Show HN: Lobsim: L3 Limit Order Book replay and paper execution simulator

LOBSIM — Limit Order Book Simulator I was doing HFT deep RL research using L3 data and needed a simulator that’s deterministic, correct, fast, and fully observable (fills, events, diagnostics). Python-only workflows were too slow and painful to get right at scale, and other open-source tools didn’t give me the inspectability/ergonomics I needed. So I built LOBSIM: a C++20 core with Python bindings, event-by-event replay, paper trading with queue behaviour + partial fills, and a sink interface that streams structured facts—built to handle tens of millions of events while staying simple and comprehensible.

LOBSIM comes with multiple examples and straightforward docs (check README). I especially recommend trying the 3 Streamlit demos — they’re small apps built directly on top of the engine and they make the flexibility really obvious. The goal is to show how easily you can layer real research tooling on top of LOBSIM: replay exploration, strategy injection, live metrics, and observability, all in a clean workflow.

If you work with L3 order book data — microstructure research, execution modelling, or RL/HFT prototyping — I’d love for you to try LOBSIM. If you give it a spin, I’d really appreciate feedback on API ergonomics, missing edge-cases you hit in real feeds, and anything that would make the research workflow smoother. Even a quick “this was confusing/this felt great/I expect X“ is extremely valuable.

Demo videos - https://github.com/kpetridis24/lobsim/releases If you’d rather try it hands-on, the README has quick commands to run the Streamlit demos locally.