To Determinism and Beyond
A non-physicist's two-week (and a bit) expedition into quantum surface code decoding
This site documents a two-week exploration of quantum error-correction decoding. The goal was not to produce a new state-of-the-art decoder, but to understand the structure of the problem through ~50 experimental approaches.
I'm Baruchi, co-founder and CTO at Substrata. My background is in speech processing, signal processing, mathematics, AI/ML, and I have been coding for most of my life. I am no quantum physicist, though I did read about it and understand part of the physical theory of it.
This project started as a casual conversation with Claude after work, kids, in one of my "I have to clear my mind from everything" moments. I pivoted the conversation to quantum computing, learnt about the existence of error correction, and I realized the problem was built from things I'd spent years working with - just arranged in a way I'd never seen before.
The question of why can't we create a deterministic, time-bounded QEC decoder buzzed in my head. I couldn't let it go. Over the next two weeks, working after hours with AI agents from my phone, I explored roughly fifty different approaches to decoding surface codes in bounded time. Some worked. Many didn't. All of them taught me something.
In a proper research project I'd start with a literature review - but this was a late-night side project, and honestly I just wanted to build things and see what happens. All the references on each experiment were found after the fact.
Surface code lattice · syndromes · matching
Most interesting results
In 2D, a neural decoder beats MWPM by 10% at d=3. In 3D, Sinkhorn + 3-opt reaches 1.03x MWPM at d=3 and 1.18x at d=5 - all with deterministic, bounded time.
After ~50 approaches and many dead ends. Understand the problem or jump to the full results.
I'm sharing this because I think some of the results are meaningful - maybe someone will find it interesting what did not work and why, or find flaws in my work. I don't have a lab or a quantum computer to take this further, though I would love to see some of these ideas in action on a real quantum computer.
Feel free to contact me if you found this interesting, got questions, or any other insights.
P.S. If you published related work and I missed it - let me know, I'll be happy to add the reference.