From idea to a NinjaTrader chart, with a verdict in between.

Step 1 · Forge

Compose the setup from blocks

Three ways to start one. Answer a short form that reads like a sentence — only look for trades when…, then take it when…, get in by…, risk it like this…, get out when… — and Forge builds the setup from your answers, then shows you what it built in plain English. Duplicate a setup that already works, name the copy for whatever you are changing, and experiment on it while the original stays untouched. Or open the canvas and place the blocks yourself.

A setup is a short pipeline: gates that must hold, a trigger that fires, an entry style, brackets (stop and target), exits (trail, time stop, thesis-loss), and sizing. Every block has typed parameters with sensible defaults drawn from the data itself.

• 150+ signals as blocks: higher-timeframe bias, fair-value gaps, swing failures and their invalidation, order-flow imbalance, session and previous-day levels, VWAP and its standard-deviation bands, changes in the state of delivery, the previous day’s verdict on the day before it, consolidation, ATR bands, time-of-day and day-of-week windows.
• Nested AND / OR condition groups — express “aligned bias AND (displacement OR grab chip)” without code.
• A signal library of typed presets so you start from a working block, not a blank form.
• Composite events: a trigger can fire on the bar a whole condition set turns true, take its direction from a signal, and place its limit at a level the signal names.
• Stops that sit where the idea is wrong: a fixed distance, or beyond the protected extreme of the recent swing, with the target following the distance actually risked.

Try changes freely: Prove it runs your unsaved edits as a draft. Nothing is recorded until you save. A duplicate never inherits the original’s verdict — a copy starts unproven and has to earn its own.

Step 2 · ProofBench

Prove it on data the setup never saw

ProofBench is a backtest engine that refuses to run without a holdout. You declare training months and holdout months; the engine runs both and compares them. The result is a stamp — VALIDATED, WEAK or REJECTED — with its reasons spelled out: does the holdout replicate the training rate, does the equity behave the same way, are the trades enough to mean anything.

Fills are honest by construction (see Proof): an order can exist only after the signal bar closes, and a target is never awarded on the fill bar from its extremes. Costs are charged per leg. Results ship as an equity curve, monthly and weekly tables, entry-hour maps, drawdown and worst-day markers.

The engine runs on our servers against a curated dataset that grows by a month at a time; you never handle files.

Step 3 · Bridge

Run only what passed

The Bridge is a single NinjaTrader 8 strategy. It has no strategy logic in its parameters — it polls ConnectQ for the setup assigned to that machine and chart slot, shows you the full picture on the chart, and executes the setup exactly as it was tested: same gates, same entry level, same brackets, same exits.

• Pair a machine with its NinjaTrader Machine ID; ConnectQ approves it; you deploy a proven setup to a chart slot from Forge.
• Trading is enabled server-side. There is no checkbox on the chart that can arm it.
• Stale config disarms: if the Bridge cannot reach ConnectQ for longer than its time-to-live, it stops arming new trades.
• A setup the Bridge cannot execute exactly is refused with a logged reason — never approximated.

Read about the NinjaTrader client and the parity method that keeps the engine and the chart honest with each other.