Technology
Every layer of the pipeline, documented.
From raw tick data to a resolved signal, we publish the reasoning behind each step — not just the output. That includes the architectures that didn't work and why.
What we actually build
Four areas of applied research, documented in the open so anyone can evaluate the reasoning, not just the results.
Model research
We evaluate how transformer and time-series architectures behave on noisy, non-stationary market data — and publish what actually holds up.
Signal processing pipelines
Open documentation on the feature engineering and denoising techniques used before any model ever sees a data point.
Backtesting methodology
Guides on avoiding lookahead bias, overfitting, and survivorship bias — the failure modes that make results look better than they are.
Responsible framing
Every research note ships with its assumptions and limitations. We do not publish guarantees, because markets don't offer any.
Go deeper in the Learning Center
Structured lessons on the same methodology, from time-series foundations to model design and responsible backtesting.