OneScore: same credit score, different cash flow

OneScore reads an applicant's real transaction history instead of a bureau-style credit score, and explains every decision in Regulation B adverse-action language. It was built for VT Hacks 2026 on Capital One's Nessie sandbox API. All customers and outcomes are synthetic.

Example: two applicants, one traditional score

Applicants A and B both have a traditional credit score of 654, which sees only bill history and account age, so it cannot tell them apart. Their cash flow differs sharply:

• Spending volatility: A $264, B $938
• Cash buffer: A 226 days, B 17 days
• Balance trend: A +$1018/mo, B +$98/mo

Decision: A is approved with a $950 credit line. B is declined, with these Regulation B reasons: (1) insufficient cash reserves relative to monthly expenses; (2) irregular or unpredictable monthly expenses.

How it works

1. Capital One Nessie API: accounts, deposits, withdrawals and purchases are read live.
2. Cash-flow features: income regularity, cash buffer, spending volatility and more, computed from a year of transactions.
3. OneScore model: a simple, inspectable model scores repayment likelihood out of 100, so every score traces back to its causes.
4. Regulation B reasons: the factors that lowered the score become plain-language adverse-action reasons.

Why it matters

Credit scores miss cash flow: they cannot see how steady income is, how big the cash cushion is, or how often the balance runs dry. Regulation B requires lenders to tell declined applicants the specific reasons, and OneScore builds every decision from named factors.

Explore

• Compare applicants: pick any two of 600 customers and see who is approved and why.
• Fairness review: approval rates by group against the four-fifths rule.
• Model and metrics: accuracy, precision, recall and the simple model behind it.

In production, OneScore would read a bank's core transaction stream, and the reason wording would go through compliance and legal review before use in a real adverse-action notice.