fix(lib): CLI-side XIRR uses Excel's 365-day basis (was 365.25)

lib/irr.mjs

@@ -156,8 +156,10 @@ export function computeXIRR(cashFlows, guess = 0.10, maxIter = 1000, tol = 1e-8)
   const sorted = [...cashFlows].sort((a, b) => a.date - b.date);
   const d0 = sorted[0].date;
 
-  // Convert dates to year fractions from first date
-  const yearFracs = sorted.map(cf => (cf.date - d0) / (365.25 * 24 * 60 * 60 * 1000));
+  // Convert dates to year fractions from first date.
+  // Excel XIRR uses a 365-day basis (matching the engine-side computeXIRR) —
+  // a 365.25 basis drifts every solved rate by ~1e-5 per leap span.
+  const yearFracs = sorted.map(cf => (cf.date - d0) / (365 * 24 * 60 * 60 * 1000));
   const amounts = sorted.map(cf => cf.amount);
 
   function xirrNPV(rate) {

tests/lib/test-lib.mjs

@@ -46,6 +46,19 @@ console.log('Testing: lib/irr.mjs');
     { date: new Date('2021-01-01'), amount: 1100 },
   ]);
   near(xirr, 0.10, 2e-3, 'XIRR of -1000 → +1100 one year later ≈ 10%');
+
+  // Excel XIRR uses a 365-day basis (not 365.25) — discriminate on a leap span:
+  // -100 on 2020-01-01 → +110 on 2021-01-01 spans 366 days (leap 2020), so
+  // Excel solves (1+r)^(366/365) = 1.1 → r = 1.1^(365/366) − 1. The old
+  // 365.25 basis lands ~3.6e-5 away, well outside the 1e-6 gate.
+  const leapXirr = computeXIRR([
+    { date: new Date('2020-01-01'), amount: -100 },
+    { date: new Date('2021-01-01'), amount: 110 },
+  ]);
+  const xirr365 = Math.pow(1.1, 365 / 366) - 1;
+  const xirr36525 = Math.pow(1.1, 365.25 / 366) - 1;
+  assert(Math.abs(xirr365 - xirr36525) > 1e-5, 'negative control: 365 vs 365.25 basis differ measurably');
+  near(leapXirr, xirr365, 1e-6, "XIRR on a 366-day leap span uses Excel's 365-day basis");
 }
 
 // ── Waterfall ────────────────────────────────────────────────────────────────