[j-bryan]

Related to #38 (does not close it). Closes #50. Stacked on #46, in stack #51.

Debt, construction, depreciation, and PTC schedules now date each period from the schedule start using add_periods, instead of stepping from the previous date, and each one takes end_of_month (keyword-only, default True).

• Amortization: AmortizationSchedule.builder / .build, amortize, and AmortizationBuilder take end_of_month. Payment dates and interest_free(from_date=..., to_date=...) both use the anchored dates. When payments are spread over calendar periods, the final payment covers the period up to the next anchored boundary.
• Construction: ConstructionSpendConfig.end_of_month, also accepted by ConstructionSpendBuilder and construction_spend_schedule. Interest servicing periods are anchored at the construction start. The implicit escalation uses the same flag; a user-supplied escalation_policy keeps its own setting.
• Depreciation: macrs_schedule and both vdb_schedule date loops (the "none" convention, and the half-year/mid-quarter conventions without schedule_dates) use anchored dates. A Feb 29 placed-in-service date no longer raises in macrs_schedule.
• PTC: the eligibility end is the anchored anniversary, so 3 years from Feb 28, 2029 ends Feb 29, 2032 with the flag on. incentives.py no longer imports dateutil.
• Tests: tests/unit/test_finance_tax_anchoring.py checks every schedule's dates against the anchored boundaries, plus worked examples for Apr 30 starts and February placements under both settings. Here the boundaries come from a temporary anchored_boundary test helper; Property tests use independent references instead of reimplemented calendar oracles #52 switches these tests to add_periods (tested against hand values in Anchored schedule boundaries with an end-of-month rule #45) and removes the helper.

MACRS table correction

A separate commit, a7dd644, corrects two columns of the mid-quarter tables to match IRS Publication 946. In 15-year, Q4, years 9–15 had the 0.0590/0.0591 alternation shifted by one position, and the column summed to 1.0001. In 20-year, Q4, year 18 was 0.04458 instead of 0.04459, and the column summed to 0.99999. Every table now sums to exactly 1.0. This changes depreciation amounts for Q4 mid-quarter placements of 15- and 20-year property. It was found by the MACRS property test and isn't related to anchoring.

Behavior changes

With end_of_month=True as the default:

• Schedules starting on the 29th–31st no longer drift after a short month: amortization payments, VDB dates, and construction interest periods all return to the start's day (or month-end) after February.
• Schedules starting on the last day of a 30-day month or of February follow month-ends: Apr 30 monthly gives May 31, Jun 30, Jul 31. Examples: monthly construction interest from Apr 30 is booked May 31/Jun 30/Jul 31, not May 30/Jun 30/Jul 30 plus a one-day stub. A MACRS or PTC schedule from Feb 28 of a year before a leap year reaches Feb 29.
• A Feb 29 MACRS placement returns a schedule dated Feb 28 in non-leap years and Feb 29 in leap years.

Not in this PR

• PTC escalation: ptc(end_of_month=...) sets only the eligibility end. Its implicit rate escalation keeps the default, because that helper (_generation_escalation) lives in generation.py, which Property-based tests for CashFlowStream #41–Consistent capacity factor validation #44 also change. That work moves to PR 4, as noted in the plan on Recurring schedules drift after a month-end or leap-day start #38.
• Callers inside EnergyProject: the debt compiler, depreciation, PTC, and construction sections use the default. Passing a project-wide setting through them is PR 5.
• time_delta_per_period and dateutil: still used by the builder's scheduling and contract settlements, until PR 7.
• Construction interest accrual (Construction debt draws don't accrue interest until the next servicing boundary #48): draws made between servicing boundaries still don't accrue until the next boundary. This PR moves the boundaries but doesn't change how accrual works within a period.

Testing

uv run ruff check, uv run mypy dcaf/, and uv run pytest pass (961 tests), and the Hypothesis properties also pass at 3,000 examples each. No existing test changed result. ruff format --check flags one unwrapped parametrize entry in tests/unit/test_finance_tax_anchoring.py; #52 reformats it.

🤖 Generated with Claude Code

[dylanjm]

Changes are good, basically adds end_of_month bool to relevant methods and fixes MACRS tables.