0.4.0: spending insights, safety net follow-through, payment method, stale-rate transparency, tests

[kenrinzero]

0.4.0 — finishing the post-audit roadmap

Implements every kept item from the improvement audit, landed as one release. Branches off main after PR #13 (safety net) was merged; bumps 0.3.0 → 0.4.0.

Data insight (mining the renewal_log + categories)

• by_category totals mode (t) — per-category monthly-equivalent subtotals in base, largest first (top-5 + "+N").
• forecast totals mode (t) — concrete, non-amortized 30/60/90-day cash-out (walks each active sub's real billing cycles; cumulative and horizon-inclusive), so clustered annual renewals stop hiding behind the amortized average.
• Month-grouped history (h) — per-month subtotal headers; the current month's header also shows your amortized estimate, so actual-so-far reads against what you committed to.

Features

• Duplicate a subscription (D / Shift+D) — clones the highlighted sub into a prefilled modal (lowercase d stays Delete).
• payment_method field — optional funding source ("which card is this on?"): additive column + migration, full CRUD/CSV/JSON round-trip, a modal field, shown on the row, and matched by the / filter.
• Budget warning — when a monthly income is set and amortized commitment exceeds it, the notice board shows a calm ▲ BASE x over budget (n%) line under the banner, regardless of the active totals mode.

Robustness

• Stale-rate transparency — currency.get_rate_status returns (rate, fresh|stale|missing). A failed fetch now reuses the last-known rate of any age (db.get_last_rate) instead of a fake 1:1, and the title bar distinguishes ⌛ stale (old-but-real) from ⚠ rate (never cached). A genuine 1.0 fetched today reads as fresh, not flagged.

Code health & tests

• gakkari/totals.py — the money/sort math lifted out of the MainScreen god-object into pure, explicitly-parameterized functions; MainScreen delegates.
• GAKKARI_DB env seam in get_conn() (read at call time) so tests/tooling can point at a temp DB.
• First test suite — tests/ with 57 pytest tests (models date/tax/factor math, totals incl. forecast/by-category, io validation + round-trip, db round-trips + backup rotation, EN/JA i18n key-parity) + .github/workflows/ci.yml running them plus a non-UTF-8-locale CLI smoke (which guards the --help/--notice em-dash fix) on Python 3.12 / 3.13.
• Polish — UTF-8 stream guard moved to the top of __main__.main() (fixes --help/argparse mojibake under a legacy codepage); dead SCREENS dict + redundant CSS_PATH removed from app.py.

Docs

CLAUDE.md (data model, totals cycle, rate-freshness, project layout, a 0.4.0 deltas section) and README (features + credits) refreshed.

Verification

• 57 tests pass; py_compile + import smoke clean; --help and --notice exit 0 under LC_ALL=C.
• An adversarial multi-agent review of the full diff (UI wiring, history/notice signature changes, data layer + SQL placeholder counts, totals parity) returned zero correctness findings — the one low-severity note (export default path ignoring the new env seam) is fixed in this branch.

🤖 Generated with Claude Code