Kurzharr — Glossary

The terms the scorecard and these pages use, defined plainly. Where a term names a validation check, the full description is in the validation stack.

ATR (average true range): A rolling measure of how much price typically moves in one bar, including gaps. Used to place stops and targets in units of current volatility rather than at fixed distances.
Attribution: The layer beneath the scorecard that translates a pattern of passes and failures into a plain-language account of what it usually means. The pattern is matched by deterministic code; a language model only phrases the matched template and fills in your numbers.
Backtest: Running a fully specified strategy over historical data to see what it would have done. A backtest is a measurement, not evidence on its own — the validation stack exists because measurements are easy to inflate.
Buy-and-hold benchmark: What buying the traded symbol at the start of the test window and holding it to the end would have returned, drawn as a dashed line next to the strategy’s equity curve. The “would you have been better off doing nothing” comparison. A strategy can clear every check and still have underperformed it.
CAGR (compound annual growth rate): A total return expressed as the equivalent steady per-year rate, so runs of different lengths can be compared at a glance. Shown for the strategy and the buy-and-hold benchmark, both over the full test window.
Cluster-aware inference: Significance recomputed treating trades that share a day or a market event as correlated rather than independent. A naive per-trade t-stat overstates confidence when a few cascading days do most of the work.
Cost realism: The result re-run net of realistic costs — spread on entry and exit, commission, and overnight swap or financing accrued per bar held — shown next to the gross figure.
Deflated Sharpe ratio: A Sharpe ratio adjusted downward for how many strategy variants have been tried, how non-normal the returns are, and how short the sample is. It guards against the best of many attempts looking significant by chance. The platform tracks it across your entire run history, not just the current run.
Drawdown: The decline from a peak in the equity curve to the next trough, expressed in percent or in R. The usual measure of how much pain a strategy puts you through to earn its return.
Economic significance: Whether an edge is large enough to be worth trading — distinct from statistical significance, which is only whether it is real. Every fixed check tests the second; none tests the first, because a small edge at a large trade count passes a significance test just as a big one does. An optional pre-registered floor on net-of-cost profit factor is the one check that tests it, and only when you commit one.
Equity curve: Cumulative account value over the course of a run, trade by trade.
Floor (pre-registered): A threshold a run must clear — minimum trade count, minimum significance — set either by the platform from the strategy’s geometry and sample size, or committed by you, before the run starts. Immutable once the run begins.
Forward test: Tracking a strategy on data that did not exist when it was built. The one test a backtest cannot fake, and the natural next step after a strategy clears.
Gambler’s-ruin baseline: For a fixed stop/target ratio, the win rate a zero-skill entry would produce — 1 / (1 + ratio) for a driftless random walk. The null-information check compares your actual hit rate against this.
Geometry: The stop, target and position-sizing rules of a strategy, considered separately from the entry signal. Two strategies can share an entry and behave completely differently because of their geometry.
Hit rate: The fraction of trades that reached their target before their stop.
Leak (lookahead bias): The backtest using information that would not have existed at decision time — an indicator reading a future bar, a higher-timeframe value that had not closed yet, an off-by-one in a previous-period join. The most common single reason a backtest beats live trading.
Leak audit (truncated rebuild): Recomputing each decision using only data up to that decision’s own timestamp, then checking it against the full-series run. Any divergence is reported down to the specific decision and field.
Null-information diagnostic: A comparison of your strategy’s actual hit rate against the pure-chance baseline implied by its stop/target ratio. If entries do not beat that, nothing downstream can create an edge that is not there. Also available as a standalone tool, the Null-info diagnostic.
Placebo test (permutation): 200 random reshuffles of your signal timing over the same data. Your real result has to sit meaningfully outside the distribution those produce, or it is within the range of what selection artifacts on this dataset can generate.
Pre-registration: Fixing the floors a run is judged against before the run produces any result. What stops “re-run until it passes”: every run is recorded, and floors cannot be revised after the fact.
Profit factor (PF): Gross profit divided by gross loss. Above 1 means profitable before costs; the platform also reports it after costs.
R-multiple: Profit or loss measured in units of the risk taken, where 1R is the distance from entry to stop. A trade closed at +2R made twice what it risked; a strategy’s expectancy is often quoted in R per trade.
Sharpe ratio: Return divided by the volatility of that return, annualised. A risk-adjusted performance number — and, on its own, one of the easiest to inflate, which is why the platform deflates it and splits the sample.
Spec: The structured, declarative description of a strategy: its universe, its indicators, its entry and exit conditions, its geometry and its floors. Strategies are composed as specs; there is no strategy-level code.
Temporal-halves gate: Splitting the sample at the median trade and requiring the edge to hold independently in both halves. Distinguishes a live edge from a regime that existed and ended.
Trial: One recorded run against your account. The first run of a strategy and every later refinement each count as a trial, and each one raises the deflated-Sharpe bar.
Twin control: A deliberately broken copy of your strategy, with a lookahead bug planted in, run through the leak audit before your real strategy. If the audit does not catch the twin, your run is halted — the checker cannot be trusted on the real strategy either.
Universe: The symbol or symbols, timeframe and date range a strategy trades over. Set once when the strategy is built.
Veto: A check whose failure invalidates the entire run rather than counting as one gap among several — the leak audit and the twin control. When a veto fails, the rest of the scorecard is not interpreted.
Verdict: The one-line result at the top of a scorecard: Cleared, Cleared with one gap, Has not cleared, or Not enough trades to grade. See the scorecard & verdicts.
z-score: How many standard errors an observed value sits from its baseline. Positive means above the baseline; a magnitude past about 2 is the conventional line for “unlikely to be chance”.