Language for the question. Mathematics for the numbers.

Large language models are remarkable at language. They read a question, weigh billions of learned patterns and produce the most plausible continuation, word by word. That is what makes them feel fluent, and it is genuinely useful: they summarise, explain, translate and reason about text better than any software before them.

But plausible is not the same as true. A language model is a probabilistic system: when it writes a number, it is predicting what a number should look like in that sentence, not calculating it. Ask twice and you may get two different answers, each delivered with equal confidence. The industry calls the failure mode hallucination. In most settings it is an inconvenience.

In portfolio management it is disqualifying. A risk figure that is merely plausible is not a risk figure. You deserve numbers you can trust and defend, not numbers that rely on the probabilities of a text generator.

Harry uses the language model for exactly what it is good at: mapping the wording of your question onto a defined analysis, and putting the result back into readable prose. The moment a figure is needed, Harry hands the question to Cadran's quantitative engine, the same deterministic engine behind every dashboard on the platform, and passes the verified result through untouched.

The consequence is simple but rare in AI products. Every figure Harry reports is rendered from a computed result and linked to the calculation that produced it. Harry does not generate numbers in prose. Ask about concentration, drift or factor exposure and the figures in the answer are the figures in the platform, reproducible on demand and traceable to their data and method. If something cannot be computed, Harry says so instead of improvising.

We make no claim that Harry understands your portfolio. It matches questions to analyses and puts numbers into sentences. The judgement stays with you; the arithmetic stays with the engine.

The Cadran name points the same way: the dial of a Swiss mechanical watch. A movement does not estimate the time, gears either mesh or they do not. That is the standard Cadran holds its numbers to, Harry included. The machinery is deterministic; only the phrasing is generated.