Flow. Add a symbol → when the live-engine toggle above is ON, the engine evaluates it on every completed 5-minute bar using your Alpaca keys from Integrations → Broker → signals are delivered into the same pipeline as your TradingView webhooks: they appear in Raw payloads received, are deduped and risk-checked, then auto-dispatch or queue for approval per your Execution mode. Pause suppresses new entries but exits keep protecting an open position. Archive permanently stops evaluation.
Params. Leave overrides empty to run the strategy exactly at its Pine defaults (Balanced preset). Overrides are JSON — the chips below the field show each value as default → override, and the full table in Engine parameters shows the complete effective configuration. Note: individual filter toggles only take effect with "signalPreset": "Custom" — the Balanced / Strict / Loose presets force their own filter set, exactly like the Pine.
Examples:

{"dirMode": "Long only", "maxTradesPerDay": 1}

{"signalPreset": "Custom", "useMarketFilter": true, "volMult": 1.2}

{"useFixedDailyCapital": true, "fixedDailyCapitalUsd": 25000, "riskPerTrade": 0.5}

{"useScaleOut": true, "scaleOutPct": 50, "scaleOutR": 1.0}

Flow. Pick a symbol and a date range (max ~1 year). The engine fetches 5-minute SIP bars (split-adjusted, extended hours — same data shape as your TradingView chart), warms indicators on ~25 prior days, and replays the strategy with the same code that trades live. Fills follow the Pine's model: signals on bar close, brackets fill intrabar, 1-tick slippage, $0.005/share commission.
Fill engine: "OHLC path" matches TV's 4-tick heuristic (use for parity against TV exports); "Tick replay" matches orders against the real recorded tape — stop-vs-target ordering, timing and prices come from actual prints, with a 250ms latency model and limit fills on touch. It is MORE accurate than TV, so expect legitimate differences from TV exports; judge live-readiness on it. Slower (downloads tape where orders were live).
Fill model controls the $-profit-take: Realistic intrabar exits at the first touch of the threshold price (what live execution achieves — judge live-readiness on this one); Bar close checks it once per bar like the literal Pine; TV-optimistic fills at the bar's best price to approximate TV's on-order-fill emulator — use it ONLY to compare against a TV export made with "Script execution: On bar close, On order fill".
Overrides use the same JSON schema as alerts — chips below the field show default → override. The date range maps to the Pine's "Date range" test window, so indicators still see prior history.
Verify against TradingView: run the same symbol / window / settings in the TV Strategy Tester, export its List of Trades, then on the server run

npm run backtest -- --symbol AAPL --from 2026-01-02 --to 2026-06-30 --out res/aapl
npm run parity  -- --engine res/aapl.trades.csv --tv tv-export.csv

A clean parity report is the proof the engine reproduces your Pine before you trade it live.
Examples:

{"signalPreset": "Strict"}

{"usePyramiding": true, "maxAddsCap": 4, "useDailyTradeLimit": false}

Each trial is a full config (base params + sampled search values) scored on every walk-forward fold with the costs above (realistic intrabar fills). Constraints (min trades/fold, max drawdown) filter; score = average fold net profit; the top 8 finalists are re-run with ±10% jitter and ranked by the plateau (median of neighbors) — a config that only works at exact values loses to one that works in a region. Max 12 searched params; pin everything else. Examples:

{"profitTakeUsd": {"min": 10, "max": 100, "step": 5}, "useProfitTakeUsd": {"options": [true, false]}}

{"signalPreset": {"options": ["Balanced", "Strict"]}, "maxTradesPerDay": {"min": 1, "max": 8}, "rMultiple": {"min": 1, "max": 3, "step": 0.25}}