Strategy DSL Format

Everything the builder collects ends up in one JSON document. That document is the strategy: the form and node editors are views of it, and the JSON editor (Code mode) edits it directly. Once you can read it you can diff two versions, generate variants in a script, and review a strategy the way you review code.

Nothing reinterprets the document on the way. The builder saves it, backtests send it to dslx_cli, and deployed strategies send the same JSON to the live engine.

• It opens on the whole strategy as engine JSON. Every keystroke is validated against the engine catalog: unknown names, bad parameters and conditions that can never be true are underlined at their line, with the engine's message and a “did you mean” hint.
• Apply DSL (or Ctrl+S) loads the document into the builder, so the form and node editors show it straight away. Apply is disabled while there are errors; warnings do not block it.
• Autocomplete offers indicator names, parameters, outputs, operators, bar fields and timeframes from the catalog. The side panel shows the reference for the indicator under the cursor, or an outline of the document. Ctrl+Shift+F formats.
• Changes made in the form or node editor show up in the JSON the next time Code mode opens, because all three editors read and write the same builder state.

Only entry is required. Everything else has a default, so a hand-written document can be much smaller than what the builder saves.

Every field, default and allowed value is in Strategy DSL — Language Reference.

// a condition: left operand + operator + right-hand side
{
  "indicator": "RSI",              // a catalog name, case-sensitive
  "params": { "period": 14 },      // optional; missing params take the catalog defaults
  "output": "value",               // optional; multi-output indicators default to the first
  "offset": 0,                     // optional; bars ago
  "interval": "1h",                // optional; read on a higher timeframe
  "operator": "crossesAbove",      // one of the eleven operators
  "value": 30                      // number | bar field ("close") | another operand { "indicator": ... }
}

Conditions combine with and, or and not. Each node has exactly one of the three and any child can be another node.

Entries and exits are lists of signals. Each signal has an id, its conditions, and optionally a cooldown, a higher timeframe, Heikin-Ashi candles and a flow. The flow decides what happens once the conditions hold: if / elif / else blocks with their own conditions, or a switch on an operand such as DayOfWeek. In the example the entry is skipped on Thursdays.

entries adds layered entry sets. Each set opens one more position when its own signals fire and price has moved by its gap from the previous layer (layering: adverse for DCA, favorable for a pyramid, any). A set's exitOverride gives its positions their own stop, targets, time exit and max loss in place of the global ones.

Any exit signal closes every open position at the bar close. Stops and targets are checked first on every bar and fill at their level (or at the open if price gapped through). A trailing stop follows the best price. With takeProfit.targets each target closes its exitPercent of the position, nearest first.

The document below uses most of the schema: a signal with a cooldown and a switch flow, a layered add with its own stop and time exit, a trailing ATR stop, two staged targets and a daily loss limit. It validates with no errors or warnings.

{
  "version": 2,
  "name": "Supertrend Scalper",
  "description": "5-minute Supertrend flip with an RSI filter, one add on a pullback",
  "symbol": {
    "exchange": "zerodha", "symbol": "NIFTY", "segment": "futures", "timeframe": "5m",
    "sessionMode": "intraday", "sessionStart": "09:15", "sessionEnd": "15:15",
    "utcOffsetMinutes": 330
  },
  "quantity": { "mode": "fixed", "quantity": 50, "capital": 1000000, "maxPositions": 2 },
  "costs": { "commissionPerOrder": 20, "slippagePercent": 0.01 },
  "entry": {
    "side": "long",
    "signals": [{
      "id": "flip",
      "name": "Supertrend turns up, RSI not stretched",
      "repetitive": false,
      "cooldownBars": 10,
      "conditions": { "and": [
        { "indicator": "Supertrend", "params": { "period": 10, "multiplier": 3 }, "output": "direction",
          "operator": "crossesAbove", "value": 0 },
        { "indicator": "RSI", "params": { "period": 14 }, "operator": "<", "value": 70 }
      ] },
      "flow": [
        { "type": "switch", "operand": { "indicator": "DayOfWeek" },
          "cases": [{ "value": 4, "action": "none" }], "defaultAction": "entry" }
      ]
    }],
    "layering": "adverse",
    "entries": [{
      "id": "add",
      "label": "Add on a pullback to the EMA 20",
      "gap": { "type": "percentage", "value": 0.3 },
      "signals": [{
        "id": "pullback",
        "conditions": { "indicator": "Low", "operator": "<=",
          "value": { "indicator": "EMA", "params": { "period": 20 } } }
      }],
      "exitOverride": {
        "enabled": true,
        "stopLoss": { "enabled": true, "type": "percentage", "value": 0.8 },
        "timeExit": { "enabled": true, "minutes": 45 }
      }
    }]
  },
  "exit": {
    "signals": [{
      "id": "flip-down",
      "conditions": { "indicator": "Supertrend", "params": { "period": 10, "multiplier": 3 }, "output": "direction",
        "operator": "crossesBelow", "value": 0 }
    }],
    "stopLoss": { "enabled": true, "type": "atr", "value": 2, "atrPeriod": 14,
      "trailingEnabled": true, "trailingValue": 1.5 },
    "takeProfit": { "enabled": true, "targets": [
      { "type": "riskReward", "value": 1.5, "exitPercent": 50 },
      { "type": "riskReward", "value": 3,   "exitPercent": 50 }
    ] },
    "maxDailyLossPercent": 2
  }
}

The parser is strict. Nothing falls back silently, and every error carries the JSON path of the mistake.