wegamekinglc / Derivatives-Algorithms-Lib / 27506602444

80%

master: 79%

Build Type

Pull #106

github

Committed by wegamekinglc

Commit Message

Add DF-bump fallback tests to analytic Jacobian suite

The analytic DRateDDiscount overrides never return empty in the existing
analytic-vs-FD and solve-equivalence tests, so the per-instrument DF-bump
fallback (FillRowByDFBump) inside YieldCurveCalibrationFunc_::Gradient
was uncovered -- a dark path that the production code reaches whenever an
instrument's TradeDateIsAnchor() check fails or a future Rate_ subclass
opts out of the analytic derivative.

Close the coverage gap with two tests that force every row to fall back
(swap tradeDate_ = T+2 spot != anchor, so SwapRate_::DRateDDiscount
returns empty):

- TestDFBumpFallbackMatchesCentralDifference: the fallback row must
  still be a correct column-by-column finite-difference sensitivity of
  F(x) w.r.t. each free node's logDF, checked against an independent
  two-sided central difference at step 1e-6. Tolerance 1e-6 (relative),
  justified because FillRowByDFBump itself uses a one-sided 1e-7 step.
- TestDFBumpFallbackSolveMatchesBumped: the fallback must produce a
  solvable calibration -- BUMPED and ANALYTIC_LOG_DISCOUNT (with every
  row falling back) must converge to the same node log-DFs at 1e-8.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Pull Request Pull Request #106: feat: analytic LOG_DISCOUNT curve Jacobian (CP1)

Coverage Stats

349 of 359 new or added lines in 5 files covered. (97.21%)

20 existing lines in 4 files now uncovered.

5800 of 7247 relevant lines covered (80.03%)

3446313.07 hits per line