NomiKit: Tiny Things You Build

Inspiration

Gunpla runners, blind-box toys, and fidget toys. The joy is in the fingers: the snip of a part coming off the sprue, the click when it seats.
Day-one rule: opening a kit should feel like opening a beautifully designed little present, not pulling a slot machine.
Shipaton was the deadline that turned a RealityKit weekend experiment (first commit July 22) into a shipped app.

Unbox. Tap the carton, the flaps fold back, pull out the collector card and the runner.
Snip, Snap, Collect Sweep a finger along the sprue and every gate it touches pops loose. Haptics tuned like a fidget toy. Drag each part into its ghost outline. A tick when it aligns. Finished toys line up on your shelf!
A CAT runs the one-minute tutorial and heckles every unboxing. ("Could be chrome. Could be gold. Will be loved.")

Stack

Swift, SwiftUI, RealityKit, custom Metal shaders (iridescent plastic, crease-edge wireframe ghost via LowLevelMesh), Core Haptics, AVFoundation.
RevenueCat SDK (StoreKit 2, anonymous users): three coin packs + Nomi Pro monthly, one Nomi Pro entitlement, customerInfoStream feeding game state.
RevenueCat Ads: AdMob rewarded through the RevenueCatAdMob adapter, so ad revenue lands in the same RevenueCat charts as purchases.

Content pipeline (one person, 45 3D models, four weeks)

Animal → SD mecha concept sheets (image model) → Meshy for 1,250-face A-pose models and textures.
My own local web tool, GLB Lab: Parts Studio splits each model into parts, a sprue generator lays out the runner and gates. Blender CLI turns GLB into USDZ.
Cards and cartons are printed at runtime in Core Graphics per series. The foil pass is the metal map, so the gold rule actually reflects the room.

Game logic

One pure reducer: reduce(state, action, moment, config, rng). 100+ self-checks run at every debug launch.
The economy was tuned in a Python Monte-Carlo before anyone played it: a free player finishes all 45 in about 38 days, everything in about a year; Pro is about 1.4× faster.
[optional] Pair-programmed with Claude Code. I owned art direction, economy, consoles, and taste; Claude wrote most of the Swift.

App Review 2.1: "unable to locate the ATT permission request." ATT lived inside Google UMP's console-scripted flow, which the reviewer's device never showed. Fix: request ATT directly on the first "Watch AD" tap, then UMP. Resubmitted with a device recording.
RealityKit in the simulator: scenes stop presenting unless some material value changes every frame, and transform writes on the root entity are ignored. Fix: a hidden light under the shelf that "breathes" by one millimetre.
AdMob's post-launch ramp: limited fill until the store listing links, the readiness review passes, and app-ads.txt is crawled. Which brings us to the dollar.

Shipped. First commit July 22, App Store August 21, 41 commits, solo.
The RevenueCat Ads chart reads $1.00 lifetime. My first ad dollar, and it flowed through the same pipe as the purchases. The pipe works.
5.0 stars from the first 3 ratings!

Consent UX is game design, not a legal checkbox. Ask on the first "Watch AD" tap and nobody feels ambushed.
RevenueCat's ad adapter makes ads a first-class revenue stream: one placement id, one dashboard, LTV that includes ad revenue, no backend.
Simulate the economy before shipping it to people. The simulator found a design bug no playtest would have.

Growth: every build is a 30-second satisfying clip. One kit, one video, find the one that travels. App Store featuring nomination.
Coins → RevenueCat Virtual Currency with server-verified rewarded ads via RevenueCat's AdMob SSV webhook, once the online version lands.
More series. The pipeline makes a nine-toy volume in an afternoon.