Live Strategy Engine

Like the alerts engine, the strategy engine runs as one long-lived process (dslx_cli --serve) instead of one evaluator per strategy. A client opens one connection and sends any number of requests on it — add, replace, pause, resume and remove strategies, push market data, run backtests — with many requests in flight at once. Replies are matched by id; entry, exit and error events are pushed to every connected client.

• Shared feeds. Strategies on the same market symbol and timeframe share one bar history. Bars are built from ticks, or taken from closed candles when those arrive (a feed that receives candles stops building bars from ticks).
• Shared indicators. When a feed confirms a bar, every strategy on it is evaluated with one indicator cache, so an indicator used by many strategies is computed once. The strategies are then stepped in parallel; events come back in a fixed order (exits before entries).
• Fresh requests at any time. A strategy added to a feed with history is warmed over it without trading — edge-triggered signals learn whether they were already true — and trades from the next bar. History is trimmed at 5,000 bars to 3,000 by default.
• Backtests alongside. Backtests run on their own thread pool and reply when done, while ticks keep flowing on the same connection.

Each live strategy runs the same bar step as a backtest: signals at the bar close, stops and targets against the confirmed bar's range, the same sizing and costs. Streaming a history bar by bar reproduces the backtest's trades; the checks compare hundreds of strategies trade for trade. The differences are deliberate: a live strategy keeps positions open at the end, and stops and targets are evaluated when the bar closes, not tick by tick.

→ {"id":1,"op":"add","strategyId":"42","feed":"BINANCE:BTCUSDT","strategy":{…},"seed":[…candles…],"meta":{"userId":7}}
← {"id":1,"ok":true,"strategyId":"42","feed":"BINANCE:BTCUSDT","timeframe":"5m","bars":1000,"sharedWith":12,"warnings":[]}

→ {"id":2,"op":"ticks","ticks":[{"symbol":"BINANCE:BTCUSDT","price":64010.5,"volume":0.3,"time":1717000000000}]}
← {"event":"entry","strategy":"42","feed":"BINANCE:BTCUSDT","timeframe":"5m","barTime":1716999900000,
   "side":"long","price":64001.2,"quantity":0.39,"stop":62721.1,"targets":[66561.2],"openPositions":1}
← {"id":2,"ok":true,"events":1}

→ {"id":3,"op":"backtest","strategy":{…},"candles":[…]}        // replies when done; live traffic continues
← {"id":3,"ok":true,"result":{"trades":[…],"stats":{…},"signalStats":[…]}}

→ {"id":4,"op":"stats"}
← {"id":4,"ok":true,"stats":{"strategies":500,"feeds":5,"strategySteps":300000,"seriesRequested":900000,"seriesComputed":21000,…}}

Errors reply {"id":…,"ok":false,"error":"…","path":"…"} — an invalid strategy is refused at add with the same JSON path the editors show.

The bot API's strategy runner keeps one engine connection. Every deployed strategy saved with an engine DSL is added to it; ticks from the market-data hub are forwarded in batches and closed candles as they arrive; history is seeded from the candle aggregator; engine entry and exit events become strategy signals. If the engine restarts, the runner reconnects and re-adds its strategies. Strategies saved before the builder wrote the engine DSL keep the older evaluator until they are saved again.