Eleven years helping teams decide what to build. This year, I started building.

• Eleven games: Egyptian Rat Screw, Slapjack, Beggar-My-Neighbour, two new games designed for this app, Hearts, Spades, Blackout, Euchre, Gin Rummy and Rummy.
• Easy, standard or hard computer players, or 2 to 4 people sharing one phone laid flat in the slap games. I’ve tested that with my own two hands, not yet with a full table.
• An AR mode deals the table onto a real surface. Placement and tracking work on my phone, but slapping in AR still has a bug I haven’t pinned down.
• Custom slap rules, saved as modes you can share as files. No account, no analytics, and it plays fully offline.

Under the hood

• The rules suite found four real bugs in the engine the day it was written. The app now has more than 2,000 automated tests.
• The core card library imports nothing, not even Foundation. Trick-taking rules are two small functions, so adding Euchre’s odd bowers needed no engine change, and Gin and Rummy share one meld engine.
• Easy computer players sometimes miss a pattern instead of just reacting slower. Self-play testing caught what every single-move test missed: a Gin bot that never declared gin, so the easy computer beat the hard one.
• No image files beyond the app icon, and even that is rendered from the same code that draws the in-game hand. The AR card textures are rendered at runtime from the same SwiftUI view the 2D table uses, and every sound is synthesized on the device.

Built with SwiftUI · RealityKit · ARKit · AVAudioEngine · Core Haptics · 2,000+ tests

• Click-lock planks and ceramic tile. Measure, mark and cut at the saw, including notches around floor vents and holes for pipes.
• Tile work start to finish: thinset, leveling, grout and sponge. On plank jobs, foam goes down first, and quarter-round and thresholds go on last.
• A 20-job career where what you earn pays for the next job’s materials and tools, plus a relaxed Zen mode.
• Draw your own room. The game lays it automatically first, to prove it can be done, before it lets you play it.
• Every step has a button, so you never have to drag. It has VoiceOver labels, a left-handed layout and Reduce Motion support, though I still owe it a full VoiceOver playthrough on a real phone.

Under the hood

• All 23 sound cues are built at launch from noise, oscillators and filters. Tests check that every cue is audible, that loops meet cleanly at the seam, and that the output is identical on every run.
• Every length is stored as a whole number of eighths of an inch, so seams and cuts come out exact, and all randomness is seeded, so a saved job resumes exactly where it left off.
• Wood grain is modeled as a tree’s growth rings sliced by a board, which is what produces real flat-sawn “cathedral” patterns.
• Seven simulated player personalities play whole jobs with no screen attached. They found an objective no player could ever meet, a way to get free money out of the shop, and a pay rate that rewarded skipping the foam and trim, all before a person did.
• A four-chapter tutorial, three difficulty levels, and a 29-article manual where most of the numbers are pulled straight from the game engine, with a test that fails if they drift.

Built with SpriteKit · SwiftUI · AVAudioEngine · Core Graphics · Core Haptics · 440+ tests

• Nested tags: tag a photo Beach and it also shows up under Summer and Travel.
• Smart albums that fill themselves from rules like untagged, a given year, or contains text.
• Tidy Up finds duplicate stacks, burst frames and old screenshots, and iOS asks before anything is deleted.

Under the hood

• Search works by tag, by what’s in the photo, by text in the photo, or by date.
• A test fails if networking or iCloud sync code shows up in the app. The only thing that ever comes in is a full-size original that lives only in iCloud, which iOS fetches for the photo viewer the same way Photos does.
• Built for a 20,000-photo library. The first scan skips a slow per-photo lookup and fills in file names later. Image analysis works from small, temporary versions of each photo, and runs on its own while the phone is charging, or right away if you tap Analyze Now.
• “Crazy slow” got measured, not guessed at. On a test library of 20,000 photos, one count on the home screen took 0.92 seconds, and the screen redid it for every photo analyzed. Caching the count brought it to 0.002 seconds.
• A crash at launch was traced from the crash log to the photo library scan, which was meant to run in the background but was running on the main thread. Tests now check that the background workers really are off the main thread.

Still to check Face groups and a Hidden album behind Face ID are newer, and still waiting on a real check on my phone.

Built with SwiftUI · SwiftData · PhotoKit · Vision · 139 tests

• One chart from a week back to 10 days ahead: recorded pressure, forecast, alert windows and logged migraines.
• Log a migraine in two taps, or add a past one by date and let the app look up the weather from history.
• First patterns show up at ten migraines. At thirty, your own alert thresholds are ready to switch on with a tap, or you get a plain “no strong pressure effect found yet”.

