Strategy System

"configuration": {
  "timeframe": "1H",
  "warmupBars": 21,
  "maxOpenPositions": 1,
  "direction": "long",
  "entryFeePct": 0.1,
  "exitFeePct": 0.1,
  "timezone": "America/Sao_Paulo"
}

"inputs": {
  "emaFast":   "ema(close, 9)",
  "emaSlow":   "ema(close, 21)",
  "rsiValue":  "rsi(close, 14)",
  "atrValue":  "atr(14)",
  "volAvg":    "sma(volume, 20)",
  "volSpike":  "volume > volAvg * 1.5"
}

"conditions": {
  "trendUp":    "emaFast > emaSlow",
  "rsiHealthy": "rsiValue > 40 AND rsiValue < 65",
  "breakout":   "crossUp(close, emaSlow)",
  "volConfirm": "volSpike"
}

"score": {
  "trendUp":    20,
  "rsiHealthy": 30,
  "breakout":   40,
  "volConfirm": 10
}

"decision": {
  "entry": "score >= 50"
}

"decision": {
  "entry": "score >= 50 AND breakout"
}

"riskManagement": {
  "stopLoss":   { "type": "percent", "percent": 1.0 },
  "takeProfit": { "type": "riskReward", "multiple": 3.0 }
}

"riskManagement": {
  "stopLoss":   { "type": "atr", "period": 14, "multiplier": 1.5 },
  "takeProfit": { "type": "percent", "percent": 2.0 }
}

{
  "configuration": {
    "timeframe": "1H",
    "warmupBars": 21
  },
  "inputs": {
    "emaFast":  "ema(close, 9)",
    "emaSlow":  "ema(close, 21)",
    "rsiValue": "rsi(close, 14)",
    "volAvg":   "sma(volume, 20)",
    "volRatio": "volume > volAvg * 1.3"
  },
  "conditions": {
    "trendUp":    "emaFast > emaSlow",
    "rsiHealthy": "rsiValue > 40 AND rsiValue < 65",
    "volConfirm": "volRatio",
    "crossover":  "crossUp(emaFast, emaSlow)"
  },
  "score": {
    "trendUp":    20,
    "rsiHealthy": 25,
    "volConfirm": 15,
    "crossover":  40
  },
  "decision": {
    "entry": "score >= 60"
  }
}

"conditions": {
  "breakout": "close > resistance"  // ERROR: 'resistance' not defined in inputs
}

# EmidLabs Strategy DSL — Reference for AI Strategy Generation

A strategy is a single JSON object with six top-level keys: configuration, inputs, conditions, score, decision, riskManagement.
Output must be valid JSON only — no markdown, no comments, no explanations.

## configuration
{
  "timeframe": "5M" | "15M" | "30M" | "1H" | "2H" | "4H" | "1D",
  "warmupBars": <int, optional, default 0>,
  "maxOpenPositions": <int > 0, optional, default null = unlimited>,
  "direction": "long" | "short", optional, default "long",
  "entryFeePct": <number 0-5, optional, default 0>,
  "exitFeePct": <number 0-5, optional, default 0>,
  "timezone": <IANA id string, optional, default "America/Sao_Paulo">
}
warmupBars should be set to the largest lookback period among the indicators used in "inputs"
(e.g. if the strategy uses ema(close, 21) and rsi(close, 14), set warmupBars to 21) — the backend
now raises an unset/too-low value up to that same floor automatically, but still set it explicitly
for best results, since the auto-floor is a conservative minimum, not a guarantee of a fully
converged indicator value. Candles inside the warmup window are excluded from trading — entry
is forced false and score to 0 — because their indicator values have not fully stabilized yet.
maxOpenPositions caps how many trades can be open at the same time. Omit it (or leave it null)
for unlimited concurrent positions (the default). Set it to 1 for single-position mode — a new
entry is skipped whenever a position is already open.
direction picks which side the whole backtest trades — every position opened by "decision.entry"
is a long (buy) or a short (sell) accordingly. A single strategy is one direction, not both at
once. "entry"/"exit"/"riskManagement" keep the exact same meaning either way — only the
stop-loss/take-profit price placement and win/loss sign are mirrored for "short" (stop above
entry, take below entry, profit when price falls).
entryFeePct/exitFeePct simulate a per-leg exchange fee as a percent of price (e.g. 0.1 = 0.1%,
a typical taker rate at Binance VIP0). Both default to 0 (no fee). When set, every trade's
pnlR/pnlPct in the result are net of this fee, so expectancyR/winRate/profitFactor account for it
automatically — a trade that looks marginally profitable before fees can correctly flip to a loss.
Real exchanges can charge different maker/taker rates per leg, so the two fields are independent,
not assumed equal. Caveat: a trade's fee cost in R-units scales inversely with that trade's own
stop distance, so once a fee is set, expectancyR is only a fair comparison within one strategy's
own stop convention — prefer pnlPct-based comparisons across strategies/timeframes with different
typical stop widths.
timezone is an IANA id (e.g. "America/Sao_Paulo") that localizes the hour()/minute()/dayOfWeek()/
isWeekend() functions above — use it for time-of-day or weekday gating conditions.

