You set the goal. Our agents do the quantitative work.

You write one sentence — a goal, an idea you want tested, a strategy you already trade. What comes back is not an opinion. It is a strategy package with the reasoning behind every decision, the evidence that supports it, the tests it survived, the tests it barely survived, and a starting capital level that was chosen by a posterior distribution rather than by enthusiasm. Then you decide whether any of it is deployed.

Using AI is not the same as delegating to it

Retail investors already use AI to decide: 62% of those surveyed say soVERIFIED — checked against a primary source^F20. Far fewer hand over the wheel: only 31% fully trust AI for financial adviceVERIFIED — checked against a primary source^F19. Read together, those two numbers are not a contradiction. They are a description of a reasonable adult: happy to use a tool that reads faster than they do, unwilling to let it sign for them.

Most products in this category argue with that instinct. They ask for trust up front and show the evidence afterwards, if at all. Quant24 is built the other way round. The agents do the work that is genuinely tedious — reading the research, proposing the mechanisms, running the statistics, writing down what was rejected and why — and then stop at the point where a person should decide. The gap between using and delegating is not a marketing problem to be argued away. It is the product boundary.

What actually happens when you write that sentence

The Strategist turns the goal into a contract. A sentence like "steady returns from index futures, nothing held overnight" is not testable. The Strategist rewrites it as a Goal Spec: a target metric, a benchmark, a drawdown tolerance, a capital figure, a universe, and an explicit list of every ambiguity it resolved on your behalf. It asks at most one round of questions. If the goal cannot be measured at all, it proposes a reformulation instead of pretending it understood.

The engine, in code, produces every number. This is the part worth being pedantic about. The language agent decides what to test and interprets what came back. It never computes a result and never writes the final package. The history is split once — training, test, and a holdout sealed on the most recent periods that the design process never sees — and six engines run against it: walk-forward out of sample, the sealed holdout, a bootstrap over trade order, a resample conditioned on regime transitions, a stress test over the worst events in the record, and a Bayesian posterior that combines them. A backtest narrated by a model, with no code behind it, is not a backtest.

The Auditor reviews the process, not the score. Look-ahead, survivorship, how many hypotheses were tried before this one, whether costs and slippage were modeled in every engine, whether the holdout stayed sealed, whether the stated reasoning matches the actual rules. A mediocre result is delivered with its diagnosis. A dishonest process is sent back to design, no matter how good the curve looks.

The Executor runs only what you approved. Execution sits in code with its own circuit breakers. Above it, a risk guardian with a mathematical veto over size and exposure. Above that, a strategic layer that reviews context on a fixed cadence and whose adjustments must pass back through the guardian. No layer can skip the risk layer. That is architecture, not policy.

You approve, and you can revoke. Nothing reaches a market on an agent's judgment.

What we are not going to tell you

We are not going to tell you what you will earn. The figure that matters here is the one about the segment: around 70% of day traders lose moneyPROXY — an indirect figure standing in for one we do not have^F16F17F18, measured by other people, and a validation pipeline does not reverse it. Anyone who suggests otherwise is selling you the part they cannot deliver.

We are also not going to tell you this is finished. Quant24 is pre-launch. There is no broker connection, no capital and no exchange market data in this build. Every result you can reach is computed on synthetic data and labeled SYNTHETIC on the number itself, and where something cannot honestly be computed, the page says so instead of printing a figure.