Capped Martingale

The practical answer to "I want martingale recovery but I want to survive". Size doubles like exponential martingale until it reaches a ceiling, then holds flat forever. You keep most of the fast-recovery benefit while converting an unbounded convex tail into a linear one — which is the difference between a strategy you can leave running and one you cannot.

Size at level i is min(2ⁱ, cap), with the cap defaulting to 8× base. The first few rungs double; every rung past the cap is the same size, so the ladder becomes a flat ladder at cap × base. Capital is therefore convex early and linear late.

Below is the wizard's own ladder table and the trading terminal's chart engine, both driven by buildLadderPlan for this mode. Change the level count, spread or target and the rung prices, sizes, capital ramp and chart lines recompute exactly as they would in the grid designer.

Bounded growth: the exponential part contributes a fixed amount and then depth costs cap × base per level. That makes deep ladders affordable and, more importantly, makes the worst case a number you can state before deploying.

Read the cost vs 5 levels column as the escalation test. A flat ladder doubles when the depth doubles; anything materially above that is a curve whose coverage has to be bought rather than simply configured.

With every rung filled, the volume-weighted average entry is 93.91 against a base of 100 and a deepest rung of 90.00 — so the average sits 61% of the way down the covered range. Getting back to flat from the bottom therefore needs a 4.35% bounce, against 5.00% for a flat ladder of the same depth — 0.65% less work.

Doubles like exponential martingale until the multiplier reaches the cap (default 8×), then holds flat. You keep most of the fast-recovery benefit while guaranteeing a maximum per-level size, so a long adverse run cannot compound into a blow-up.

The last row is the metric that separates modes. Profit for a full cycle is the target percentage of deployed capital in every mode, so it cannot discriminate. What differs is how much capital has to sit reserved to earn one shallow cycle: this mode reserves 70.0× base to collect 2.9 bps of it when the first rung completes.

Live mean-reversion on instruments that occasionally trend — which is to say, most instruments. The mode is designed for the case where the regime assumption will sometimes be wrong.

Doubles like exponential martingale but clamps the multiplier at a ceiling (default 8×). You keep most of the fast-recovery benefit while guaranteeing a maximum per-level size — one of the most practical, survivable martingale variants for live trading.

The tail is heavy but linear past the cap. The honest risk is that the cap is a dial: raising it toward 32 or 64 progressively re-introduces the exponential behaviour the mode exists to prevent.

This is the product's grid backtester, not a summary of one. It fetches public candles for the instrument you pick, rebuilds the ladder anchored on the first candle of the window, then applies the engine's own fill, take-profit, refill and stop-loss rules bar by bar. The caveat block inside the panel is part of the output — read it before trusting a number.

First the arithmetic, then the market. The table below is computed from the size curves alone; the sweep underneath replays every one of those ladders over the same real candle window, with grid_mode as the only variable.

• Live mean-reversion where surviving a trend matters more than maximal recovery.
• You want martingale behaviour you can actually leave running unattended.
• Any situation where you were tempted by exponential martingale.

• You need the true exponential tail — but you almost certainly do not.
• The cap has been raised so high that the mode is exponential in all but name.

• The cap is the whole strategy. 4–8 is conservative, 16 is aggressive, 32+ is exponential with extra steps.
• Depth past the cap is cheap in relative terms: extending coverage costs cap × base per level, predictably.
• Compare against capped linear in the sweep below — often the gentler ceiling performs comparably at a fraction of the commitment.

This mode is parametric. The wizard and the settings panel show only the dial(s) it actually uses, and changing one is a structural edit — the resting ladder is cancelled and re-placed at the new sizes.

• Record the cap in every deployment note; the mode name without the cap is not a description.
• Alert when the deepest fill passes the cap level — from there the ladder behaves flat and coverage becomes the binding constraint.
• Watch the deepest filled level, not the P&L. Depth is the leading indicator; P&L is the lagging one.
• Compare peak committed notional against the account, weekly. A growth curve that has never been deep has never been tested.
• Treat every level increase as a capital decision: adding levels to a growth curve is not a linear cost.

1. 1
   Cap inflation

   An operator raises the cap after a period of good results, reasoning that the doubling never hurt. The tail returns in full and the mode's central protection is gone.
2. 2
   Post-cap coverage exhaustion

   Past the cap the ladder is flat, so very deep trends still exhaust the range. The stop and the level count remain necessary.

• Pairs well with a stop loss placed just beyond the covered range — the bounded curve makes the loss at the stop calculable.
• Spread sets how fast the ladder consumes depth. Widening it buys coverage for the same level count at the cost of fewer fills.
• Target sets what each cycle books. Below round-trip fees plus slippage, a grid is a fee generator regardless of the size curve.
• The stop loss is the only bound on the tail. Without it, the curve's worst case is whatever the market does, not what you planned.

1. 1
   What cap should I start with?

   The default 8×. Then decide from measurement, not intuition: run the sweep at 4, 8 and 16 and compare return on capital and max drawdown together.
2. 2
   Is this the best default martingale?

   For unattended production, it is the strongest candidate in the catalog — bounded worst case, most of the recovery profile, one legible dial.

• Sizes and capital: levelMultiplier / gridModeCurve in lib/grid/gridModes.ts, mirrored from grid_modes.hpp and the Python engine's lot calculator.
• Rung and exit prices, capital ramp, init targets and the stop line: buildLadderPlan — the same builder the wizard and grid designer use.
• Backtest fills, cycles and P&L: runBacktest / runBacktestSweep over public broker candles, applying the engine's fill, exit, refill and stop rules.
• Price charts: the trading terminal's chart engine, with the ladder drawn through its own price-line API.

• 1.Grid Modes — Overview — The full taxonomy and how to choose
• 2.Capped family — What every curve in this family has in common
• 3.Capital Allocation — Sizing an account to a Capped Martingale ladder
• 4.Stop Loss — Bounding this curve's tail
• 5.Reading the Ladder Plan — Verifying the curve before deploying
• 6.Grid Structure — Levels, init batch and base size