Glossary

Every term on this page, explained your way

Definitions follow the setting at the top of the page. Type to filter.

Backtest: Running a rule on past prices to see what it would have done. It's a replay, not a prediction.; Historical simulation of a fully specified rule under a stated fill and cost model. Informative only to the extent the rule was not fit to the data it's run on.
Bootstrap: Re-shuffling your own trades thousands of times to see how much the result could have varied by luck of the draw.; Resampling trades with replacement (5,000 draws) to estimate a 95% interval on net and P(net > 0). Order-shuffle variant used for the drawdown distribution.
Bracket: A stop loss and a target placed together around a trade.; Stop (% of entry) plus target (R × stop distance), with the time exit as backstop and an explicit intrabar fill policy.
Breakeven slippage: How much extra cost per trade the rule could absorb before it stops making money.; Ticks per side at which net expectancy reaches zero, found by re-simulation at each rung.
Coin-flip twins (the null): Pretend versions of your rule that trade the same days and times but choose long or short at random. If your rule can't beat them, it isn't deciding anything.; Random-side null: matched eligibility, timestamps and costs; direction drawn per trade; 5,000 seeds; p is the empirical share matching or beating the real net.
Concentration: How much of the total profit came from a single trade or a single year. High concentration means the result leans on a few lucky days.; Top-trade, top-5 and best-year shares of net; scored in Exam IV (top trade) and Exam III (largest unseen year).
Continuous contract, unadjusted: Futures contracts expire every few months. A continuous series stitches them together so you can test many years. "Unadjusted" means the prices are the real prices people traded, not shifted to hide the gaps.; Front-month splice without back-adjustment. Used because percentage gates on additively adjusted series are distorted across time; roll-week days can diverge between vendors and are disclosed.
Drawdown: The biggest drop from a high point in your running profit to the low that followed. It's what a losing stretch feels like in dollars.; Peak-to-trough decline of cumulative net. Reported actual and 95th percentile under order shuffle; the ratio scores drawdown realism.
Edge: A real, repeatable reason a rule makes money, as opposed to having been lucky.; Positive expectancy attributable to the rule's decision, distinguishable from a matched null on data the rule was not developed on.
ETH: "Electronic trading hours": the nearly-24-hour session futures trade in, not just 9:30 to 4:00.; Full-session bars. Time gates are evaluated on bar close stamps in US Eastern.
Fade / Follow: Fade: bet the move reverses. Follow: bet it keeps going.; Mode mapping from sign of move to side of trade; it is the single decision the null tests.
Filter: An extra condition bolted onto a rule: only trade with the trend, only on busy days, never on Fridays. Each one makes the backtest look better and the rule more fragile. You can add up to three here; every one is named in the report, counted against your score, counted as one more version of the rule, and shown with the trades it removed.; A predicate on the signal list, evaluated at the signal bar or on completed prior days. Under Protocol Amendment #2 any number may be declared; each adds its tunables to the parsimony count, one lineage step, a ±10% perturbation in Exam VI, and a row in the condition-by-condition table.
Squeeze: A day whose Bollinger bands sat inside its Keltner channel: the market was unusually still. The condition only takes signals on the day after one. A reader handed us this one; on the plain opening-range rule it beat the coin flips, and the report shows what it cost in trades.; Prior daily bar with Bollinger(20, 2.0) inside Keltner(20, 1.5, Wilder ATR). Four tunables, all perturbed.
Breakeven (move to): Once a trade has gone your way by a set amount, the stop moves up to your entry price so the trade can no longer lose. It feels free. Sometimes it saves a runner; sometimes it scratches a trade that was about to pay. The report shows the rule with and without it.; Trigger in R, a fraction of the first target, points or ticks; the stop moves to entry ± an offset. Trigger and entry touched in one bar resolve against you and are counted as ambiguous.
Trailing stop (chandelier): A stop that follows the best price the trade has reached, a fixed distance behind it, and never loosens. Recomputed at the close of each bar, not tick by tick. When the distance is an ATR multiple this is the chandelier exit.; Distance in points, ticks or an ATR multiple; activates at the fill or after a trigger; only ever tightens the working stop.
Range projection: A target set by laying the opening range's height on top of its edge, the "measured move." Different from a fraction of the range measured from your entry, which sits a little farther out because your entry is past the edge.; Target = range edge + k × H, resolved from the real side and mirrored to the twin; the range-fraction target is entry + k × H.