Under the hood

• A WeatherKit failure shows up as a banner with the real error, a rule I made after an earlier app’s weather card sat blank for weeks. On the first real install that banner showed a sign-in error, which turned out to be a setting I’d missed, and I fixed it the same day.
• WeatherKit is the only thing the app talks to, for the forecast and the Apple Weather logo Apple requires, and a test fails if any other networking code shows up. Home coordinates are rounded before they’re stored.
• Each migraine hour is compared with the same hour on each of the seven days before and after it, and a small statistical model turns that into a chance, like “pressure drops probably raise your risk (87% chance)”. In simulations at thirty migraines, my first rule told about 1 in 6 people with no real pressure effect that they had one, and the model did that 2 times in 300. Alerts keep the standard thresholds until thirty migraines, and switch over only after a tap.
• A live reading from the iPhone’s own barometer sits beside the forecast. It’s converted to sea level with the barometric formula, and on my phone it matched the forecast within 0.8 hPa.

Still to check Heads-up alerts, the Apple Health sync and the history lookup pass their tests, and I’m still checking them on my phone.

Built with SwiftUI · Swift Charts · WeatherKit · Core Motion · HealthKit · 150+ tests

• Build a quote room by room, watch the price and breakdown update as you go, then send it as a text or a one-page PDF.
• Job checklists with before and after photos, and a report PDF when the job’s done.
• Short-term rental turnovers: restock, damages, lost and found, and a summary emailed to the host.

Under the hood

• Tests keep door codes and key locations out of every shared quote and PDF.
• Photos are re-encoded on import, which strips location and camera details. A test proves it with a photo that had them.
• The quote math lives in its own package and uses decimal arithmetic, so a total never picks up a stray fraction of a cent.

Still to check It all passes its tests, and I’m still checking the photos, job reports, turnovers, email to the host and Add to Calendar on my phone.

Built with SwiftUI · SwiftData · Keychain · ImageIO · 121 tests

• Tap Start at bedtime. In the morning it shows how many minutes to rewind, and when the count is small enough to be useful, how many jump-back taps that is.
• An optional sleep timer that plays brown, pink, white or surf noise under the book, and a Shortcuts action for one-tap bedtime.
• Forgot to tap Start? Give it your best guess at when you started and it works out the night from there.

Under the hood

• The rewind math is a pure function over plain dates, and Apple Health is touched in exactly one file, so the logic is tested without a device or a night’s sleep. You count as asleep at the first stretch where the next three minutes are at least 60% asleep.
• You pick your player’s jump-back step: 10, 15, 20, 30, 45, 60 or 90 seconds. The never-longer-than-the-timer rule came from a bug that only showed up on a real phone.
• The background noise is generated in code, so the app ships no audio files, and the morning notification arrives quietly without ever showing a permission prompt.

Still to check The sleep timer, the Shortcuts action, the morning notification and the forgot-to-tap recovery pass their tests, and still need real nights on a phone to prove them.

Built with SwiftUI · HealthKit · AVFAudio · App Intents · 78 tests

• Drag to aim, let go to drop, and real physics settles every block.
• Blocks melt a little every second and vanish when they get too small, so hesitating costs you.
• Perfect drops lined up on the block below build a streak multiplier of up to 5x.

Under the hood

• Melting is one swappable theme that covers shrink rate, colors, particles and share text, so the game scene never has to know which theme it’s running.
• Swift’s built-in hash changes on every launch, so it couldn’t be trusted to give everyone the same day. The daily blocks come from a small seeded random number generator keyed to the date instead.
• There’s no debugger attached to a test build on my phone, so the app keeps its own diagnostics: a self-trimming log that also catches Objective-C exceptions before a crash, and a stall watchdog, all readable in Settings.

Still to check My last two fixes, for a freeze on Replay and a black screen after tapping Play, still need a real check on my phone.

Built with SpriteKit · SwiftUI · AVFoundation · Core Graphics · no automated tests yet

• Trainer: lessons where you feel a letter, guess it, and level up after five in a row.
• Keyer: a big tap pad that adjusts to your rhythm as you tap and decodes as you go.
• Compose: type a message and feel it played back.

Under the hood

• The Morse engine (encoding, timing and the tap decoder) lives in a pure Swift package that’s tested on the Mac, no simulator needed.
• The decoder calibrates from your own taps with a moving average instead of fixed thresholds, so it keeps up when your speed drifts mid-message.

Still to check I still need to check how the buzzes actually feel on my phone.

Built with SwiftUI · Core Haptics · Swift Testing · 36 tests