Anton König

Small, correct data tools. Two areas: prediction markets (Polymarket, Kalshi) and European data sources that most actor developers skip.

I spent months capturing and auditing Polymarket order-book and oracle data for my own research. The lesson was humbling: most public numbers in this space are wrong for boring reasons.

• Fees ignored. Polymarket and Kalshi both charge taker fees that scale with price. A 1% "spread" is often negative after fees.
• Top-of-book treated as executable. The best price usually covers a few shares, not your stake.
• Book order assumed. Polymarket's CLOB returns asks descending, so a naive reader takes the worst price and calls it the best.
• Silent data gaps. My own capture had 33 feed outages that quietly corrupted a backtest until I audited it.

Every one of those mistakes produces a confident, impressive, useless number. So my actors are built the other way around:

1. Walk the real order book at the size you actually trade.
2. Apply each venue's real fee schedule, per market.
3. Show confidence scores and flag results that need human review.
4. State the limitations in the README instead of hiding them.

If an output looks too good, my actors say so rather than letting you find out with money. If you need a field an actor does not return, or something breaks when a site changes, open an issue. I read them and I respond.

Based in Switzerland. Backend engineer (JVM, Go, Python), long-time home-lab operator, which is where the taste for reliable long-running data capture comes from.