Petio Petrov

Low-latency trading infra. Order book engines, feed handlers, chasing microseconds.

[Fede Cardoso @cardosofede]

we are running an agent trading hackathon with a total prize pool of 20000... are you going to join or just watch from your computer?

[hummingbot @_hummingbot]

🔌 New Lambdaplex spot connector is live! 🚀

Hedera-native onchain exchange with offchain matching + onchain settlement, now tradable from Hummingbot.

Thanks @lambdaplex 🙏

Thanks to community dev, @petioptrv_ca

📰 https://t.co/DN1aEpjcmP

[Petio Petrov @peteptrv]

@techie__ghost Looks like we’ll be colleagues!

[Petio Petrov @peteptrv]

Takeaway: if a venue mirrors its book, single-market YES/NO arb is a measurement artifact, not a trade. Always check both prices were quoted at the same instant before you call it arbitrage.

[Petio Petrov @peteptrv]

Was looking through some Polymarket arb articles this week and hit one with a $40M headline. It’s an old paper now (Aug 2025; arXiv:2508.03474v1), but the mechanics are still relevant, and one of its biggest findings fails on the most misunderstood prediction market mechanism.

[Petio Petrov @peteptrv]

Their realized-profit section is more honest, it tracks users assembling YES and NO across a ~1h window. But that’s cross-time accumulation with real leg risk, not instant risk-free arb. Paper labels both “arbitrage” and that’s where the "magic" part happens.

[Petio Petrov @peteptrv]

@sachaarbonel Cheap code is cheap

[Petio Petrov @peteptrv]

Caveat: ran locally, no colo, no geo tuning. Absolute numbers would tighten a lot from a well-placed box, but the spread between connections is the interesting part to me.
Script here: https://t.co/BmzhM8LB85

[Petio Petrov @peteptrv]

Spun up 20 parallel websocket connections to the same Binance BTCUSDT trade feed to see how much they'd disagree.
Over ~2800 trades, one client was first more than 40% of the time. It came last twice.
Same feed, same machine.

[Petio Petrov @peteptrv]

So fan out. Several connections, take whichever tick lands first, timestamp everything so you know what you're actually getting.
Costs you nothing but sockets.

[Petio Petrov @peteptrv]

But no connection was always fastest. Every one of them led at some point.
That's the actual finding. If you're listening on one socket you're not sampling a slow feed, you're sampling a random one.

[Petio Petrov @peteptrv]

Per-trade delay behind whichever connection got there first.
9 of the 20 sat over a second behind the leader at the median.

[Petio Petrov @peteptrv]

@bored2boar Prediction markets mirror the order books. The prediction market itself makes sure that the best bid for YES exactly equals 1 - the best ask for NO and vice versa. This means that the arb you described will never happen by function of the market’s design.

[Michael Feng @fengtality]

https://t.co/8jPbnY5Kvq

[Petio Petrov @peteptrv]

Interesting (evolving) AI benchmark: https://t.co/4AxpbP64j6