The Signal

Sterling Crispin shipped a Polymarket bot with a single thesis : humans systematically overestimate the probability that dramatic things will happen. The bot — called nothing-ever-happens — does one thing. It sc ans non-sports prediction markets and buys the 'No' side. No ML model. No sentiment analysis. No complex signal pipeline. Just a hard-coded contrarian bet against collective human anxiety . It hit the Hacker News front page with 143 points, which means the strategy is interesting enough that smart people stopped scroll ing. The full source is public on GitHub.

Builder's Take

This is a first -principles exploit, not a trading system. Here's the actual thesis:

  • Prediction markets price attention, not probability. A market asking "Will X politician be arrested this month?" gets created because Twitter is on fire about it — not because the base rate justifies it.
  • Non-sports markets skew toward dramatic, low-base-rate events . Wars, assassinations, regulatory collapses, tech company failures. These are interesting to bet on, which means they attract recreational bettors who overprice them.
  • Sports markets have sharps . They have historical data, closing line value, and professional arbit rageurs. Non-sports political/macro markets are softer.

The leverage calculation is blunt: if "No" is mi spriced at 60 cents when the true probability is 80%, you have a 33% edge on every dollar deployed . Compound that across dozens of markets and you don't need a PhD — you need a cron job.

What this destroys: the moat of complexity . Every solo builder assumes they need a fancy model to compete in algo trading. This bot has zero ML. Its edge is a behavioral insight encoded in three lines of logic . That's Naval's leverage principle applied ruthlessly — write the code once, let it run forever.

What this creates: a template for opinion arbitrage bots. Any market where human emotion infl ates one side of a binary is exploitable with this pattern. You don't need Polymarket specifically. You need a market , an API, and a thesis about human bias.

Tools & Stack

The Core Stack (from the repo)

  • Polymarket API — their CLOB (Central Limit Order Book) has a REST API and Python SDK. Check docs.polymarket.com for current access requirements.
  • Python — the repo uses Python for market scanning and order placement . Standard requests + web3 libraries.
  • Polygon blockchain — Polymarket settles on Polygon. You need a wallet funded with US DC. Gas fees on Polygon are negligible (fractions of a cent per tx).

Replicating This Yourself

#  Clone and inspect the core logic
git clone https://github.com/sterlingcrispin/nothing -ever-happens
cd nothing-ever-happens

# Install deps
pip install -r requirements.txt

# You 'll need: a Polygon wallet private key + USDC
# Set env vars before  running
export PRIVATE_KEY=your_wallet_key
export RPC_URL=https://polygon-rpc.com

Alternatives & Extensions

  • Manifold Markets — play- money prediction market, zero financial risk. Perfect for backtesting this exact strategy before going live. Free API, no wallet needed.
  • Metaculus API — question-based forec asting platform with a public API. No direct betting, but great for training a model on base rates vs. crowd estimates.
  • Kalshi — US-regulated prediction market. Has an official REST API. More conservative market selection but legally cleaner for US builders.
  • LLM filter layer — you could add a cheap GPT-4o-mini or Claude Haiku call to classify markets as " hype-driven" vs. "data-driven" before placing bets. At roughly $0.15 /1M input tokens (check current OpenAI/Anthropic pricing), screening 1000 market titles costs literal pennies.

Ship It This Week

The Project: "Hype Tax" — A Base Rate Tracker for Prediction Markets

You don't have to risk real money to build something valuable here. Here's what you can ship in 3- 4 days:

Day 1-2: Hit the Polymarket API (or Manifold for zero risk) and pull all currently open non-sports markets. Store title, current 'Yes' price, and expiry date in a SQLite table.

import requests
import sqlite3

# Manifold API - no  auth needed for reads
r = requests.get('https://api.manifold.markets/v0/markets? limit=100')
markets = r.json()

# Filter for non-sports, high-drama keywords 
hype_words = ['arrested', 'resign', 'crash', 'collapse', 'ban', 'war', 'fired ']
for m in markets:
    if any(w in m['question'].lower() for w in hype_words):
        print( m['question'], m['probability'])

Day 3: Add an LLM classification step. Feed each market title to Claude Haiku or GPT-4o-mini with a prompt like: "On a scale of 1-10, how much is this prediction market question driven by Twitter hype vs. historical base rates? Answer with just the number." Flag anything scoring 7+ as a 'No' candidate.

Day 4: Build a simple dashboard ( Streamlit or even a plain HTML table) that shows your flagged markets, their current 'No' price, and the implied edge if you assume true probability is 20% higher than market price. Share it publicly . This is a tool other traders will use, and that's a distribution channel.

The monetization path is obvious : paper trade it for 30 days, track your theoretical P&L, then decide if you want to fund a real wallet . Or charge other traders $20/month for the filtered feed. Either way, you've shipped something real.

Start here: git clone https://github.com/sterlingcrispin/nothing-ever-happens and read the core loop before you write a single line of your own.