Build Story

FatTail Markets was built by a single developer using Claude Code as the core building partner. Every phase — from architecture to deployment — was AI-assisted.

The Builder

Suhong Sung — Solo builder + DeFi YouTube creator (Corini, 21K subscribers).

Years of analyzing DeFi protocols, AMM mechanics, and yield structures on YouTube led to the core insight: 90:10 extreme probability markets can create a natural two-sided market — stable carry for the 90% side and explosive convexity for the tail. FatTail is the product of that domain expertise combined with AI-assisted full-stack development.

YouTube | X/Twitter | GitHub

By the Numbers

Metric	Value
Phases completed	15+
Unit tests	99
API routes	45+
Chains supported	5
Time to MVP	Weeks, not months

Phase Timeline

Phase 1: Foundation

LS-LMSR math engine in pure TypeScript (packages/ls-lmsr)
Cost function, pricing, liquidity, vigorish, safety guard, Kelly criterion
69 unit tests passing from day one
Log-sum-exp trick for numerical stability with large values

Phase 2: DB & Auth

Supabase (PostgreSQL) with Row-Level Security
Profiles with auto-creation trigger + 10,000 K-Point welcome bonus
Email/password auth via Supabase Auth

Phase 3–4: Trading & Markets

LS-LMSR automated market maker with buy/sell
Dynamic vigorish: probability-weighted fees (0.3% for 90/10, 3% for 50/50)
Market creation with admin + optimistic oracle resolution
Safety guard with 4-tier health system (informational only, never blocks trades)

Phase 5–8: Oracle & Resolution

Three oracle types: Admin, Price Feed (CoinGecko), Optimistic (UMA-style)
Optimistic oracle flow: propose → challenge → vote → resolve
Bond system (500 KP), KP-weighted voting for Sybil resistance
Resolution criteria with yes/no conditions, evidence rules, void conditions

Phase 9–11: UX & Wallet

Dual persona UI: Tail mode (speculator) vs Yield mode (market maker)
Mobile responsive optimization
Solana wallet integration (Phantom, Solflare)
K-Point ↔ on-chain bridge

Phase 12: Multi-Chain Expansion

5 chains: Solana, Base, Arbitrum, Optimism, Polygon
wagmi v3 + viem v2 for EVM chains
Chain-agnostic deposit/withdraw with per-chain verification
MoonPay fiat on-ramp across all chains

Phase 13: Telegram Bot

grammY framework with Vercel webhook
Commands: /balance, /positions, /markets, /alert
Push notifications on trades, settlements, vig earnings
Price alert cron with 1-hour cooldown

Phase 14: Bot API

REST API with API key authentication (HMAC-SHA256)
Endpoints: markets, trade, portfolio, positions
Python and TypeScript SDK stubs
Rate limiting and usage tracking

Phase 15: Landing & Docs

Astro 5 landing page (0KB JS, i18n EN/KO, SEO)
Starlight documentation site (14 pages, Pagefind search)
Multi-subdomain deployment (app/www/docs on Vercel + Cloudflare)

AI Development Philosophy

Claude Code was not used as a simple autocomplete tool. It served as:

Architect: Designed the LS-LMSR engine, oracle state machine, and multi-chain abstraction layer
Implementer: Wrote production code across Next.js, Supabase RPC, Solana programs, and EVM integrations
Debugger: Diagnosed issues like anchor.Wallet ESM export failures, Supabase generic type limitations, and Turbopack polyfill conflicts
Reviewer: Caught security issues (RLS gaps, missing input validation) and suggested improvements
Documenter: Generated migration files, API docs, i18n keys, and this documentation site

Every phase was planned as a structured session with clear goals, implemented incrementally, tested, and documented before moving to the next phase. The result: a production-grade prediction market platform built at startup speed by one person.

Why Extreme-Only? The Structural Moat

The technology is replicable — the moat is in the network and positioning:

Creator-Builder Fusion: A DeFi YouTuber with 21K subscribers building the product directly. The trust and distribution channel are not replicable by adding a feature.
Two-sided Liquidity Network: Once yield seeker ↔ tail hunter liquidity forms, late entrants face a cold start problem on both sides simultaneously.
Bot API Ecosystem: Traders who build strategies on FatTail’s SDK have switching costs.
Category Ownership: “Extreme probability = FatTail” — category creators have first-mover advantage.

Yield Seeker APR

Yield seekers (90% side) earn from vig distribution. Mathematical example at TVL $10M, 10% daily turnover:

Daily Volume: $1M
Daily Vig Pool: $1M × 0.6% = $6,000
Yield Seeker Share (50%): $3,000/day
90% Side Pool (~70% of TVL): $7M
$10K Position → 0.143% share → $4.29/day → ~$1,566/year → APR ~15.7%

Turnover Rate	APR
5% (early stage)	~7.8%
10% (growth)	~15.7%
20% (with bot traders)	~31.4%

Real APR is ~70-80% of above after accounting for market expiry and reinvestment cycles. Auto-compound feature is already implemented to minimize reinvestment friction.

Tech Stack Decisions

Choice	Reason
Next.js 16 (App Router)	Server components + API routes in one framework
Supabase	PostgreSQL + Auth + RLS + RPC in one service
LS-LMSR (custom)	No existing library for liquidity-sensitive LMSR
pnpm workspace	Monorepo for shared math library
Astro 5 (landing)	Zero-JS static site for maximum performance
Starlight (docs)	Built-in i18n + search + Astro ecosystem
wagmi + viem	Type-safe EVM interactions
grammY	Lightweight Telegram bot framework