How we decide a signal is real.
The same artifacts that gate our live trading are what we publish. This is the method — community hypotheses turned into reproducible tests, walk-forward gates, a C++ real-fill L3 simulator as the only authority for any P&L claim, RIGOR adversarial review, dispersion-first reporting, and a chain you can verify yourself.
*ZN is carried as a deliberate control: a non-retail instrument where the method should — and does — produce a clean zero. A method that can't find "no edge" where there is none can't be trusted where there is.
The evaluation loop
-
Intake hypotheses from the desk
Ideas can come from the internal AI research loop, member observations, failed trades, structural market notes, or the falsification library. They do not become signals by opinion; they become explicit hypotheses with a named mechanism, a known failure mode, and the data needed to test them.
-
Detect at tick resolution
Opening-range breaks/flips, prior-day & overnight level crosses (PDH/PDL/PDC, ON-H/ON-L), and retest state machines are detected from the order book seconds ahead of bar-close-watching retail — not approximated from public OHLC.
-
Gate each event for quality
Energy gate: trade only when the opening range is wide (top-tercile) or pre-break volatility is elevated — fades want energetic, expanded tape, not compression. Confluence gate: veto when another level sits ≤20 ticks away (stacked levels were the single worst-performing cell, −13.0t, and stable across two years).
-
Walk-forward regime gate (the on/off switch)
Each family trades iff its own trailing-K-session P&L is positive — a strictly causal rule that uses only sessions already closed. A family that goes cold is benched before the cold streak, on the record. This is the literal mechanism behind "honesty as a feature," not a marketing label.
-
Score against a real-fill L3 simulator
Signals are not P&L. Every economic claim is run through a passive-touch FIFO fill simulator on real Level-3 (MBO) order data — queue position, trade-through price priority, latency, fees, and marketable exits. The simulator is the only authority; a setup that looks great on mid-price marks but doesn't survive real fills is not deployed.
-
Run RIGOR before anyone celebrates
RIGOR attacks every conclusion before it reaches the product: split the populations before averaging, treat regime as the first partition, compare against the dumbest baseline, check sample power, name the causal mechanism, and log negative ablations. The point is not to validate the idea; it is to find the crack before the market does.
-
Re-rank every fortnight
At each 10-session boundary the whole book is re-scored on a rolling window and re-ranked by current-regime numbers. We trade the intersection of (gate ON) ∧ (positive recent lift) ∧ (hot on the trailing window). The book is never "set and forget."
-
Report dispersion, not just means
Means hide the risk. Every published figure carries per-event / per-day standard deviation, max drawdown, and losing-streak length. The per-event edge is honestly thin and regime-conditional; we sell timing, levels, and a discipline gate — not a money printer.
What members are joining
Infrastructure as leverage
The expensive part is not the chart idea; it is clean MBO data, feature caches, replay correctness, parallel sweeps, holdout discipline and production monitoring. qMachina carries that burden and publishes the artifacts.
AI-accelerated research
AI agents help explore hypotheses, build detectors, generate review artifacts and surface suspicious assumptions. Human engineering still owns the protocol: no proxy fills, no hidden tuning loops, no result without a reproducible path.
Open-source quant community
The goal is to make research workflows that used to live inside closed hedge funds and quant shops legible to a broader community: public records, falsified ideas, verifiers, dashboards and a shared research queue.
What the walk-forward gate produces (hypothetical)
| Family | Gate | Gated result | Coverage | Win rate |
|---|---|---|---|---|
| RETEST-FAIL | trailing-K10 PnL>0 | +28.8t/day | 120 / 212 traded days | 0.56 |
| ORB-ANTI (flip) | trailing-K10 PnL>0 | +9.8t/day | 247 / 510 traded days | ≈2× ungated |
| PDH-FADE | trailing-K21 PnL>0 | +16.7t/day | — | — |
Honest caveats, stated up front: K∈{10,21} is lightly post-hoc; ungated ORB-anti averages only ≈+2.6t/event over two years — the exceptional months are exceptional, not typical; the edge is regime-conditional (by year: 2024 −7t / 2025 −2t / 2026 +26t). A magnitude floor (gate on trailing net > +40–50t, not merely >0) lifts the carrier book from −$4,555 to +$18,180 with improved drawdown (SIMULATED, 3-lot).
Verify the record yourself
Every signal is hash-chained at emission time — the instant it fires, before the outcome is known. Each record commits to the previous one, so the history is tamper-evident: nothing can be altered, inserted, or back-dated without breaking the chain.
Each line of record/track_record.jsonl is:
Re-walk the chain and confirm no record was altered or inserted:
Exit code 0 and "chain verified" means intact. The rolling chain head is posted publicly (X / git) so the timestamps cannot be back-dated — a third party timestamps it for us. See the live track record →
Required Disclosures
CFTC RULE 4.41 — SIMULATED PERFORMANCE
HYPOTHETICAL OR SIMULATED PERFORMANCE RESULTS HAVE CERTAIN LIMITATIONS. UNLIKE AN ACTUAL PERFORMANCE RECORD, SIMULATED RESULTS DO NOT REPRESENT ACTUAL TRADING. ALSO, SINCE THE TRADES HAVE NOT BEEN EXECUTED, THE RESULTS MAY HAVE UNDER-OR-OVER COMPENSATED FOR THE IMPACT, IF ANY, OF CERTAIN MARKET FACTORS, SUCH AS LACK OF LIQUIDITY. SIMULATED TRADING PROGRAMS IN GENERAL ARE ALSO SUBJECT TO THE FACT THAT THEY ARE DESIGNED WITH THE BENEFIT OF HINDSIGHT. NO REPRESENTATION IS BEING MADE THAT ANY ACCOUNT WILL OR IS LIKELY TO ACHIEVE PROFIT OR LOSSES SIMILAR TO THOSE SHOWN.
Trading futures involves substantial risk of loss and is not suitable for all investors. Past or hypothetical performance is not indicative of future results. This is impersonal, broadcast information, not individualized trading advice.
Read the complete required disclosures — Rule 4.41, full futures & options risk, the impersonal-publisher statement, and data/simulation limitations — on the Disclosures & Risk Disclosure page.