Formula
For each criterion: pass if file value ≥ minimum (or ≤ minimum when the name contains "inverse"). Margin = (value − minimum) ÷ |minimum| for normal criteria, or (minimum − value) ÷ |minimum| for inverse ones.
Points when passing = weight × min(1, 0.6 + 0.4 × clamp(margin, 0, 1)): 60 percent of the weight for passing plus up to 40 percent for margin, maxing out at a margin of 100 percent above the minimum.
Points when failing = weight × clamp(0.6 + margin, 0, 0.5): at most half the weight, falling to zero when the shortfall reaches 60 percent of the minimum.
Weighted score = (sum of points ÷ total weight) × 100. Decision: Approve at or above the approve threshold, Refer at or above the refer threshold, otherwise Decline.
Worked examples
Example 1: the default seven-criterion matrix scores 84.6 and approves
Inputs: seven lines with total weight 100; approve at 75, refer at 55. Months in business, weight 15, minimum 12, value 30: margin = (30 − 12) ÷ 12 = 1.5, clamped to 1, so points = 15 × (0.6 + 0.4) = 15. Average monthly deposits, weight 20, minimum 15,000, value 42,000: margin 1.8, clamped to 1, points = 20. Negative days (inverse), weight 15, minimum 5, value 2: margin = (5 − 2) ÷ 5 = 0.6, points = 15 × (0.6 + 0.24) = 12.6. NSF count (inverse), weight 10, minimum 3, value 1: margin 0.667, points = 10 × (0.6 + 0.267) = 8.7. Owner FICO, weight 15, minimum 550, value 640: margin = 90 ÷ 550 = 0.164, points = 15 × (0.6 + 0.065) = 10.0. DSCR, weight 15, minimum 1.1, value 1.35: margin = 0.25 ÷ 1.1 = 0.227, points = 15 × (0.6 + 0.091) = 10.4. Positions (inverse), weight 10, minimum 2, value 1: margin 0.5, points = 10 × 0.8 = 8. Total points = 15 + 20 + 12.6 + 8.7 + 10.0 + 10.4 + 8 = 84.6; score = 84.6 ÷ 100 × 100 = 84.6 out of 100. All seven pass, zero criteria failed, and 84.6 is above 75, so the decision is Approve. Note that FICO and DSCR, both passing comfortably in absolute terms, earn only two thirds of their weight because their margins are measured relative to the minimum: 640 is just 16 percent above 550.
Example 2: five failures produce a decline at 40
Inputs: the same seven criteria and thresholds with weaker file values: months 8, deposits 22,000, negative days 9, NSFs 4, FICO 560, DSCR 1.0, positions 3. Months: 8 < 12, fail; margin = (8 − 12) ÷ 12 = −0.333; points = 15 × clamp(0.6 − 0.333, 0, 0.5) = 15 × 0.267 = 4.0. Deposits: pass; margin 0.467; points = 20 × (0.6 + 0.187) = 15.7. Negative days: 9 > 5, fail; margin = (5 − 9) ÷ 5 = −0.8; 0.6 − 0.8 is negative, clamped to 0; points 0. NSFs: 4 > 3, fail; margin −0.333; points = 10 × 0.267 = 2.7. FICO: 560 ≥ 550, pass; margin 0.018; points = 15 × 0.607 = 9.1. DSCR: 1.0 < 1.1, fail; margin −0.091; 0.6 − 0.091 = 0.509, clamped to 0.5; points = 15 × 0.5 = 7.5. Positions: 3 > 2, fail; margin −0.5; points = 10 × 0.1 = 1.0. Total = 4.0 + 15.7 + 0 + 2.7 + 9.1 + 7.5 + 1.0 = 40.0; score 40.0, five criteria failed, and 40 is below the refer line of 55, so Decline. The DSCR line shows the cap in action: a narrow miss still earns half the weight, which is deliberate, since a 1.0 coverage against a 1.1 minimum is an exception a human might approve, while the negative-day line, at nearly double the limit, earns nothing.
Example 3: a three-criterion matrix approves a file with a hard fail
Inputs: Years in business, weight 40, minimum 2, value 6; DSCR, weight 40, minimum 1.25, value 1.25; Guarantor FICO, weight 20, minimum 680, value 650; approve at 70, refer at 50. Years: margin = (6 − 2) ÷ 2 = 2, clamped to 1; points = 40. DSCR: 1.25 ≥ 1.25, pass with zero margin; points = 40 × 0.6 = 24. FICO: 650 < 680, fail; margin = −30 ÷ 680 = −0.044; 0.6 − 0.044 = 0.556, clamped to 0.5; points = 20 × 0.5 = 10. Total = 40 + 24 + 10 = 74; score 74.0; one criterion failed; 74 is above the approve threshold of 70, so the decision is Approve. This is the result the fail column exists to catch. If a 680 guarantor score is a genuine requirement of the policy, a weighted matrix will not enforce it, because a strong showing elsewhere can outscore the shortfall. The fix is to treat FICO as a knockout checked before the matrix, or to raise its weight until a fail cannot be outscored, or to set the policy so that any fail caps the decision at Refer. The matrix is a ranking device; knockouts are a separate layer.