Members · 05
The third setup piece, and the second way into the factory. The AI is how we build and ask questions. The console is how we look. It is the dashboard that turns the results database into pages a human can read at a glance: every system, every backtest, every live deployment.
The results database is the truth, but nobody enjoys reading a database. Early on we found ourselves asking the AI the same questions every morning: how are the live systems doing, which backtests ran overnight, what changed since yesterday. Those are questions for a screen, not a conversation. So we had the AI build us one. It runs on our own computer, it is read-only, and it reads from the results database and nowhere else.
The home page is the morning glance: which systems are live and healthy, how long until each broker login needs renewing, the combined live equity curve, a feed of what changed since yesterday, and the latest backtest runs. The systems page is the bench: every system, which stage of the gauntlet it has reached, and for each one a leaderboard of the variants tried, its equity curves, and, once live, its real results overlaid on what the tests predicted.
A run page is one backtest in full: the headline statistics, the equity and drawdown curves, a trade-by-trade curve, a timeline of what it held and when, and the full trade list, exportable. The compare page puts several runs side by side, with drawdown depth and duration statistics, which is where most "is variant B really better?" arguments get settled. And the live page is the monitor: every deployment's heartbeat, positions and recent orders, plus the twice-weekly broker re-login workflow, which is the one thing the console can actually do rather than merely show.
The console never calculates a financial number. A separate numbers step, written in Python, computes every statistic using the same library code the research itself uses, and writes the results to a read model the console displays. If a page needs a figure the numbers step does not produce, the numbers step grows, not the page. This sounds fussy. It is the reason the figure on the dashboard is, by construction, the same figure a research comparison would produce, and the reason we can trust what we glance at.
Ours is built in TypeScript because the AI is fluent in it and it makes good web pages. You do not need to build this on day one, or in this language. Start with QuantConnect's own web pages for viewing backtests, and when you find yourself asking your AI the same questions every morning, that is the moment to have it build you a console.