Methodology — Overview
This section documents exactly how every number in the API is produced. It is written to be checked against the data: the formulas and conventions below are the ones the exporter and the API actually run, including their limitations.
:::warning Factual historical data — not advice Every figure is backtested / modelled historical performance computed from public market data. Past performance is not indicative of future results. This is factual historical data for professional / institutional use under a global license — not investment advice, not a recommendation, not a forecast, and not an offer. :::
What the dataset is
- 14 strategies, all trading
BTC-PERPETUALon Deribit — an inverse, BTC-settled contract (PnL and balances accrue in BTC; see Provenance). - Backtest window: indicators warm up from 2018-08-14; the published series run to the dataset's generation date (currently 2026-06-24) — about 7.8 years, ~2,871 daily return points per strategy.
- Source: public Deribit chart history (candles). The exporter is keyless — it authenticates to nothing and reads no credentials.
- Deliberately diverse: the strategies are built on different, largely independent return drivers — the source of the book's low cross-correlation.
- Each strategy is sized per-strategy at its own base risk % (
book_size = 1) — this is a single-strategy backtest, independent of any live portfolio allocation.
How a figure is produced
For each strategy the exporter re-runs the engine's keyless DRY_RUN backtest bar-by-bar
over the public candles, producing a mark-to-market equity curve. Everything else is
derived from that curve and the trade list:
- Returns — last equity per UTC day / ISO week, then period-over-period % change (Metrics).
- Stats — CAGR, max drawdown, Sharpe, Sortino, time-underwater, win rate, profit factor (Metrics).
- Correlation — Pearson on the aligned return series across strategies (Correlation).
- Costs (commission + slippage) are applied inside the backtest, so all figures are net of them (Fees & Slippage).
Verification states
verification_state is our internal validation depth — how many independent ways we confirmed
a strategy's numbers in-house:
| State | Count | Meaning |
|---|---|---|
dual_verified | 12 | Results confirmed two independent ways inside our pipeline — two implementations agree within tolerance. |
engine_backtested | 2 | Validated by the engine's keyless backtest (a single method). |
Both are real backtests over the same public candles and the same cost model. This is a separate
axis from tv_status, which tracks the external TradingView cross-check (all strategies are
currently provisional). The two are complementary, not contradictory — both are published on
every strategy and are explained in full at Verification & data confidence.
How risk_pct linearly rescales returns — and why it's modelled, not re-backtested.
Exact definitions: CAGR, drawdown, Sharpe, Sortino, time-underwater.
Pearson on returns, risk-invariance, effective bets.
Source, engine, denomination, base risk, dataset versions.