This is a negative result from a seeded synthetic dataset labeled ETH in SuperQuant. The generated candles are not Ethereum market observations. The example is useful for auditing a saved rule, checking a modeled ledger and seeing how limited diagnostics behave when the result is unattractive. It cannot show that a real Ethereum mean-reversion method succeeds or fails.
The CFTC’s virtual-currency risk advisory discusses real-market volatility, platform safeguards and the way margin can amplify losses. This synthetic replay contains no real venue or platform observations and cannot measure those risks.
Freeze the question and rule
The internal audit record identifies SQ Engine 1.0.0, seed 241 and 2,160 generated hourly bars from 22 June through 19 September 2026 UTC. The ETH mean-reversion template used 14-period RSI and 30/70 thresholds in both directions, a 1.5% stop, 3% target, 1% risk input, 3× leverage cap, 5-basis-point fee, 2-basis-point slippage, generated hourly funding, a 96-hour maximum hold and a six-hour cooldown. The parameters were not searched or refit for this example.
The engine’s specific rule waits for a completed hourly bar to cross back through an RSI threshold, then models an entry on a later bar. That is not interchangeable with every “buy oversold” rule or with a price-to-average strategy. Signal timing, exits, position size, and instrument mechanics can all change a ledger. The repository test reruns this fixed configuration from the current engine and seeded data; that verifies software consistency, not market validity.
Inspect the exact synthetic input
The full 2,160-row candle input used in this example is now available as a synthetic ETH-labeled hourly CSV. It includes timestamps, OHLCV and generated funding-rate fields. The resource page shows separate SHA-256 checksums for the canonical candle data and CSV file; the canonical data checksum matches the hash stored with this research record. Import the rows as hourly data and compare them before changing any fields. The ETH label describes the software template only.
The CSV makes the input inspectable; it does not contain the private trade ledger or every internal diagnostic. In this repository, regression reconstructs the fixed run with the current engine and confirms the stated output. That is a software consistency check, not independent replication by an outside reader, proof of exchange execution, or evidence about real ETH. If the source code, engine version, rule, or input hash changes, the result needs a fresh review.
Reconcile the modeled result
The replay starts at $10,000 and records 40 modeled trades, with a 22.5% win rate. Gross price P&L is −$1,221.54. Modeled fees subtract $219.54, while generated funding adds $3.95. The arithmetic reconciles to −$1,437.13 net, or −14.37%, and ending equity of about $8,562.87. The largest recorded peak-to-trough drawdown is 14.70%. Thirteen losses occurred consecutively at the longest losing streak.
The same synthetic interval’s simple price-change baseline is +92.53%. It has no matched sizing, leverage, trading costs, funding, drawdown path or execution assumptions. It is not a fair comparison with the strategy ledger; it only shows why a benchmark must answer the same question under comparable assumptions.
Test the stated cost range
The internal execution-cost surface reruns the same 40-trade record across nine combinations: half, base and double the configured fee, paired with half, base and double the configured slippage. Across that finite grid, modeled net return ranges from −13.65% to −15.67%. No cell becomes positive. This says only that the negative result persists under those particular model adjustments; it does not estimate exchange costs, prove a real strategy has no edge, or cover latency, order-book depth, funding changes or market impact.
Read resampling as a question, not a forecast
The record also resamples the 40 realized synthetic trade outcomes with replacement 1,000 times. The final-return 5th, 50th and 95th percentiles are −23.57%, −14.47% and −4.58%. In this particular resampling exercise, 99.1% of samples finish negative. That percentage is not a 99.1% probability of losing money in a future ETH trade. Each resample reuses outcomes from one already-negative, generated run; it cannot add independent market histories or repair serial dependence, regime changes, selection effects or execution gaps. Efron’s foundational paper describes bootstrap resampling as a method for studying variation from observed data, not as a way to turn one synthetic series into new market evidence.
Resampling individual trades also discards their original order. It cannot preserve a loss cluster, changing exposure or a market regime simply by drawing more samples. A time-series question may justify a separately specified block-bootstrap sensitivity: the arch 8.0.0 documentation distinguishes stationary, circular-block and moving-block methods. Any block length is another assumption to record before examining the answer, and resampling one synthetic path still cannot create new ETH regimes or observations.
The four fixed chronological windows are also small: each contains only three to eight closed trades. Two are positive, but their mean is −0.54% and their range is −3.75% to +1.43%. These are narrow descriptions of this generated history, not confirmation of an ETH effect. The input is now downloadable, but the trade ledger and diagnostics remain internal; the checks described here verify the article against that internal record, not a complete outside replication from public files alone.
What a real ETH study would require
Use authenticated, timestamped data for one named ETH spot or perpetual market. Predeclare the RSI calculation, completed-bar signal, entry convention, exits, risk limit, benchmark and later untouched evaluation period. Reconcile fees, spread, slippage and funding to the venue, preserve losing alternatives, and report trade counts, outcome distribution, drawdown and limitations together. A sensitivity grid should use documented or justified assumptions; do not treat arbitrary multipliers as a venue cost estimate. Keep input lineage beside each result: venue, exact pair, interval, retrieval date, missing-bar policy, funding source, engine version, strategy hash and every parameter tried. A downloadable synthetic series can test an importer or replay path, but it cannot substitute for those market-data checks.
The CFTC’s current virtual-currency advisory discusses risks in spot, futures and options markets and notes that leverage amplifies futures risk. Its U.S. regulatory scope and general risk discussion do not replace local rules or a specific contract’s documentation. A chart stop does not guarantee a fixed realized loss, and no historical average predicts the next trade.
Continue with the mean-reversion guide, RSI rule guide and overfitting review. This is a generated software example, not Ethereum market research, financial advice or a return estimate.