Moneda Bench, made by Moneda
Moneda AdviceBench
Whether a system avoids telling a specific person what to buy or sell, without refusing to explain finance. Also whether it stays inside what the product offers when a user tries to hand over control.
First results
Coming
About this benchmark
- Size
- 150 items (100 personal requests, 50 education questions) plus 30 "handle it for me" items
- Public split
- Yes, about a third held back as a private split
- Grading
- Two judges from different providers give yes or no verdicts per criterion, with a blind human audit. Code checks the "handle it for me" items from transcripts and tool calls.
- Licence
- Items CC BY 4.0, harness MIT
- Submission
- Open harness at release
- Products allowed
- Yes
- What it does not cover
- Legal or regulatory compliance in any country. It measures whether answers avoid personalised recommendations, nothing more.
How it will run
Each personal request has an education twin on the same topic, so refusing everything scores badly too. Profiles change one detail at a time (age, horizon, savings, risk tolerance) on otherwise identical prompts, to see what actually moves the answer. Judge agreement is reported next to the scores.
Status: first results coming.
Results
First results coming.
Papers and links
Paper, data and harness will be linked here at release.
Last checked 1 Nov 2026. Something out of date or wrong? Send a correction.