Results

Same events. Who should you trust?

We gave Claude and GPT the same made-up event packs that ARGUS’s formulas saw. Models were fine at spotting surges and ranking “which place looks hotter.” They were weak at naming the risk level, testing a claim, and saying how sure you should be.

Job Plain question What we saw Better choice
Spot a surge Is activity really jumping? Models matched 100% Either
Rank places Which country looks hotter? Models ranked well Models (for sorting)
Name the risk level Is it LOW or CRITICAL? Only ~11% matched ARGUS ARGUS
Test a claim Does this report support or contradict? Only ~40–53% matched ARGUS
Say how sure How confident should the brief be? GPT overclaimed ARGUS

Takeaway. Use a model to help write or sort. Use ARGUS for the grades that decide what goes in the brief.

How the formulas work: methods.

Get ARGUS

Free. Open source. Clone it and run it on your own computer.

Open on GitHub · How it works · About