Validate a data analysis with AI
Review the calculations, comparison periods and conclusions behind a report, then decide which issues need attention before it is shared.
To validate a data analysis with AI, supply the reasoning behind the chart as well as the finished slide. If an order total is joined to every item in the basket, summing those rows can count the same money several times. The validate-data guide asks for checks that can expose problems like this before a decision rests on the figure.
Other faults survive a perfect calculation. The latest month may still be incomplete, or the narrative may attribute a change to an event that merely happened at the same time. The review can distinguish a calculation error from a missing caveat and identify the evidence needed to resolve either.
What it does
- Check the unit represented by each row
- Spot duplicated totals after a join
- Compare date coverage and percentage denominators
- Recalculate a sample against a reference source
- Flag a claim that overstates what the data shows
Install and use
- Install the skill and provide the report, metric definition and authorised query or source extract. Recalculating values requires access to the relevant underlying figures.
- Ask “Check this average order value report. Recalculate three orders after the basket-line join and explain any difference.”
- Resolve findings that change the conclusion and retain the scope and limitations with the revised report.
FAQ
Source
Made by Anthropic. Original skill: validate-data, published on GitHub under the Apache-2.0 licence.
View on GitHub