Per-leg deleveraging for split-allocation books (D-028)

[HouseofTyrell]

What & why

Split allocation and deleveraging were mutually exclusive (D-016/D-026): with both enabled, the deleverage plan was dropped and the UI showed a "plan ignored" warning. You wanted to run a deleveraging scenario on a split book, so this builds the feature out.

Per the decision logged as D-028 (owner-approved scope: per-leg plans, engine depth = full combined branch with per-leg embedded-gain pools), each split leg — core and overlay — can now carry and glide its own deleverage plan independently.

Engine

• SplitAllocation gains optional coreDeleverage / overlayDeleverage plans. The top-level deleveragePlan stays single-strategy-only (still ignored when split is on).
• deleverage.ts: extracted resolveLegDeleveragePlan(plan, source) and buildDeleverageSchedule(resolvedPlan, …) so a leg can be scheduled from its own strategy. The single-strategy path is bit-identical.
• core.ts: new combined split+deleverage branch that collateral-weight-blends each leg's per-year rate / financing / extension, and runs a separate embedded-gain pool per leg. The book's pre-existing gain is assigned to the overlay leg first (the appreciated-stock sleeve), capped at its market value, remainder to core — so a concentrated overlay's unwind cost is attributed to the right sleeve. Unwind gains stay endogenous (netted with strategy flows, charged against taxSavings, per D-016/D-017). Per-leg financing follows each leg's glide.

Surface area

• ResultsTable (both orientations) and Excel export light up in split mode with no schema change on the aggregate columns — data-driven on extensionFraction < 1, and the book-level outputs are now the per-leg sums.
• Per-leg breakout columns (second commit): YearResult.coreDeleverageGain / overlayDeleverageGain (defined only in split mode, summing to deleverageGainRealized) surface as conditional "Core Unwind" / "Overlay Unwind" columns in both table orientations + Excel, each with a popupContent.ts entry. Also fixed a pre-existing latent bug found here: the table "Start" row emitted 3 placeholder cells for the 5-column expanded Deleverage group (now 5), which rendered ragged when expanded.
• CSV round-trips splitAllocation.coreDeleverage.* / .overlayDeleverage.* via a shared serializer/parser.
• Workspace rail: in split mode the single plan editor becomes two per-leg target selectors; the two "plan ignored" warning chips are removed (a softer note remains only when a stray top-level plan is set with split on).
• Sensitivity grid unchanged (still excluded, D-013/D-016 — cells stay comparable).

Tests & verification

New engine + CSV tests: split equivalence (two identical legs == single-strategy deleverage), overlay-only collateral-weighted extension, independent per-leg schedules, per-leg pools (overlay basis raises the overlay's unwind gain), endogenous-netting invariant, financing-glide, per-leg CSV round-trip, and per-leg attribution (core+overlay sum to the book gain; undefined in single mode). Updated the WorkspaceTab test to assert the rail offers per-leg editors instead of the conflict warning.

• npm run test:run — 419 passed (was 408)
• npm run build — passes (tsc + vite)
• npm run lint — 0 errors
• Browser-verified (Playwright): per-leg rail editors render in split mode; enabling the overlay plan moved the headline $2,686,020 → $1,993,074; the expanded Year-by-Year Deleverage group shows Core Unwind $418,966 + Overlay Unwind $586,552 = $1,005,518 ≈ Unwind Gain $1,005,517; no console errors.

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying enhanceddirectindexingcalc with    Cloudflare Pages

Latest commit:	cb1887f
Status:	✅  Deploy successful!
Preview URL:	https://0dd19ac0.enhanceddirectindexingcalc.pages.dev
Branch Preview URL:	https://claude-indexing-calculator-d.enhanceddirectindexingcalc.pages.dev

View logs