## inputs
Named, reusable computed values. Evaluated once per candle, before conditions.
Available market data: close, open, high, low, volume
Built-in functions:
  Indicators: ema(series, period), sma(series, period), rsi(series, period) [0-100], atr(period),
    adx(period) [0-100, trend strength], adxPlusDi(period), adxMinusDi(period) [+DI/-DI, trend direction —
    compare adxPlusDi vs adxMinusDi alongside adx]
  Signals: crossUp(a, b), crossDown(a, b)
  Series operators: highest(series, n), lowest(series, n), change(series)
  Lag / window aggregation: shift(series, n) [look-back only, NaN before enough history],
    any(boolSeries, n), all(boolSeries, n), count(boolSeries, n) [over the n candles before the current one]
  Structure: swingHigh(series, confirmBars), swingLow(series, confirmBars) [confirmed N-bar swing point —
    true confirmBars candles AFTER the actual peak/trough, never at the peak itself, so it can't repaint
    between backtest and live]
  Candle anatomy: body(), range(), upperWick(), lowerWick(), isBullish(), isBearish()
  Math: abs(x), min(a, b), max(a, b)
  Time: hour() [0-23], minute() [0-59], dayOfWeek() [0=Sunday..6=Saturday], isWeekend() [boolean] —
    all localized to "configuration.timezone" (default "America/Sao_Paulo"), for time-of-day or
    weekday gating, e.g. "hour() >= 9 and hour() < 13" or "not isWeekend()".
  Volume: there is no volumeSma()/volumeSpike() — volume is a market-data series like close/open/high/low,
    so use sma(volume, period) for a volume moving average, and volume > sma(volume, period) * multiplier
    for a volume-spike condition.
  Candlestick patterns (all no-arg, boolean, read only closed-candle OHLC — same in backtest and live):
    Single-candle: hammer(), shootingStar(), doji(), bullishMarubozu(), bearishMarubozu(), spinningTop(),
      dragonflyDoji(), gravestoneDoji(), longLeggedDoji()
    Two-candle: bullishEngulfing(), bearishEngulfing(), piercingLine(), darkCloudCover(), bullishHarami(),
      bearishHarami(), haramiCross(), tweezerTop(), tweezerBottom()
    Three-plus-candle: morningStar(), eveningStar(), threeWhiteSoldiers(), threeBlackCrows(),
      threeInsideUp(), threeInsideDown(), threeOutsideUp(), threeOutsideDown(), risingThreeMethods(),
      fallingThreeMethods()
    Caveat: hammer()/shootingStar() are shape-only and don't know the prior trend (the same shape is a
    Hammer after a downtrend but a Hanging Man after an uptrend, and vice versa for shootingStar/Inverted
    Hammer) — pair with a trend/momentum condition (e.g. rsi, ema) rather than using the shape alone.
Rules:
  - Names must be unique.
  - An input can only reference inputs defined above it (order matters).
  - Circular references are not allowed.
  - Must resolve to a numeric value.
Example: { "emaFast": "ema(close, 9)", "rsiValue": "rsi(close, 14)" }

## conditions
Named boolean expressions using inputs, market data, and operators.
Operators: >  <  >=  <=  ==  !=  AND  OR
Rules:
  - Every variable referenced must be defined in inputs.
  - Must resolve to a boolean (0 = false, non-zero = true).
  - No future data — only current or past candle data (no lookahead bias).
  - No randomness — strategies must be fully deterministic.
Example: { "trendUp": "emaFast > emaSlow", "rsiHealthy": "rsiValue > 40 AND rsiValue < 65" }

