Money questions
- Who paid for whom, splits, and contributions
- For a person, both people, or someone else
- Several categories on one entry, such as online + personal
A private expense tracker for two people. The problem was keeping shared money honest — who paid, who it was for, splits, accounts, and necessity — not adding more features. This case study covers the problem, the stabilisation-first build, and where it stands now.
Problem · solution · current status — screen archive below.
Scope
Private app • Two people
Coverage
23 screens • 06 areas
Status
Web stabilisation • Android next
Expense Tracker is a private app for two people. The hard part was not adding more screens; it was keeping shared spending honest: who paid, who it was for, 50/50 splits, contributions, accounts, categories, and necessity versus want. The spreadsheet was a legacy system of record, and the web app needed correctness, clarity, and protections before anything else.
The web app is built with Next.js 16 App Router, React 19, TypeScript, motion, hand-written CSS tokens, and Vitest. It deploys as a Cloudflare Worker through @opennextjs/cloudflare, with Cloudflare Access in front. Google Sheets is the only backend: reads and writes go through /api/v1, values are written RAW and read UNFORMATTED_VALUE, and auth is a service-account JWT signed with Web Crypto so it runs on Workers unchanged.
Without credentials it runs on a bundled synthetic household and saves in the browser. With credentials, the spreadsheet is the truth and writes go through the API behind one optimistic store.
The JSON backup is versioned with schemaVersion, kind, exportedAt, and source. The native import reads the same shape, so a backup taken here becomes the migration path later.
CI checks types, tests, build, and secrets on every pull request. A release checklist and golden domain fixtures keep the money rules consistent. Gestures are additions, never replacements for reachable controls.
Current discipline: the web app receives correctness, clarity, and data protections — no new screens, no new features, no redesign. The long-term target is a native Android app; the spreadsheet will become an import path and read-only backup, not its backend.
The web app is in stabilisation, not feature development. The core ledger, accounts, splits, filters, backup, API, CI, and Cloudflare deployment are done; the next major step is the native Android app.
No new screens or redesign while the web app stabilises.
Correctness, clarity, and protections that keep the data honest.
Dashboard, day-grouped ledger, month navigation, person switcher, accounts, who-paid-for-whom, categories, necessity, activity filters, add/delete, Sheets API, backup, CI, and deploy.
Unit and golden tests cover the domain, API, and backup contract.
Phase 1 is built; CI produces a sideloadable APK. Install check is pending.
Not started; kept in the future-plan material.
Filter by area, then open any screen to inspect it full-size. Full-page captures scroll inside the viewer.
The Web Design page collects every case study — briefs, decisions, and deliverables.