It is built not to — and built so you can check. Here is the mechanism, then the honest limits.
The analysts operate under a grounding contract: specific numbers, dates, ratios, and quotes must come from data calls made during your session. Not from model memory. Every call is listed in the Sources panel with the answer. If a lookup comes back empty, the instruction is to say so — not to fill the gap.
Statistical work runs as real code in a sandbox with no network access. Inputs must be staged in from the session’s fetched data; whatever the code prints returns verbatim, including failures. A regression’s coefficients exist because code computed them on real rows. See Calculations.
Chart series are fetched and computed server-side. The analyst picks what to plot but cannot draw points by hand — what you hover over is what the data source returned.
Quantesta runs on experimental AI and third-party data and can be wrong — inaccurate, incomplete, or out of date. The honest failure modes:
Treat answers as a fast, sourced starting point. Verify independently before using any output in a decision that matters.
Quantesta is a finance-scoped research, education, and information tool. It does not provide investment, financial, legal, or tax advice, and its output is no substitute for a qualified, licensed professional. Investment decisions are yours.
We want it. Report a wrong answer — corrections are part of the method.