Inspiration Before making a purchase, investors should be able to check it against rules they chose themselves—not rely on a black-box recommendation. ETHOS makes those personal constraints visible and open to question.
What it does ETHOS simulates a proposed portfolio purchase and checks it against the investor’s confirmed rules. It shows the returned results and evidence, including when missing or partial data makes an answer uncertain. Users can inspect the evidence, challenge a figure, or change a rule and compare results. It is a fictional simulation—not financial advice, a recommendation, a forecast, or a trading tool.
How we built it We built a static-exported Next.js app with a typed frontend and shared API contracts. A deterministic Python engine calculates the checks and verdicts; AI-assisted services support parts of the workflow, but do not decide the verdict. The app also has a fixtures mode for replaying recorded test scenarios.
Challenges we ran into We had to make incomplete data show up as uncertainty rather than silently treating it as zero. We also needed to keep user-confirmed rules distinct from system-proposed drafts, and make asynchronous results and evidence understandable without letting generated text determine the verdict.
Accomplishments that we’re proud of ETHOS puts the investor’s own rules at the center of the check. It makes the reasoning inspectable, distinguishes “unknown” from “pass,” and lets users challenge recorded figures or compare results under a changed rule.
What we learned Trust depends on showing not just an answer, but what supports it and where its limits are. We also learned that turning a preference into a rule should involve clarification and explicit confirmation—not guesswork.
What’s next for ETHOS Finish and verify the English/French experience across the full app, validate the browser journeys, and continue improving how evidence sources and uncertainty are presented.