Frequently asked questions
The questions we hear most about ANC Labs and its work, with short answers.
01What is ANC Labs?
ANC Labs is a one-person, independent research lab that builds machine learning for financial data with a very low signal-to-noise ratio. Its work covers noise-robust model architectures, a synthetic lab for measuring robustness, and a market-neutral live system called milliways01. The research is a means, not the end: the goal is systems that earn sustainable returns in live markets, after costs.
02What does "noise-robust machine learning" mean?
In financial data only a small part of future returns is predictable; the rest is noise. A noise-robust model keeps recovering the true signal as noise increases, instead of memorising the noise. ANC Labs measures this as efficiency: the information coefficient (IC) a model captures, divided by the best linear IC achievable on the same data.
03What is DA-xLSTM?
DA-xLSTM is a noise-robust variant of the xLSTM architecture developed at ANC Labs. It combines a clipped training target, a smaller hidden state (16 → 4) and an average of five independently trained models. In the synthetic lab it delivered 16–36% higher signal efficiency than a standard xLSTM in all 5 test cells.
04What is the synthetic SNR lab?
It is a data generator calibrated to the statistics of real crypto perpetual markets, with fat tails, cross-sectional correlation, GARCH-style volatility, regime shifts and a funding component, plus a planted signal of known strength. Because the true signal is known, the lab can measure how much of it a model recovers. All 13 of 13 calibration targets fell within tolerances set in advance, and 1,583 lab measurements were run.
05Why is financial data hard for machine learning?
Returns have fat tails (a Student-t tail parameter of about 2.8 in the data used), assets move together (average correlation of about 0.58), and volatility clusters. In this setting flexible models start memorising noise early: a standard xLSTM reached its best validation score within the first two epochs in 86% of runs. That is why ANC Labs measures robustness before anything else.
06How does ANC Labs avoid misleading backtests?
Every test follows the same rules: no look-ahead; results count only after fees, spreads and market impact; the universe is built without survivorship bias from 832 USDT perpetual symbols, delisted ones included; hypotheses and pass criteria are written down before a test runs; and every number that drives a decision is reproduced independently. Critical safeguards are themselves tested by breaking the code on purpose.
07What is milliways01?
milliways01 is a market-neutral, cross-sectional portfolio on roughly the hundred most liquid Binance perpetuals, combining one family of signals from order flow with one from funding. It has been in a live forward test since October 2026. In a historical simulation over 2020–2025 with realistic cost and execution assumptions (model v1), it showed a Sharpe ratio of 1.76, an annual compound return of 25.3% and a maximum drawdown of 14.4%.
08Are these results from real trading?
No. The performance figures on this site come from historical simulations run under realistic cost and execution assumptions. They are not the results of actual trading and do not guarantee future results.
09Is any of this investment advice?
No. ANC Labs publishes research; nothing on this site is investment advice or an offer to buy or sell any financial product.
10How can I contact ANC Labs?
By email at info@anclabs.co.