The nine order gates

An absolute ceiling on how large any one order can be. It is a pure order-shape check: it needs no market state, no position history and no capital read, which is why it runs first.

When it trips: the order is rejected with MAX_ORDER_VALUE_EXCEEDED. The strategy keeps running; the next signal that produces a smaller order is placed normally.

Set it when you want a hard cap on single-order size regardless of what the sizing logic computes, the protection against a sizing bug, a bad price tick, or a strategy that scaled up more than you intended.

The gate that makes "never risk more than 1% on a trade" a real constraint rather than an intention. It computes the actual money at risk between entry and stop and compares it to a budget derived from your allocation.

It only applies to an entry that carries a usable stop price. A market entry with no stop has no measurable risk distance, so the gate cannot evaluate and skips rather than guessing. That is a real limitation worth knowing: a strategy with no stop-loss is not protected by this gate.

When it trips: the order is rejected with PER_TRADE_RISK_EXCEEDED.

Also skips when the strategy's allocated capital is unknown. It cannot resolve a percent into an amount, and it will not invent one.

Caps how many positions the strategy can hold at once. Working orders count, an entry sitting unfilled on the book occupies a slot, so the cap cannot be exceeded by racing two entries.

When it trips: the entry is rejected with MAX_OPEN_POSITIONS_REACHED. Exits are always permitted, so the strategy can close a position to free a slot and continue.

Historically this control was called max_trades in the strategy configuration, and that key is still read for compatibility.

A velocity brake. Counts entries, not round trips, and resets at UTC midnight. Not at the local market open. For an Indian equity strategy this means the counter boundary falls at 05:30 IST, inside the overnight gap rather than at the session open.

When it trips: the entry is rejected with MAX_TRADES_PER_DAY_REACHED until the next UTC day.

Set it when you want a bound on how much a strategy can churn, the protection against a signal that starts firing every candle in a choppy market.

Halts new entries after N losing trades in a row. The streak is derived from the realised sell-P&L timeline and resets on the next winning round-trip. You do not have to clear it manually.

When it trips: the entry is rejected with MAX_CONSECUTIVE_LOSSES_REACHED.

How trading resumes: this gate works with a cooldown. Rather than blocking until a win that cannot happen while entries are blocked, the halt is time-boxed: after cooldown_bars_after_loss bars at the strategy's timeframe have elapsed since the losing trade, the violation is suppressed and entries resume. The default is 3 bars when you have not set a value. See Alerts & breaches.

Set it when you want the strategy to stand down in conditions it is clearly misreading, rather than paying for the same mistake ten times.

The classic circuit breaker: stop trading once today has cost more than a set fraction of the strategy's capital.

Realised only. An open position that is deeply underwater does not count here. This gate reads closed-trade P&L for the day. If you want unrealised loss to halt entries, that is gate 7.

When it trips: the entry is rejected with DAILY_LOSS_LIMIT_EXCEEDED. New entries are blocked; the open position is left to its own stop and target. Blocking exits would trap capital in a losing position, which is the opposite of what this gate exists to do.

Skips when allocated capital is unknown, because a percent cannot be resolved into an amount.

The unrealised-loss brake. Halts new entries once realised plus unrealised loss is down by a set amount.

Stated as an amount, not a percent, deliberately. A percent silently moves the breaker when your capital changes; an amount is the number you actually meant. It mirrors the backtest control of the same name, so a strategy behaves the same researched and deployed.

The mark it uses is the incoming entry's own price, the entry is for the same instrument the strategy already holds, so its limit or trigger price is the current market for the position being valued. A market order with no price hint carries no usable mark, and the gate then skips rather than guessing a valuation.

Two known narrowings versus the backtest control, stated rather than hidden:

• "Realised" here is today's realised P&L. The backtest measures from the run's starting balance. Closing that gap needs a lifetime-realised metric the live path does not carry yet.
• Open exposure is derived from the buy-lot FIFO ledger, so a short-side position contributes quantity but not a mirrored unrealised sign.

When it trips: the entry is rejected with MARK_TO_MARKET_LOSS_EXCEEDED. As with the daily cap, the open position is left to its own exits.

The ceiling that makes a capital allocation mean something. Rejects an opening order that would push total deployed capital past the limit.

This one is not optional in practice. A configured value caps it explicitly; absent one, the strategy's allocated capital is still the effective ceiling, enforced through gate 9. A strategy cannot trade capital it was not given, and cannot reach into another strategy's allocation.

When it trips: the order is rejected with MAX_CAPITAL_ALLOCATION_EXCEEDED.

The last gate, and the one that cannot be turned off. An opening order the strategy cannot fund from its free capital is rejected.

This mirrors the ledger-backed capital check used everywhere else in the platform, so the same rule governs whether an order can be funded regardless of which path asks.

When it trips: the order is rejected with INSUFFICIENT_AVAILABLE_CAPITAL, reporting what was needed and what was available.

With gate 8, this pair is why capital isolation holds. Together they are the guarantee that a strategy stays inside its allocation, the one risk property on this platform that does not depend on your configuration.