A backtest example is useful only when you can tell what it tested. This walkthrough audits one SuperQuant BTC-labeled template against the exact seeded demonstration record in the repository. The candles look like hourly Bitcoin data in the interface, but every price and funding value was generated for software practice. Nothing here measures Bitcoin’s historical or current performance.
Freeze what the run means
The research record contains 2,160 synthetic hourly candles from 22 June through 19 September 2026 UTC, generated with seed 73. Its SHA-256 data hash and the engine and data-generator file hashes are recorded alongside the run. The short engine fingerprint is only an accidental-mismatch check; it is not a cryptographic integrity proof.
The submitted rule uses a 20-hour simple moving average and a 50-hour simple moving average. It reads the previous completed candle for a crossover, allows long and short positions, and models entry at the following bar’s open. The template also sets a 2% stop, 5% target, 1% risk input, a 3× leverage cap, a 96-hour maximum hold, and a six-hour cooldown. The replay applies the configured 5-basis-point fee and 2-basis-point slippage inputs and includes generated funding. These are one template’s assumptions; the 1% risk input is not a guaranteed maximum loss.
Before reproducing a result, record the template version, every setting, data identity, coverage and timestamp convention. If you change a setting after looking at results, treat the next run as another research choice. Bailey and co-authors’ paper on the probability of backtest overfitting discusses the risk of selecting strategies after repeated investment backtests. It is relevant as a general research warning; SuperQuant’s walkthrough does not calculate that paper’s PBO measure.
Read the result against its limits
This replay begins with $10,000 and finishes at $10,202.21: a 2.02% net result in the generated sample. It records 23 modeled trades, 34.8% winners, 4.46% maximum drawdown, $106.94 of fees and $33.04 of funding. Gross trade P&L is $276.11; after the recorded costs and funding, net is $202.21. The exact figures are reproducible from the source artifact and current engine.
The engine also reports a 12.74% simple price-change reference over its measurement window. That line does not match the strategy’s changing exposure, leverage, trade timing or fees, so it is not a fair portfolio-versus-strategy comparison. It is still a useful prompt: this particular net strategy replay ended below the reference. That result should invite questions about the rule and measurement, not be hidden behind a positive headline.
Zero modeled liquidations do not establish that the strategy is safe. The research record explicitly says liquidation uses trade-price OHLC as a proxy for mark price and a flat maintenance fraction; it omits venue tiers, cross margin, auto-deleveraging and insurance-fund behavior. The same record excludes order-book depth, queue position, latency, partial fills and market impact. Stops can gap, and same-bar exit ordering follows the engine’s conservative adverse-exit convention. These boundaries make the output a software and assumption rehearsal, not a venue-validated execution study.
Ask whether the fixed rule survives small checks
The engine’s chronological diagnostic applies the same submitted rule to four later windows without refitting its parameters. This record reports +0.41%, +2.54%, +1.52% and +0.41%, with only 2, 3, 3 and 4 trades. Every window is positive, but twelve trades in a generated series are far too little evidence to infer a durable Bitcoin effect. The windows have also been inspected and recorded here; they are no longer untouched holdouts for a researcher who reads this page and then tunes against them.
A separate trade-order diagnostic resamples the 23 observed net trade-dollar outcomes with replacement 1,000 times. Its 5th, 50th and 95th percentile final returns are −6.08%, +1.59% and +11.27%; 37.9% of those resampled sequences finish below zero. This percentage describes that particular resampling exercise. It is not the chance that a live strategy will lose: the procedure reuses the same small set of trades and does not model serial dependence, new regimes or execution conditions.
The nine-cell cost surface changes only the configured fee and slippage assumptions. At twice each baseline input (10 bps fees and 4 bps slippage), the replay reports +0.56%; at half each, it reports +2.83%. That is a bounded sensitivity check, not a complete execution stress test. It does not add order-book depth, latency, queue priority, funding changes or market impact.
Reproduce the walkthrough, then replace the data
In the Research Lab, select the BTC trend template and confirm the data provenance is marked synthetic before running it. Keep the submitted settings fixed, then inspect the trade ledger: compare each crossover signal’s completed-bar time with its modeled next-open entry, and reconcile price P&L, fees, funding and net. Open the disproof diagnostics and record which assumptions each one changes. Do not choose the most favorable diagnostic and present it as the main result.
To study a real Bitcoin question, start a separate experiment with a source whose market, contract, timestamps and coverage you can verify. Preserve the original file and a content hash; inspect missing bars, duplicate timestamps, OHLC consistency, timezone and funding units; and state how you model fees, spread, slippage, partial fills, stop gaps and liquidation. Reserve later observations before choosing parameters, and write down when any holdout has been viewed or reused. The public backtesting audit checklist and guides to crypto backtest costs, walk-forward testing and performance metrics provide the next checks.
Download the exact generated 2,160-row hourly input as BTC-labeled synthetic backtest data. It matches the built-in example series and can be checked against the resource page’s SHA-256 fingerprints. The detailed research record at content/research/bitcoin-trend.json remains internal, so the public file exposes the candle input, not the saved result ledger and diagnostics. SuperQuant’s BTC label is only a software symbol; every price and funding value is generated. This page demonstrates how to inspect one research workflow and offers no Bitcoin forecast, trade recommendation or evidence of a future edge.