Last swing inside the range: A stop at the most recent dip (for a long) or bump (for a short) that formed inside the opening range before the break. If nothing formed, the far side of the range is used, and the report says how often that happened.; Most recent 3-bar pivot low/high between the range's last bar and the break bar, inside [range low, range high]; fallback to the far side, counted.
Condition by condition: A table in the report that shows the trade count, net and unseen-era net after each condition is added, in the order you added them. It's how you see which condition did the work and which one just removed days.; Cumulative prefix runs of the filter stack; the base-versus-filtered comparison is printed for the full stack.
Gate: A problem so serious the score doesn't matter. Too few trades is one. The rule only working on years you looked at is another.; Auto-fail conditions G1–G7 in Protocol v1.1. A tripped gate yields Rejected at any point total, with the reason class recorded.
Lineage (multiplicity): How many versions of an idea you tried before this one. The more you tried, the more the winner has to be discounted.; Effective variant count n (declared vs. submissions per address, whichever is larger). Adjusted p = 1 − (1 − p)^n.
Livability: Whether a real person with a real account could have stuck with the rule through its slow years. A rule can be real and not livable.; Report section 9, lines not points: longest stretch without a new equity high, share of negative rolling 12-month windows and the worst one, max drawdown against the median year, and share of profit from the best two years. Kept out of the score so a real-but-slow edge and a fake-but-smooth one can never look alike.
One-minute bars: Price summarized once per minute: where it opened, the high, the low, where it closed. Fine enough to time entries to the minute.; OHLCV at 1-minute resolution, bar-start stamped by the vendor and evaluated on close stamps by the engine.
Overfitting (curve-fitting): Tuning a rule until it fits the past perfectly, which usually means it fits nothing else. The most common way a good-looking backtest turns into a losing account.; Selection of parameters on the evaluation data. Detected by unseen-era collapse, plateau failure, and null indistinguishability; partially corrected by lineage adjustment.
p-value: The share of coin-flip twins that did as well as you. Smaller is better. 0.05 is the usual bar.; Empirical tail probability under the random-side null. Reported raw and lineage-adjusted; unseen-era value governs.
Plateau vs. needle: If nudging a setting a little barely changes the result, you're on a plateau, which is good. If a small nudge ruins it, you found a needle, which means the setting was tuned to the past.; ±10% perturbation of the primary trigger; retention of base net and sign-flip check. Exam VI.
Profit factor: Total money made on winning trades divided by total money lost on losers. Above 1 means profitable; 1.3 is solid for a simple rule.; Gross wins ÷ gross losses. Used for cost-survival at 2× friction and for unseen/seen retention in Exam III.
R multiple: How much you made or lost measured in units of what you risked. Risk 20 points, make 40, that's +2R.; P&L normalized by initial stop distance. Targets are specified in R.
Seen era / Unseen era: Seen: the years you've already looked at while building the rule. Unseen: years before that. Only the unseen years can prove anything, because you couldn't have tuned to them.; Declared contamination boundary. Unseen era governs the null p, the bootstrap CI, and era consistency (Design B holdout).
Sharpe (per-trade): Average profit per trade divided by how much trades vary. Higher means steadier.; Mean ÷ standard deviation of per-trade P&L. Retention unseen/seen scores Exam II on a stated basis.
Slippage: The small amount you lose on each trade because the price you get isn't quite the price you saw. We charge one tick per side and then test worse.; Adverse fill offset in ticks per side on market orders; stress-tested at 2 and 3 and searched to breakeven.
Stop loss: An order that closes your trade automatically if it loses a set amount. It limits a single loss, but it can also cut off the moves that made the rule work.; Market-on-touch at a percentage of entry; slippage applied; gaps fill at the open.
Tick and point value: A tick is the smallest price step a contract moves. A point is the full unit. Each is worth a fixed number of dollars, different for every market.; Per-instrument constants from the verified instrument table; used for slippage, cost stress and dollar P&L.
Validated / Marginal / Rejected: Validated: the rule is real and holds up. Marginal: something real may be there, but something specific is broken and the report names it. Rejected: don't.; Score bands 70 / 40 with the No-Crater Rule and gates. A verdict is a statement about statistical reality on the tested data, not about future returns or livability.