## score
Integer weight per condition. On each candle, the engine sums the scores of all conditions
that evaluate to true. Values are arbitrary — only their relative weight matters.
Example: { "trendUp": 20, "rsiHealthy": 30 }

## decision
{ "entry": "<expression>", "exit": "<expression, optional>" }
"entry" is evaluated per candle after the total score is computed. May reference "score" (the
candle's total) and any condition by name. When true, a trade opens at that candle's close.
"exit" is optional and uses the same expression rules as "entry". When true on a candle where a
position is open, that position closes at that candle's close — independent of "riskManagement".
If omitted, positions only close via stop-loss/take-profit.
Example: { "entry": "score >= 50 AND breakout", "exit": "crossDown(emaFast, emaSlow)" }

## riskManagement
Optional. Configures how each open trade's stop-loss and take-profit are calculated. If omitted
entirely, the engine defaults to { stopLoss: percent 1.0, takeProfit: riskReward 3.0 } — the
same as older strategies that predate this section.

stopLoss (pick one type):
  { "type": "percent", "percent": <number > 0, e.g. 1.0 for 1%> }
  { "type": "atr", "period": <int > 0>, "multiplier": <number > 0> }

takeProfit (pick one type):
  { "type": "riskReward", "multiple": <number > 0> }  // take-profit distance = stopLoss distance * multiple
  { "type": "percent", "percent": <number > 0> }

Example: { "stopLoss": { "type": "atr", "period": 14, "multiplier": 1.5 },
           "takeProfit": { "type": "percent", "percent": 2.0 } }

## Execution model
  - Entry fills at the close of the candle where "decision.entry" turns true — no lookahead.
  - Risk per trade is governed by "riskManagement" (stop-loss and take-profit levels calculated
    at entry). If "riskManagement" is omitted, the default is a fixed 1% stop-loss and a 1:3
    risk-reward take-profit (win = +3.0R, loss = -1.0R). With a custom "riskManagement", win/loss
    R magnitudes can vary per trade (e.g. an ATR-based stop-loss).
  - A position closes on the first of three events: stop-loss hit, take-profit hit, or
    "decision.exit" turning true, checked in that order each candle. If a candle's stop-loss (or
    take-profit) and "decision.exit" would both trigger on the same candle, the stop-loss/
    take-profit wins and "decision.exit" is not evaluated for that position that candle.
  - "configuration.direction" picks long or short for the whole backtest (default long) — see
    "configuration" above. There is no per-trade direction; one strategy trades one side.
  - By default, multiple positions can be open at once — every candle where entry triggers opens
    a new, independent trade, regardless of other open trades. Set "configuration.maxOpenPositions"
    to cap concurrent trades (e.g. 1 for single-position mode).

## Invalid patterns — do not generate
  - Referencing an input or condition that isn't defined.
  - A condition that doesn't resolve to a boolean.
  - Referencing future candle data.
  - Circular input references.
  - Randomness or any non-deterministic logic.

## Full example
{
  "configuration": { "timeframe": "1H", "warmupBars": 21 },
  "inputs": {
    "emaFast": "ema(close, 9)",
    "emaSlow": "ema(close, 21)",
    "rsiValue": "rsi(close, 14)",
    "volAvg": "sma(volume, 20)",
    "volSpike": "volume > volAvg * 1.3"
  },
  "conditions": {
    "trendUp": "emaFast > emaSlow",
    "rsiHealthy": "rsiValue > 40 AND rsiValue < 65",
    "volConfirm": "volSpike",
    "crossover": "crossUp(emaFast, emaSlow)"
  },
  "score": { "trendUp": 20, "rsiHealthy": 25, "volConfirm": 15, "crossover": 40 },
  "decision": { "entry": "score >= 60" },
  "riskManagement": {
    "stopLoss": { "type": "percent", "percent": 1.0 },
    "takeProfit": { "type": "riskReward", "multiple": 3.0 }
  }
}

## Submitting to the API
POST https://backtest.emidlabs.com/api/public/v1/backtest
Nest the strategy object under "strategySnapshotJson", alongside "assetPair", "initialDate",
and "finalDate".