The game’s tagline is “The Greatest Prize Wheel Ever Made.” I wrote that. I am not the most reliable narrator here.
Winning Bigly is a prize wheel you play in the browser, with six segments: BIGLY, HUGE, X2, CHAD, TAXED, and SAD! The theme, to put it politely, owes a certain debt to the current political moment. BIGLY pays out 150,000 points and three free spins. TAXED removes fifty percent of your bank. SAD! wipes it entirely. There is a Tax Collector mechanic that triggers when your multiplier hits a hundred percent, because the Tax Collector always gets there eventually.
Why This Exists
Earlier in the year I built the BaudyPositive blog using Claude Code from scratch. It is a site powered by AI that writes, reviews and iterates on its own content. That project was about understanding what coding assistant development actually looks like in practice, end to end. Winning Bigly was the next question: what happens when you start with a different tool, hand the first draft to Vercel’s v0, and then switch to Claude Code to take it somewhere more interesting?
The answer, it turns out, is a prize wheel with a tax mechanic and adverts in the game. This was not entirely my idea. More on that shortly.
What v0 Actually Gives You
v0 is Vercel’s AI UI builder. You describe what you want, it writes React and pushes commits directly to your repository. The first working version of the wheel existed within a few sessions. It had physics, a scoring system, win streaks and a bank. That is genuinely fast. The 3D metallic effect on the wheel segments, the glowing indicator, the serif font on the title: all of those emerged from v0 looking like design decisions rather than accidents.
What v0 also gives you, whether you asked for it or not, is the full Radix UI component suite. All of it. The accordion. The alert dialog. The aspect ratio wrapper. The hover card. The command palette. I needed a prize wheel and a leaderboard; I received twenty-nine Radix packages and a command palette I will never use.
The tech debt here is cosmetic. The packages sit in package.json, unused, taking up space and looking faintly judgemental. They don’t ship in the bundle unless you import them, so the actual damage is minimal. It just looks like someone who packed for a weekend trip using a suitcase designed for a three-month expedition.
What v0 is less good at is coherence across iterations. Each session shuffled the architecture slightly, moving state around in ways that weren’t wrong exactly but that made the codebase harder to follow on subsequent visits. After four or five sessions, the component tree had the structural confidence of a building renovated by seven different contractors who never met each other.
Handing Over to Claude Code
At that point I switched to Claude Code, which works differently. Rather than prompting a UI builder and merging its output, you are working directly in your own codebase with an assistant that can read the whole thing. The difference is noticeable. The suggestions are contextual rather than generative; it fixes what’s there rather than proposing a new version of it.
The first thing Claude Code helped with was the spin physics. The initial v0 version stopped abruptly. The wheel needed power4.out easing: fast at the start, decelerating as it approached the landing segment, with a tick sound that sped up during the spin and slowed to a stop. Getting that right required iterating on the timing curve, the tick intervals, and the relationship between spin duration and final angle. This is tedious to debug manually and reasonably fast to describe to an assistant that can actually read the existing code and understand why the previous attempt felt wrong.
From there: win streaks, the multiplier system, the Tax Collector trigger, the Vault mechanic. Each feature followed the same pattern. Describe the intended behaviour, review what came back, test it, adjust the description, repeat. The iteration cycle was faster than writing it myself and faster than v0, which kept wanting to redesign the whole component rather than extend it.
The Adverts
After a first round of playtesting with a small group, one playtester’s partner asked whether you could watch an advert to earn more spins.
This is not a direction I had considered. It is also, on reflection, completely correct for a game called Winning Bigly. The game is already a parody of a certain kind of maximalist game design — points everywhere, tax the losers, keep spinning. Adverts that reward you for watching them fit so naturally into the theme that not including them would have been the stranger choice.
So the adverts went in. They are simulated rather than real: a placeholder modal with a countdown timer. The mechanic is there. Watch the ad, earn extra spins or a multiplier boost. The Tax Collector still fires. It is not inconsistent.
The playtesting feedback loop, by the way, is one of the things that works well with coding assistant development. The gap between “here is a suggestion” and “here is the feature” is short enough that you can act on feedback before the person who gave it has forgotten they said it.
The Stack That Actually Matters
Next.js 16 / React 19. App router, server components, the standard arrangement.
Upstash Redis. The leaderboard has to persist somewhere across Vercel’s serverless instances. Upstash gives you Redis without the infrastructure. The leaderboard API route reads and writes to it; the client fetches the top three scores on the homepage and the full board at /leaderboard. It works without thinking about it, which is exactly what you want from your persistence layer in a side project.
Framer Motion. The spin animation. Worth every kilobyte.
canvas-confetti and html2canvas. Gold coins explode across the screen when you land on BIGLY. The Brag button captures the screen to a PNG and passes it to the native share sheet. These are the two most frivolous libraries in the project. They are also the reason anyone would open it a second time.
next-auth. Sign-in is optional and does not currently unlock anything beyond your name appearing on the leaderboard. An odd design choice I am aware of but have not yet resolved.
RESULT
It’s live at winningbigly.baudypositive.com. The leaderboard works. The confetti works. The tax collector fires at the worst possible moment, as intended. The Brag button produces a screenshot that is roughly sixty percent of what I hoped it would be, which is about average for html2canvas.
As a learning exercise it did what it was supposed to do. v0 is excellent for getting a thing on screen quickly. Claude Code is better for making a thing that already exists into something coherent. Start messy, refine with context. That combination is probably the right way to use both.
The twenty-nine unused Radix packages are still there. I will get to them.
The Greatest Prize Wheel Ever Made. Probably.