Backtest System — Dependency Map

This document maps every dependency in the backtest pipeline: what data is required, how the simulation works per entry strategy, how SL/TP levels are derived, and what the output means. Read alongside Trading System — Dependency Map for live trading context.

1. System Overview

Admin UI ──► POST /api/ohlc/optimise (mi-casa) ──► POST /backtest (ohlc-service)
                                                         │
                                               backtestService.runBacktest
                                                         │
                                    ┌────────────────────┴──────────────────────┐
                                    │                                           │
                               saved_patterns                           walk_forward
                               (ohlcPatternRepo)                   (walkForwardScanner)
                                    │                                           │
                         createBacktestService                createWalkForwardBacktestService
                                    │                                           │
                              simulateTrade ◄─────── calculateSlTp ◄─── config.multipliers
                                    │
                              aggregateStats
                                    │
                            BacktestResult (SSE stream)

Two backtest engines exist, selected by optimisationMode:

saved_patterns (default) — uses patterns already stored in ohlc_patterns DB table.
walk_forward — re-scans candle history with a sliding window; no DB patterns required.

Both engines share simulateTrade and calculateSlTp.

2. Prerequisites for a Backtest to Run

#	Condition	Source
1	`OHLC_SERVICE_URL` and `OHLC_SERVICE_API_KEY` env vars set	`optimisationController.triggerOptimisation`
2	Caller is authenticated and has `admin` role	`optimisationRoutes.ts` — all three routes guard with `requireRole('admin')`
3	Valid `symbol`, `timeframe` in request body	Zod schema in `backtestRoute.ts`
4	`ohlc_patterns` table has data for symbol+timeframe (saved_patterns mode)	`ohlcPatternRepository.getBySymbolAndTimeframeWithDetectionTime`
5	`ohlc_candles` table has data for symbol+timeframe (walk_forward + prz_touch/candle_close)	`ohlcCandleRepository.getByRange` / `getFrom`

3. Entry Strategy Simulation Logic

Each pattern produces one BacktestTrade. The entry price, SL/TP anchor, and candles used for simulation depend on tradingConfig.entryStrategy:

3a. `swing_confirmation` (default)

Step	Detail
SL/TP anchor	D-point (`dPrice`)
Entry candle	Fixed: `futureCandles[PIVOT_LOOKBACK + 1]` — the 6th candle after `detectedAt`; user `entryLagCandles` is ignored for this strategy
Entry price	`futureCandles[6].open` (or `dPrice` if fewer than 7 future candles exist)
Simulation starts	From `detectedAt + 1ms`
Outcome when no candles	`'open'`

3b. `prz_touch`

Step	Detail
PRZ band	Prefer `pattern.przLow/High`; fall back to SL-derived band from `calculateSlTp`
Touch detection	`findPrzTouchCandle(candlesFromCPoint, przLow, przHigh, isBullish)` — first candle where `low ≤ przHigh` (bullish) or `high ≥ przLow` (bearish)
Entry price	Touch candle’s `low` (bullish) or `high` (bearish) — the extreme that entered the PRZ
SL/TP anchor	`przTouchPrice` (overrides `dPrice`)
Simulation starts	Candles strictly after touch candle
Outcome when no touch	`'open'` — PRZ never reached

3c. `candle_close`

Same as prz_touch for touch detection, then:

Step	Detail
Entry price	Open of the candle AFTER the touch candle
Simulation	Passes `entryLagCandles: 1` to `simulateTrade`, candles from touch candle onward
Outcome when no candle after touch	Uses `przTouchPrice` as fallback entry

3d. `abcd` (emerging ABC path)

ABCD backtests use the emerging ABC path, not swing_confirmation: they validate a consecutive XABC sequence, wait for the first PRZ touch after C, and enter at the touch price. swing_confirmation remains the behavior for the other harmonic strategies only.

4. SL/TP Calculation

calculateSlTp(input, config?) in slTpCalculator.ts:

Try candlestick pattern formulas (computeCandlestickSlTp)
Fall back to pattern-type specific formulas (harmonic ratios, head-and-shoulders, cypher, etc.)
Apply multipliers from config: sl *= slMultiplier, tp1 *= tp1Multiplier, etc. (scaled relative to D-point)

The returned SlTpLevels = { sl, tp1, tp2, tp3 } are all absolute prices.

5. Trade Simulation (`simulateTrade`)

Runs candle-by-candle after entry. Uses these config fields:

Config field	Effect
`entryLagCandles`	Skip N candles; use `candlesAfterD[N].open` as entry price
`tpTargetMode`	`'tp1'` closes at TP1; `'tp1_tp2'` closes at TP2; `'tp1_tp2_tp3'` closes at TP3
`tpCount`	Legacy alias — mapped to `tpTargetMode` (1→tp1, 2→tp1_tp2, 3→tp1_tp2_tp3)
`partialExitEnabled`	When true: close `partialExitTp1Percent`% at TP1, move SL to breakeven
`partialExitTp1Percent`	% of position to exit at TP1 (default: 33.3)
`partialClose`	Legacy boolean alias for `partialExitEnabled`

Resolution rules (non-partial path)

On each candle c:

SL hit: isBullish ? c.low ≤ sl : c.high ≥ sl — always wins on the same candle as TP
TP hit: isBullish ? c.high ≥ targetTp : c.low ≤ targetTp

No candle hit → outcome 'open'.

