The engine, running live.
This page calls the same suggestion engine that ships with the product, against the sample portfolio below. Every card is generated — not hardcoded. Refresh and the math runs again.
Sample portfolio · illustrative
$96,668
Operating cash
$60,000
Brokerage market value
$36,668
Monthly expenses
$9,500
3 suggestions proposed
Concentration · 80% confidence
VTSAX is ~17.2% of your marketable holdings — above the ~10% single-position level commonly treated as elevated concentration
Observation calculated from your own linked-account data — not investment advice and not a recommendation to buy or sell any security. A high single-position weight is a diversification observation, not a prediction; whether it is appropriate depends on your goals, time horizon, and overall financial picture. Nova is not a registered investment adviser. Consult your fiduciary before acting.
Idle cash · 85% confidence
$31,500 of cash is earning ~4.82% below money-market yield — if moved, the estimated annual difference would be ~$1,518
Scenario calculated from your linked-account data. Not investment advice and not a recommendation to buy or sell any security. Verify current rates and your own tax situation before acting.
Tax-loss harvest · 60% confidence
Expires 12/31/2026
~$2,629 in unrealized losses available — if harvested, the estimated tax benefit would be $574–$1,021
Informational scenario calculated from your own linked-account data — not investment advice and not a recommendation to buy or sell any security. Nova is not a registered investment adviser. Wash-sale rules and your specific lot accounting may change the outcome; consult your tax preparer or fiduciary before acting.
How this works
Every suggestion is a pure function over a portfolio snapshot. The reasoning field is structured JSON — inputs with provenance, named arithmetic steps, and a confidence band with plain-language drivers. The UI never invents prose; it renders the structure.
Why it matters
You can audit every suggestion. The math is in the open. No black-box LLM telling you what to do — just deterministic rules against your real data, with the work shown.