BRYME Money · Open research
95 crypto runs, 9 forex runs, 34 blind tests, one strategy freeze and one self-audit that retracted its own best result. This is what honest strategy research looks like, step by step.
Most trading content shows you a winning backtest and asks for your email. QUANTLAB is the opposite: a public research laboratory where every experiment — including the embarrassing ones — is numbered, logged and open for inspection. This page walks through exactly how the project works, in order, using nothing but its own published record. It is general information about a research process, not advice to trade anything.
QUANTLAB's founding rule is that a failed experiment is still a finding. Across its two research arcs it logged roughly 95 numbered crypto iterations (R001–R095) over twelve days in July–August 2026, 9 forex iterations (F001–F009), and then — after a deliberate strategy freeze on 9 August 2026 — a blind out-of-sample campaign of 34 tests (T1–T34) on a separate branch. Nothing was quietly deleted. The journal, the scripts, the reports and the charts are all in the repository.
Each numbered run is built around a single claim that data can prove wrong: does strategy X, on market Y, survive costs on data it was not designed on? If the answer is no, the run is written down as a no. There is no third outcome where a losing idea gets a softer name.
The crypto research worked on hourly candles across a broad symbol universe, with 2023–2026 split into per-year holdouts — years kept untouched until a verdict, so no result could quietly lean on the data it would later be judged by. Forex ran on 1-hour spot. Five-minute crypto was tested too — and it is one of the project's proudest results that it proved no cost-surviving edge exists there, seven independent ways (runs R089–R095).
Each verdict uses a fixed vocabulary so results cannot be talked up: VALIDATED, WATCHLIST (promising, sample too thin), REJECT, RETRACTED, OVERFIT.
The honest map, from the log: five-minute crypto — no edge; Deriv synthetic indices — indistinguishable from a random walk in that sandbox; forex 1-hour — the crypto trend pipeline transferred at roughly break-even (profit factor ≈1.10, no real edge); most discovered edges proved universe-specific, failing on symbols they had never seen. What survived: a 1-hour crypto mean-reversion family — an SVM-filtered setup with a volatility-ceiling regime gate — plus a VWAP-band lab explored separately.
On 9 August 2026 the project froze its best configuration — and then did the thing almost nobody does: it re-tested the frozen champion blind, on untouched per-year data, with costs. The frozen config failed. Its backtest profit factor of 1.94 collapsed to about 1.25 after costs on the blind re-test, and one static variant lost outright. The freeze did its job: it stopped a fragile result from being trusted.
In September 2026 a test (T34) auditing whether the results were implementable found something uncomfortable: the trend strategy's filter had been reading data from the exit bar of each trade — information that does not exist at entry. That is textbook lookahead bias. Re-anchored honestly to the entry bar, the trend edge fell to roughly break-even (profit factor ≈1.03). The project's response was to publish the retraction, mark the old claims superseded, and make causal audits mandatory for every future run. The framework caught its own error — which is the entire point of having a framework.
As of its September 2026 log, the surviving verdict is deliberately modest: the mean-reversion leg only, at a 2R-target / 1R-stop exit and 1% risk per trade, was the configuration the small-account study could defend — the project's own simulation logged roughly +104% on a $100 account across 2024 to mid-2026 with a realised max drawdown near −37%, while risking 2% per trade produced simulated drawdowns of −50% to −80%. Note what those numbers include: a losing year (2024) inside the winning period. And note the protocol attached: the study re-runs its full validation battery on fresh data at year-end and green-lights 2027 only if the post-cost profit factor holds at roughly 1.2 or better. These are the project's logged research results on historical data — not a promise, not a forecast, and not advice.
Do the maths your own research deserves: the expectancy calculator and the position size calculator cover the two numbers every system lives or dies by, and backtesting 101 explains the testing logic in full.
General information, not financial advice. Everything on BRYME Money is educational. Trading forex, crypto and derivatives involves substantial risk of loss and is not suitable for everyone. Past performance — including any published research — does not guarantee future results. Never trade money you cannot afford to lose.