Partial-exit path (3 phases)

Phase	Condition	Action
Phase 1	Pre-TP1	SL beats TP1 on same candle; TP1 hit → partial exit + SL → breakeven
Phase 2	Post-TP1, pre-TP2	SL (at breakeven) → remainder exits at 0; TP2 hit → close all (tp1_tp2) or become milestone (tp1_tp2_tp3)
Phase 3	Post-TP2	SL (at breakeven) → remainder exits at 0; TP3 hit → close all

6. Signal Score Filtering

When tradingConfig.minSignalScore is set:

filteredPatterns = allPatterns.filter(p =>
  p.signalScore == null  ← null scores PASS through
  || parseFloat(p.signalScore) >= minSignalScore
)

⚠️ Bug XXX: Patterns without a signalScore (null) always pass through the filter even when minSignalScore > 0. In live trading, events with null/low confidence are blocked by the engine’s minSignalScore guard. Backtest results for configs with a non-zero minSignalScore are therefore over-optimistic: they include patterns the live engine would reject.

7. Walk-Forward Mode

createWalkForwardScanner slides a window of WALK_FORWARD_WINDOW_SIZE = 300 candles:

At each position, run detectEmergingPatterns on the window.
Deduplicate by patternType:direction:x.time:c.time:fibLabel — only first occurrence (with firstEmergingAt) is kept.
After scan, resolve przTouchTime/Price for each pattern from candles after firstEmergingAt.

The walk-forward backtest then runs simulateTrade on each scanned pattern with every config combination (50 grid configs per pattern).

8. Backtest Result Shape

BacktestResult {
  symbol, timeframe, patternType,
  totalPatterns,     // all patterns (including open)
  completedTrades,   // trades with outcome !== 'open'
  openTrades,        // trades with outcome === 'open'
  wins,              // completed with outcome !== 'sl'
  losses,            // completed with outcome === 'sl'
  winRate,           // wins / completedTrades
  avgWinPips, avgLossPips,
  profitFactor,      // null when no losses
  totalPnlPips,
  maxDrawdownPips, maxDrawdownPct,  // from sorted-by-entryTime pnl sequence
  trades: BacktestTrade[]
}

Note: maxDrawdown is computed on a totalPnlPips equity curve sorted by entryTime. If entryTime is inaccurate (see Bug above), drawdown ordering may be wrong.

9. API

`POST /api/ohlc/backtest/strategy` (proxied from mi-casa)

Auth: JWT + markets access
Body: { strategyType, symbol, timeframe, from, to?, params?, entryLagCandles?, tradingConfig? }
Response: SSE stream of { type: 'progress' | 'result' | 'error', ... } events

The legacy pattern backtest route and saved-pattern engine (createBacktestService.ts) were removed in issues #920 and #921. Backtest and optimisation are now strategy-only (harmonic patterns: cypher, gartley, bat, butterfly, crab, shark, abcd).

`POST /api/ohlc/optimise` (proxied from mi-casa)

Auth: JWT + admin role
Triggers optimisation which calls runBacktest 50 times internally (see Optimisation System — Dependency Map)

10. Known Bugs / Limitations

ID	Summary	Severity
#461	`entryPrice` in output is lag=0 open even when `entryLagCandles > 0` — fixed for `swing_confirmation` (lag hardcoded to 6; `entryPrice` set to `lagCandle.open`)	Medium
#460	`minSignalScore` filter: patterns with null signalScore always pass	Medium
By design	`walk_forward` patterns pre-compute SL/TP from PRZ midpoint; actual simulation may use different anchor (przTouchPrice)	Low / cosmetic

mi-casa Docs

Explorer

Backtest System — Dependency Map

Backtest System — Dependency Map

1. System Overview

2. Prerequisites for a Backtest to Run

3. Entry Strategy Simulation Logic

3a. `swing_confirmation` (default)

3b. `prz_touch`

3c. `candle_close`

3d. `abcd` (emerging ABC path)

4. SL/TP Calculation

5. Trade Simulation (`simulateTrade`)

Resolution rules (non-partial path)

Partial-exit path (3 phases)

6. Signal Score Filtering

7. Walk-Forward Mode

8. Backtest Result Shape

9. API

`POST /api/ohlc/backtest/strategy` (proxied from mi-casa)

`POST /api/ohlc/optimise` (proxied from mi-casa)

10. Known Bugs / Limitations

Graph View

Table of Contents

Backlinks

mi-casa Docs

Explorer

Backtest System — Dependency Map

Backtest System — Dependency Map

1. System Overview

2. Prerequisites for a Backtest to Run

3. Entry Strategy Simulation Logic

3a. swing_confirmation (default)

3b. prz_touch

3c. candle_close

3d. abcd (emerging ABC path)

4. SL/TP Calculation

5. Trade Simulation (simulateTrade)

Resolution rules (non-partial path)

Partial-exit path (3 phases)

6. Signal Score Filtering

7. Walk-Forward Mode

8. Backtest Result Shape

9. API

POST /api/ohlc/backtest/strategy (proxied from mi-casa)

POST /api/ohlc/optimise (proxied from mi-casa)

10. Known Bugs / Limitations

Graph View

Table of Contents

Backlinks

3a. `swing_confirmation` (default)

3b. `prz_touch`

3c. `candle_close`

3d. `abcd` (emerging ABC path)

5. Trade Simulation (`simulateTrade`)

`POST /api/ohlc/backtest/strategy` (proxied from mi-casa)

`POST /api/ohlc/optimise` (proxied from mi-casa)