Skip to main content

Data Provenance

Where every figure comes from, how it is denominated, and how dataset versions work.

Source & engine​

  • Data source: public Deribit chart history (candles). No private data.
  • Engine: the project's own Deribit-native BTC-PERPETUAL DRY_RUN backtest — the engine's keyless validation harness, run bar-by-bar over those candles. It authenticates to nothing, reads no credentials, and is not the live trading engine. (Isolation is enforced by a guard test in the exporter's repo.)
  • The exporter re-runs this backtest per strategy and persists the equity curve, daily/weekly returns, stats, and the cross-strategy correlation matrices as static JSON, served read-only by this API.

Denomination — BTC inverse​

:::warning Returns are BTC-denominated BTC-PERPETUAL is an inverse, BTC-settled contract: margin, PnL and balances accrue in BTC. Every return, curve and statistic is therefore in BTC terms (ccy: "BTC"), not USD. A BTC-denominated return is not the same as a USD return — converting requires the BTC/USD path over the same window, which this dataset does not apply. Interpret all performance as "in BTC." :::

Sizing — base risk, single strategy​

  • Each strategy is simulated at its own base_risk_pct (per-trade risk fraction). Across the set these are 1.0%–2.5% (0.01, 0.015, 0.02, 0.025).
  • The backtest forces book_size = 1 — i.e. single-strategy sizing at base risk.

:::note Not the live portfolio These figures are per-strategy at base risk, independent of any live multi-strategy allocation. They are not the live engine's combined book (which sizes differently across several sub-accounts). Use risk_pct to view a strategy at a different risk dial, and /v1/summary for an equal-weight book view — but neither is a claim about a specific live allocation. :::

Window​

  • Warm-up from 2018-08-14 (indicators warm here); published series run to the dataset's generation date (currently 2026-06-24) — ~2,871 daily points (~7.8 years).
  • Each strategy's exact window (from/to) is on its metadata.

Dataset versioning​

  • The dataset is a static, immutable snapshot. manifest.generated_utc is the dataset_version, surfaced in every API response meta, on /v1/dataset, and used for ETag/304 cache revalidation.
  • A new snapshot is produced by re-running the exporter; the dataset_version changes accordingly. Figures are deterministic given the same public candles and engine version.

Not yet stamped​

:::warning parameters & engine_commit are null-until-stamped Two provenance fields are currently null in the dataset and on /v1/strategies/{id}:

  • parameters — the exact per-strategy indicator/rule parameters used in the backtest.
  • engine_commit — the git commit of the engine that produced the snapshot.

They are placeholders pending a planned exporter enhancement to stamp them, so the reference is honest about machine-exact reproducibility today. The documented fees & slippage and the engine/source above are fixed and accurate; the precise parameter values are not yet published per-strategy. :::

Overview

The dataset and verification states in context.

API Reference

Pull metadata, stats, returns, curve and correlation directly.