AI chatbots answer market questions with confident numbers that are often impossible to trace. These seven checks work on any AI research tool. Under each one is how AstraEye Labs meets it.
Here is one statement, word for word, from an AstraEye report of 27 October 2026, taken apart into the pieces the seven checks look for.
Within crypto, 287 of 4753 measured pairs (6.0%) had a correlation above 0.70 and the median pair correlation was 0.417, which ATLAS reads as high clustering (90 days).
Ask the tool where a figure came from. A number produced by a data lookup or a calculation can be traced; a number a language model wrote from memory cannot, and it can be confidently wrong.
In a 2023 study (Kang and Liu, 2023), open models asked for historical stock prices with no data tool scored 0%. Given a price lookup tool, the same models reached 98 to 100%.
Every figure in an AstraEye report is computed in code by a specialist AI scientist on licensed market data. The model explains the results; it does not produce them.
A credible report names the data it used, so you can open the same source and compare.
A figure without a named source cannot be checked, however precise it looks.
Every report ends with its data sources, credited to each provider, and the attributions page lists the terms behind them.
Two tools can report different values for the same idea, because they measure it differently. The method is what lets you tell which one fits your question.
Without the method, a reading like 'momentum is strong' is an opinion, not a measurement.
Each analysis carries its method in the report: what it measures, the rule it applies, and what it cannot support. SAGE answers questions about any of it.
'Up 12%' means nothing until you know over which period and as of which close. Check that every figure carries its window and date.
Markets move daily; a figure with no date may be stale, and one with no window cannot be compared.
Report statements carry their window and as-of date, for example '62 closes from 29 July to 26 October 2026' or '(1 day)'.
No source covers everything. A trustworthy tool tells you which assets or readings were missing, instead of filling the gap with a guess.
A gap reported as a gap is information; a gap filled silently looks like full coverage and is not.
Reports state their coverage per analysis and name what was not measured and why, rather than reporting a starting value as a reading.
If you follow stocks and crypto, check whether the tool researches both, or whether you would be stitching two products together.
Separate tools disagree on dates, windows and definitions, and reconciling them by hand is where errors creep in.
One report across U.S. stocks, ETFs and crypto: technical structure, fundamentals and macro, news and sentiment, on-chain data, derivatives and discovery screens.
The best check is one you will actually run. A tool inside the assistant you already use, with its price shown before it runs, is one you will keep using.
Research you have to switch apps for, or whose cost you only learn afterwards, tends to be skipped.
AstraEye shows every run's price in credits before it starts, and it works inside the AI assistant you already use: ChatGPT, Claude, Gemini, Perplexity, Grok or Glama.
Open a published AstraEye report, no account needed, and run the seven checks yourself: the data sources at the end, the method for each analysis, and the window on every figure.
A language model writes text from patterns. Without a data lookup, it can recall or estimate a figure and state it confidently. In a 2023 study (Kang and Liu, 2023), open models asked for historical stock prices without a data tool scored 0%; with a price lookup tool they reached 98 to 100%.
Ask for its source and its date, then compare it with the original: the company's filing for fundamentals, or the exchange or price provider for market data. Check the window it covers and how it was calculated.
Yes. One report can cover U.S. stocks, ETFs and crypto, from specialist AI scientists that compute every figure in code on licensed market data.
Yes. AstraEye Labs connects to ChatGPT, Claude, Gemini, Perplexity, Grok or Glama as an app, so the assistant can read your finished reports and their methods, and ask SAGE to price and start new research.
There is a free plan, and paid plans start at $19 a month (Space Analyst). New accounts get a 7-day trial with no card. Every research run shows its price in credits before it starts.
Study cited: Kang and Liu, 2023, Deficiency of Large Language Models in Finance. Informational research only. AstraEye Labs does not provide financial advice or recommendations. You decide what, if anything, to do.