This case study audits a SuperQuant breakout template on deterministic synthetic candles labelled SOL. The generated series is not Solana market history. Its purpose is narrower: show what the saved rule, modeled ledger and diagnostic outputs say, including the results that a headline return can hide.
Freeze the example before reading its result
The internal research record identifies engine SQ 1.0.0, a seeded generator, 2,160 synthetic hourly bars from June 22 through September 19, 2026 UTC, and seed 321. The example checks a completed bar against the previous 20-hour range, then models entry at a later bar open. It uses a 2.5% stop, 6% target, 1% risk input, a 3× leverage cap, 5 basis-point fee, 2 basis-point slippage, generated funding, a 96-hour maximum hold and a six-hour cooldown. The record and repository tests reproduce the calculation internally; the raw research artifact is not part of the public container, so visitors cannot independently rerun this exact seeded sample from the page alone.
| Part of the rule | Recorded setting | What a reader should keep in view |
|---|---|---|
| Signal | Prior 20 completed hourly bars; both directions | Entry is modeled after the breakout is confirmed |
| Exit | 2.5% stop, 6% target, 96-hour maximum hold | Bar-only data cannot show the true intrabar path |
| Risk control | 1% risk input and 3× leverage cap | These are engine parameters, not a live account guarantee |
| Friction | 5 bp fee and 2 bp slippage; generated funding | Neither is a measured SOL execution or funding record |
| Replay | 2,160 bars; seed 321; six-hour cooldown | One deterministic path is not a sample of market regimes |
The engine fingerprint is a compact accidental-mismatch check; the record also stores separate SHA-256 digests for the engine, generator and data. That supports a reproducibility check inside the repository, but hashes do not validate whether generated data resemble a real market or whether assumptions match a venue.
The full synthetic replay begins with $10,000 and records 41 modeled trades. Its gross result is $1,157.76. Fees subtract $163.78 and generated funding subtracts $1.42, leaving $992.56 net, or 9.93%. The reported maximum drawdown is 4.51%, with a 39.0% win rate. These numbers are rounded from the saved run; they describe this generated series and model only. A 39.0% win rate does not contradict the positive net result: the size and timing of gains and losses matter too. The stored profit factor is 1.85, and the engine’s annualized hourly Sharpe is 4.16 under a zero risk-free input and square-root-of-hours scaling. Neither statistic has a real SOL return series behind it. Annualizing synthetic hourly bars does not turn them into live performance evidence. The backtest metrics guide explains why sample size, return frequency and formula convention matter when interpreting these measures.
Read the stress checks for what they are
The fixed-rule cost surface changes only fee and slippage inputs. At half the baseline fee and slippage, the same 41-trade sample returns 11.41%; at the stated baseline it returns 9.93%; at double both costs it returns 7.23%. This shows how the modeled answer moves under those selected cost changes. It does not test queue position, depth, latency, price impact, altered funding, new market regimes or real venue fills.
Four chronological diagnostic windows produce two positive returns, and each contains only four to six trades:
| Window | Net return | Modeled trades |
|---|---|---|
| 1 | −1.49% | 6 |
| 2 | −1.13% | 4 |
| 3 | +4.15% | 6 |
| 4 | +1.36% | 6 |
Their mean return is 0.72%; the weakest is −1.49% and the strongest is +4.15%. These windows use fixed parameters on the same synthetic history. They are not four independent markets or evidence about future SOL conditions. A high result in one short window does not outweigh the weak ones or substitute for a market-relevant sample.
A separate diagnostic resamples the 41 realized trade outcomes with replacement 1,000 times, using the basic bootstrap idea of drawing repeated samples from observed outcomes. In that run, the 5th, 50th and 95th percentile ending returns are −0.57%, 9.91% and 21.82%; 6.3% of those resampled paths finish below their starting balance. This distribution asks what re-orderings of this exact set of simulated trades look like if outcomes are treated as exchangeable draws. It is not a 6.3% forecast of loss: resampling does not reproduce serial dependence, a different market regime, execution uncertainty or future trades. The original bootstrap paper by Bradley Efron introduces resampling as a statistical method; it does not establish that this trade sequence is independent or representative.
Keep the benchmark mismatch visible
The same synthetic interval’s simple price-change baseline is 63.38%. That number has no matched position sizing, leverage, fees, funding, drawdown path or execution model, so it cannot be compared as if both were equivalent portfolios. Start by choosing a benchmark suited to the research question, then disclose differences in exposure and cost treatment. This price-change figure is not proof that either method would perform that way with SOL.
The example also uses trade-price OHLC as a mark-price proxy, flat maintenance assumptions and bar-level exit ordering. It has no order-book depth, queue position, latency or partial-fill model. A real study needs authenticated, timestamped SOL spot or perpetual data, venue-specific costs and an execution model suited to the claim. Write down the instrument, venue, data source, UTC coverage, funding field, mark or trade-price field, decision delay, risk model and cost range before calculating a result. Preserve the exact rule and data version, report every chronological window and rejected alternative, and inspect whether realistic costs or contract units change the conclusion. If a necessary input is unavailable, label the claim untested rather than silently substituting a default. The data-quality checklist, backtest cost guide and audit checklist provide reusable review steps.
Read the breakout rule guide, cost sensitivity guide and data-quality checklist. This is a synthetic software audit, not a Solana finding, live fill record, trading signal or investment advice.