Loan Approval

Nine flat fields drive the decision: creditScore, annualIncome, monthlyDebt and monthlyIncome, employmentStatus with employmentYears, plus applicantId, loanAmount, and loanTerm for context. The four scoring stages downstream each read a different slice of this request.

A first hit policy walks the creditScore bands from the top: '>= 750' returns "excellent" with 30 creditScorePoints, [700..749] "good" with 25, [650..699] "fair" with 15, [600..649] "poor" with 5, and anything under 600 "very_poor" with 0. Ordering the rows from best to worst means each application lands in exactly one band.

The cutoffs track the credit-score tiers most US lenders work with, where scores in the mid-700s and above price as prime and anything under 600 reads as subprime. Weighting the top band at 30 points, the largest single contribution in this graph, reflects how heavily mortgage underwriting relies on credit history as a default predictor.

Annual income falls into one of four brackets under a first hit policy: '>= 100000' maps to "high" with 25 incomePoints, [60000..99999] to "medium" with 20, [30000..59999] to "low" with 10, and below 30000 to "very_low" with 0.

These breakpoints are a business calibration rather than a regulatory requirement: 60000 sits near a typical household income for a mid-size mortgage, and the 0-point floor under 30000 stops low earners from accumulating qualification points on income alone. The incomePoints feed both the rejection check and the risk-priced rate later in the graph.

Dividing monthlyDebt by monthlyIncome and multiplying by 100 produces dtiRatio, the percentage of gross monthly income already committed to debt service. Debt-to-income is the standard affordability measure in mortgage underwriting, and the value computed here is tested against the 40 threshold in the rejection table and folded into the risk score that prices the loan.

Stability is scored from paired conditions on employmentStatus and employmentYears with a first hit policy: full_time with 5 or more years earns "very_stable" and 20 employmentPoints, full_time at 2 or more years "stable" and 15, self_employed needs 3 years for the same "stable" 15, part_time tops out at "moderate" with 5, and unemployed is always "unstable" with 0.

Requiring a longer track record from self_employed applicants than from salaried ones mirrors standard mortgage practice, where lenders typically want around two years of documented self-employment income before treating it as stable. The part_time ceiling and the zero for unemployed reflect income continuity risk rather than any regulatory rule.

Every failing check contributes here because the hit policy is collect: creditScorePoints < 15 adds "Poor credit history", incomePoints < 10 adds "Insufficient income", dtiRatio > 40 adds "High debt-to-income ratio", and employmentPoints < 10 adds "Unstable employment history". The messages accumulate into the rejectionReasons array via the table's outputPath, so an applicant can fail on several grounds at once.

Returning specific reasons instead of a bare decline matches US adverse action practice, where ECOA and Regulation B require lenders to state the principal reasons for a denial. The 40 percent DTI cutoff sits just under the 43 percent benchmark long associated with qualified mortgage rules, while the point floors simply restate the weaker bands of the upstream scoring tables.

Applications split on a single test: len(rejectionReasons) == 0 sends the request down the approval branch to interest rate pricing, and the default branch catches everything else as a rejection. Keeping the branch condition tied to the collected reasons list means the approval logic never restates the underwriting thresholds themselves.

Approved applicants get risk-based pricing: totalRiskScore sums creditScorePoints, incomePoints, employmentPoints, and dtiRatio, then riskAdjustment is computed as 100 - $.totalRiskScore and the final interestRate adds 0.05 points per adjustment unit to a 5.5 baseRate. Stronger applicants therefore pay closer to the base rate, which is how lenders translate a scorecard into a personalized offer.

On the declined branch the response is reduced to two keys: rejectionReasons carries the collected messages forward and approval is set to false. With passThrough disabled, the caller receives a clean rejection payload instead of the full scoring state.