PokerStats

A native iOS app for live poker players: three-second hand logging at the table, HUD-style stats computed from your own history, and a leak finder — every formula verified against worked examples.

Problem

Online players get VPIP and c-bet stats for free from HUD software. Live players get a notebook — because no one logs hands at a real table unless it's nearly instant.

Built

A Swift 6 / SwiftUI / SwiftData app: crash-proof session tracking, a progressive-disclosure hand logger capped at five taps, eight stats computed by pure functions, and a leak finder that compares your numbers to reference ranges.

Proof

121 tests across 9 suites, including worked examples that trace full hand histories through every stat formula against hand-computed answers.

Context

Serious online players make decisions with HUD statistics — how often an opponent enters pots, folds to three-bets, gives up on the river. Live players generate the same information every session and lose all of it, because logging a hand mid-game has to compete with the game itself.

PokerStats bets the whole product on one constraint: if entering a hand takes more than a few seconds, nobody does it twice. Everything else — the stats engine, the leak finder, the cloud backup — only matters if that holds.

What I built

The hand logger uses progressive disclosure: a preflop fold is one tap, and the longest possible path through any hand is five — a rule stated in the README and enforced in contributing guidelines. Sessions track buy-ins, rebuys, and P&L with a timer derived from a persisted start date, so a killed app resumes mid-session without losing time. Live Activities put the session on the Dynamic Island; widgets put bankroll on the home screen.

From logged hands, a pure-function StatCalculator computes eight statistics — VPIP, PFR, fold-to-3-bet, c-bet%, WTSD%, W$SD%, hourly rate, and ROI — returning nil rather than a fabricated number when the denominator is zero. A leak finder compares your stats to reference ranges and turns deviations into coaching notes. Sync is offline-first: SwiftData locally, Supabase backup behind Sign in with Apple, row-level security on every table, and a dedicated edge function that deletes storage, metadata, profile, and auth user on account deletion. The app target has zero third-party dependencies, and Fastlane automates TestFlight, release, and an App Store screenshot pipeline across three device sizes with seeded data.

Decisions

• Five taps or it doesn't ship

  The entry-speed cap is the product thesis, written into the contributing rules so a future feature can't quietly add a sixth tap. Preflop fold — the most common hand — is one.

• Stats are pure functions with known-answer tests

  The StatCalculator has no state and no side effects, and three worked examples trace full hand histories through every formula against answers computed by hand. A stats app that's wrong is worse than a notebook.

• Return nil, never a guess

  Zero hands played means VPIP is nil, not 0%. Every formula has explicit zero-denominator tests, because a confident wrong number in a coaching tool compounds.

• Offline first, cloud as backup

  Poker rooms have bad reception. SwiftData owns the source of truth; Supabase holds an encrypted backup with per-user row-level security and a GDPR-style deletion path.

• Release engineering from day one

  Fastlane lanes for TestFlight and release, Xcode Cloud CI, and a screenshot pipeline that boots three simulators with seeded data. Shipping infrastructure isn't an afterthought you bolt on at v1.0.

What this proves

No AI in this one — that's the point. The verification discipline transfers: known-answer tests on every formula, nil over fabricated numbers, and release automation built before launch, not after.