# Crypto backtest performance-metric audit worksheet

> Blank research template. It records inputs and conventions; it does not calculate a metric, verify a market edge or recommend a trade.

## 1. Identify the run

- Strategy name, version and fingerprint:
- Dataset source, instrument, venue and date retrieved:
- Timezone, sampling interval and evaluation window:
- Account/equity base, starting balance and external cash-flow treatment:
- Software, engine version, price-mark rule and fill assumptions:
- Fees, spread, slippage, funding and other cost treatment:
- Benchmark and question chosen before seeing this result:

## 2. Choose the correct observation ledger

Use one row per fixed observation interval for periodic account-return measures. State whether each return is time-weighted, how open positions are marked, how deposits and withdrawals are handled, and whether flat periods remain in the series. Do not replace missing observations with zero.

| UTC interval end | Equity / NAV | External cash flow | Net period return | Risk-free return | Sortino target | Mark / missing-data note |
|---|---:|---:|---:|---:|---:|---|
| | | | | | | |
| | | | | | | |
| | | | | | | |

Keep a separate row per completed trade for profit factor. Include every winner and loser under one declared gross or net convention; do not mix period returns with trade P&L.

| Trade ID | Entry UTC | Exit UTC | Gross winning P&L | Gross losing P&L | Fees | Other costs | Net result | Partial / open / excluded note |
|---|---|---|---:|---:|---:|---:|---:|---|
| | | | | | | | | |
| | | | | | | | | |

## 3. Write each metric convention before comparison

| Measure | State the calculation inputs and convention | Result / unit |
|---|---|---|
| Sharpe | Periodic excess-return series; mean convention; standard-deviation convention; risk-free input | |
| Sortino | Periodic return series; target; downside-deviation denominator and treatment of zero shortfalls | |
| Profit factor | Gross or net trade P&L; treatment of open trades, costs and a zero-loss denominator | |
| Maximum drawdown | Equity-mark frequency; cash-flow adjustment; peak and trough timestamps; dollar and percentage depth | |
| Annualized statistic | Period conversion and dependence assumptions; any autocorrelation treatment | |

For the Sortino example in the related guide, the stated convention is the mean of return minus target divided by the square root of the mean squared shortfalls, including zero shortfalls across all observations. Other implementations can use different targets or denominators. Record the actual formula used by the software.

If periods are monthly, do not assume multiplying a monthly Sharpe by the square root of 12 is valid when returns are dependent. State the assumptions and inspect serial dependence. A ratio based on a short or selected sample is especially uncertain.

## 4. Preserve the search and validation record

| Candidate / version | Parameter or asset choices tried | Why retained or rejected | Date / data window inspected | Holdout touched? |
|---|---|---|---|---|
| | | | | |
| | | | | |

Record the full number and scope of trials, discarded variants and any later-period data opened during selection. A selection-adjusted statistic still cannot repair look-ahead bias, unreliable data, unrealistic fills or an unsuitable return series.

## 5. Reconcile and state limits

- [ ] Arithmetic independently reproduced from the retained input rows.
- [ ] Sampling interval, units, target, costs, open-position marks and cash flows are stated.
- [ ] Trade-based metrics use a trade ledger; periodic metrics use a periodic return ledger.
- [ ] Drawdown uses the declared equity-mark frequency; gaps and missing marks remain visible.
- [ ] Trial count, revisions, holdout use and cost sensitivity are recorded.
- [ ] Synthetic, historical, paper and actual fill evidence are labeled separately.

Result or unresolved discrepancy:

What would change the interpretation?

Next review date:

Read the [Crypto Backtest Metrics guide](/learn/crypto-backtest-metrics) for worked fictional calculations. Keep this worksheet with the source data and software version. A metric summarizes a chosen sample; it does not predict a future result.