Apps

Why this app

The Big 4 mobile apps in this space (BDO Germany, Deloitte Germany among them) are content-heavy lead-gen funnels for the firm's consulting business — useful in theory, but built to redirect you into an engagement. I wanted the opposite incentive: something genuinely useful on its own, that never tries to sell you anything. Started as a calculator with a news feed bolted on; the news feed turned out to be the more interesting problem.

Inside the code

The tax logic lives entirely in lib/tax/uk/ as pure, dependency-free TypeScript — no UI code touches a rate or a threshold directly. A cluster of calculation engines covers Income Tax (a full stacked model: salary, other income, savings and dividends, with the Personal Allowance taper worked through correctly) and Take-Home Pay, Capital Gains Tax (Annual Exempt Amount, current-year vs brought-forward loss rules per HS227), Corporation Tax with marginal relief and employer NIC, SDLT, Student Loan repayment, VAT (including the Flat Rate Scheme), a Pension Planner, and a Sole Trader vs Company comparison that models retained profit as a company-only lever rather than treating "take out of the company" as the whole picture.

Above that sits a decision-tool layer — IR35 status, Self Assessment requirement, side-hustle/trading-allowance, a VAT-registration check, a short Tax Health Check quiz — each one exportable as a PDF via expo-print and expo-sharing, which is the feature that separates this from a plain calculator: a client-shareable output with the working shown, not just a verdict.

The other half is Tax Radar: the live gov.uk Content API feed, tagged by a rule-based engine (lib/updates/tagging.ts) into life-situation topics — self-employment, property, pensions, PAYE, VAT, CGT — with a relevance score that boosts high-signal formats and followed topics while down-ranking routine "page updated" churn. A personalisation profile (employee / self-employed / director / landlord / investor, multi-select) seeds followed topics automatically and filters the Deadlines list to what's relevant. No backend, no cron job: the feed is fetched live, tagged and scored entirely on-device, with a bundled seed fallback so the tab is never empty even offline or the first time gov.uk's API is slow.

What was hard

Getting the calculators right was the easy part — the rules are published, and 144 tests pin every band, taper and edge case against HMRC's own figures. The harder problem was restraint: an early "salary vs dividend" scenario tool was trivially "always pay yourself the minimum salary," so it was reframed into Sole Trader vs Company, and even that needed retained-profit modelling once it became clear that full extraction at current dividend rates no longer favours incorporating the way it used to. A "unified position" screen that let you enter your numbers once was built, then dropped — it duplicated the existing calculators without propagating anything into them, so it was just another screen pretending to be a shortcut.

The other real constraint was privacy as a design rule, not a slogan: no accounts, everything local via AsyncStorage, crash reporting (Sentry) gated to production only and limited to crash diagnostics — no analytics, no tracking identifiers. That rule killed a planned Face ID lock outright, once it became clear that nothing persisted was sensitive enough to need one.

What's next

• Curated plain-English summaries layered onto significant Radar items — authored offline, human-reviewed, no runtime LLM cost.
• A home-screen widget and deadline notifications — both need the same move off Expo Go onto a custom EAS dev build, so they're batched together.
• Scottish income tax bands (rUK-only at launch) and a standalone UK Corporation Tax calculator.
• Weighing a light freemium model (PDF export / scenario save-compare as the paid tier) against keeping the whole thing free — undecided.

Why this app

Most chess-puzzle apps serve you a randomly generated tactic stripped of context: solve it, forget it, next. Brilliancy inverts that — every position is the real finish of a game that actually happened, with the history attached. Morphy at the opera, the Immortal Game, Fischer at thirteen: solve the position and you get the story, not just a green checkmark. The first version of the brief called for a daily-puzzle-with-a-streak format; that framing was deliberately dropped mid-build (see below) in favour of a collection explored at your own pace.

Inside the code

Every puzzle is defined once, as a move list and metadata, in src/puzzles/catalogue.ts. Nothing is hand-written from there on: chess.js replays the moves to derive the FEN and verify legality, and a vitest suite checks, for every one of the 28 entries, that the position is legal, the solution line is forced, and the final move actually delivers the stated goal — mate, in every case so far. That rule (no FEN from memory) exists because the app's entire value proposition is that the positions are genuine; a single wrong square would undermine the whole premise.

The board itself is hand-drawn SVG (react-native-svg) rather than an image-tile board, tap-to-move with legal-move dots and capture rings, animated with React Native's Animated API and haptic feedback on each move. The screen flow — masthead → player cameos → prologue → solve → epilogue — is deliberately paced like reading a short piece about the game, not a puzzle-of-the-day widget. Progress (points, solved count) is stored locally via AsyncStorage; there's no account and nothing leaves the device.

What was hard

Two decisions turned out to matter more than any line of code. First, licensing: the piece-art sets suggested in the original brief (Staunty, Fresca, Cardinal, Gioco) all turned out to be CC BY-NC-SA — non-commercial — and couldn't ship. The set actually used, Chessnut, is Apache-2.0, retinted for the app's dark, candlelit palette by a small generation script (scripts/gen-pieces.mjs); every portrait is public-domain, sourced and recorded in CREDITS.md. Second, framing: the app was originally briefed as a daily puzzle with a streak, and that was deliberately removed once it became clear it would force a permanent content-publishing obligation onto a solo, unpaid hobby project — the opposite of the calm, browse-the-greats feel the app is actually going for. Home is now the next unsolved puzzle, "where you left off," not "today's puzzle."

What's next

• Drag-to-move (currently tap-to-move only) — gesture-handler and Reanimated are already wired in, it needs the worklets babel plugin and a root gesture handler.
• Growing the verified set from 28 towards the full 150-game backlog.
• VoiceOver labels on the board ("knight on d7") — reduced-motion is already handled, full screen-reader support isn't yet.
• An Android build — the package identifier is already configured, just not built or tested.