 246 | 33.7% | -3.2 pts | p=0.47007 | CM.ESU6 · 2026-06-19 → 2026-07-03 |
| Regime | NO EDGE | 666 | 33.9% | +8.7 pts | p=0.3852 | CM.ESU6 · 2026-06-19 → 2026-07-03 |
| Channel | NO EDGE | 9 | 44.4% | +2.2 pts | p=1.0 | CM.ESU6 · 2026-06-19 → 2026-07-03 |
| Context A | NO EDGE | 494 | 33.8% | +2.8 pts | p=0.428541 | CM.ESU6 · 2026-06-19 → 2026-07-03 |
| Context B | OOS CANDIDATE | 191 | 39.3% | +24.9 pts | p=0.049508 | CM.ESU6 · 2026-06-19 → 2026-07-03 |
8 of 9 engines show no edge on this ES window — including Mean Reversion, whose 83% win-rate the gate rejects because it can't beat its own risk:reward breakeven with statistical significance. Only Context B clears the one-sided binomial gate (p = 0.0495 < 0.05) on this contract — and it does not clear on the earlier ES contract, so it is a candidate, never a guarantee. That 0.0495 is a raw, per-test value: because 9 engines were screened, the honest multiplicity-corrected figure is the Benjamini–Hochberg FDR q = 0.45 — the same across-the-grid correction the in-app screener applies — so Context B does not survive multiple-comparison correction, and seeing one engine near p = 0.05 out of 9 is expected roughly 37% of the time by chance alone. This is precisely why it is flagged a candidate, not an edge. We publish no aggregate win-rate, no blended figure, no equity curve, and no verified-live edge claim. Reproducible: 9 engines · prover e464bd505b004c3e · generated 2026-07-11. Trading futures carries substantial risk of loss; past results do not guarantee future performance.
Reference computation on historical ES bars — not your account, not live, not a promise. Paper / default-off execution. No verified-live edge is published; past results do not guarantee future performance.
You already trade the e-minis and want a disciplined second opinion that never fires on noise and never lets a dead setup keep firing.
You're working an evaluation with daily-loss and drawdown rules. Direction-lock and the drawdown guard are designed to keep you from the behaviors that blow evals.
You want to see why a signal fired — which modules voted, which gates passed — and an honest, re-runnable view of whether an edge survived out-of-sample.
You own the workflow: it runs on your own macOS machine, against your own feed, and shows its work. Execution is disarmed by default and paper-first, so there is no custody of your account and no live order is placed unless you explicitly arm that path.
One subscription, cancel anytime. The engine reads your TopstepX ES session through the bundled local bridge, writes an honest local ledger, and keeps live execution disabled until you explicitly arm it.
Cancel yourself anytime; access runs through the period you paid for.
macOS · Apple Silicon · Developer ID signed outside the App Store. The app starts empty, opens the browser bridge locally, and keeps execution off by default. Windows and mobile downloads are coming soon.
Developer ID signed · Apple-notarized · runs on your machine · your data stays with you. Windows and mobile downloads are coming soon.
Not by default. The engine analyzes the market and surfaces gated bracket suggestions — direction, size, stop, target — with execution disarmed and paper-first. Live order placement is a separate step that requires explicit arming, your own broker or platform credentials, and a setup where automation is permitted. We never take custody of your account.
There is no verified-live ES edge published right now, so we don't show you one. The public proof state stays empty until an ES candidate clears the current out-of-sample gate on eligible ES data — that same gate runs inside the app on your own captured data, so you watch proof accrue rather than trust a marketing number. No win-rate or P&L on this page is a promise — trading futures carries substantial risk of loss.
A macOS machine and your own WealthCharts or TopstepX session. The bundled bridge opens your selected platform in a product-owned browser profile and posts observed market data into the local webhook/store. Broker or platform credentials are only needed if you choose to place real orders or explicitly arm live execution.
You own the workflow. It runs on your machine, against your feed, and shows its work — which of the 9 engines fired and which risk limits passed. There's no chat room, no hype, and no inflated track record; the candidate gate stays empty until ES proof clears.
It's built with prop-firm discipline in mind: direction-lock blocks conflicting entries, the drawdown guard halts suggestions after two daily stops, and the position sizer respects a contract cap. It can't guarantee you pass; you remain responsible for every account rule, credential, and execution choice.
9 edge-gated engines, WealthCharts-ready browser bridge, honest ledger — $75/mo, cancel anytime. Execution is off by default and paper-first. No performance guaranteed.