Three agents. One honest line.

«Questions come up the whole way: what does this metric mean, why did the backtest sag here, where do I find this in the platform — and the big one: the equity curve looks great, but will it hold up outside the backtest?» That last question is exactly where most strategies fall apart.

Bellatrix is the platform’s global AI assistant — it lives on every page, not just the agents workspace. It answers any question about the product, your strategies, trading and your stats; reviews the page you’re on — a strategy, a backtest report, your journal; explains any text you select, right in place. And when you ask for the truth about a strategy, it runs the checks a disciplined quant would — out-of-sample hold, profit concentration, losses by market regime, drawdown probability — and says it straight: «In-sample looks great, but it’s overfit — here’s the proof».

The answer lives where the question appears — no docs digging, no tab switching. And it’s still cheap insurance against putting your deposit on a strategy that only ever existed in backtest: the friend who tells you the truth before you risk money, not after.

No free builder or marketplace will tell you your strategy is garbage — much less prove it. It’s not in their interest to upset you. Bellatrix will — right on the page where the problem lives. It doesn’t build or fix strategies, though — that’s Rigel’s and Betelgeuse’s job. That’s the hook: you see the problem, you want the tool that solves it.

«I have an idea — «buy the BTC dip on 1h by RSI with a trend filter» — and I want to quickly know if it’s worth anything. I don’t want to spend a week wiring up indicators and I don’t want to learn quant theory.»

Turns an idea into a runnable, tested prototype in minutes — with the safeguards a pro would insist on baked in: there’s always a real stop; entry is an event trigger, not a forever-true state (otherwise the strategy fires once and stalls); there’s a regime gate so you’re not long during a crash; no degenerate «always-true» conditions. Even a quick prototype isn’t a beginner’s trap.

Compresses idea-to-validation from days to minutes — and the result is trustworthy enough to make a real call: «take it / pass». You can vet 5 ideas in an hour, and each one is honestly testable, not a toy.

Competitors’ fast generators give you speed or discipline. Their «AI strategy» is usually naive — no stop, state-entry, fit to noise. Rigel gives you speed and a baseline of quality (anti-overfit, anti-dead-signal) together.

«I don’t need a strategy that backtested well — I need one that has a reason to work and survives in the future and on other markets. Finding that is a full-time research job I don’t have time or skill for.»

The deep work — formulates a causal hypothesis (why this edge exists, who on the other side is systematically losing), grounds the design in a knowledge base of what actually works for the archetype/asset (RAG), builds the strategy in a dialogue with you — it asks rather than assumes — and validates hard: OOS holdout, multi-asset cross-check, statistical significance (signal or noise?), iteratively eliminates blockers, remembers across sessions what was tried and what didn’t work — so it doesn’t spin its wheels. It’s not an agent that grinds on to burn tokens: if the strategy is dead, it says there’s no point going further. And honestly: if after digging there’s no durable edge, it says so — instead of handing you a fit.

Replaces weeks of quant research. Value isn’t the strategy — it’s «a strategy that survived the tests that kill most strategies, plus an honest verdict on whether it’s real».

Competitor «AI generates strategies» optimize for backtest and stop. No honest OOS+cross-asset+significance gate, no causal reasoning, no learning across sessions. Output is qualitatively different: candidates for durable edge with evidence, not just a pretty backtest.