All About AI

Polymarket AI Trading UPDATE + New Treasure Hunt Concept

By · · 4 min read

Yesterday I shipped two Polymarket strategies built off the 100x mispricing analysis: the window-switch snipe and the 24-hour preload. Less than a day later one of them already paid out — 50¢ in, $49.50 out. That's the kind of asymmetric win that makes a 1-in-16 hit rate actually positive EV. This post walks through what happened, what the second strategy is doing, and introduces a separate experiment I'm running in parallel — the "Follow the White Rabbit" treasure-hunt drops.

Watch the video:

Strategy 1: window-switch snipe — first paid fill

This is the one from the 100x strategies post: at the exact moment a 5-minute BTC up/down window closes and the next opens, fire $0.01 bids on both up and down across BTC, ETH, SOL, and XRP (eight bids), then cancel after 120 seconds so we don't get filled in the late-window stale-bid zone.

After 17 fills overnight, one of them paid:

17 fills is too small a sample to claim anything statistically. But the fact that any fill closed positive after one night confirms the mechanic exists in the wild. That's enough to keep running it.

Strategy 2: 24-hour preload — still incubating

This one places resting 1¢ bids on freshly listed future windows ~24 hours ahead of close, on both sides, cancel-on-timeout. The idea is to be first in the book when the close-time sweepers eventually pass through.

Current state:

The hit rate is a long way below what would make this profitable, but the per-trade risk is intentionally tiny — 50¢ to potentially win $50, with the cancel-on-timeout protecting against the late-window stale-bid trap. I'm letting it run another few days before judging. If two fills both losing is just bad luck on a small sample, that's fine. If the geometry of the order book actively works against early preloads (which is plausible — closer-to-close bids might cut in front), the strategy doesn't survive contact with reality and I kill it.

Follow the White Rabbit — the parallel experiment

Separate from the trading, I'm running a treasure-hunt concept I'm calling Follow the White Rabbit. Mechanic is two clues, one wallet:

I dropped the first one yesterday. Hidden the private key on a page nested inside one of my existing posts. Someone in the Discord found it inside an hour and walked away with $25 in USDC. The full mechanic and "how to claim" walkthrough lives on the whiterabbit page.

One thing I figured out from round one: the wallet needs to be pre-loaded with a tiny amount of SOL for gas. Otherwise the winner needs to fund the wallet first to sweep it — which looks exactly like a honeypot scam and kills participation. With pre-loaded gas, the winner just imports the key and sends — one click, no upfront cost. Round one took about 60 minutes from drop to claim with that change in place.

Why am I doing this at all? Two reasons. One, I had some USDC sitting around where pulling it back to fiat would cost ~50% in taxes — much more interesting to spend it on community engagement than send half to the tax authority. Two, it's a stress test of the puzzle/distribution loop: if a puzzle drives sub-hour participation across X and the site, that's a real audience signal worth understanding.

How they fit together

The Polymarket strategies and the treasure hunt are unrelated mechanically — one is a trading bot iteration, the other is a community game. But they share the same underlying setup: small money, observable outcomes, fast iteration. I learn more in a weekend of $50 trades and $25 drops than a month of reading.

If you want to follow the strategies in real time, the AI_automata Discord is where I post the dashboards and the white-rabbit drops as they go live. The X account is where wallet addresses go.

Resources