Blog · 8 June 2026
BitterQuant is live
BitterQuant runs autonomous trading strategies on a custom-trained AI forecaster. The forecaster, Centaurium 1.0, reads short-horizon odds for every name in a US mid-cap universe. A deterministic policy turns those odds into target weights. An executor mirrors the resulting book to a copy-trading broker, 24/5, with no human in the loop.
The bet behind that design is the Bitter Lesson: general methods that scale with learning and compute beat hand-crafted rules, and the gap widens over time. So we put the learning in the model and keep the trading policy small, deterministic and auditable. Why we build it this way.
Three strategies launch on Centaurium 1.0 — one forecaster at three points on the risk frontier: Americano 1.0 (vol-managed, low drawdown), Boulevardier 1.0 (balanced, index-relative) and Old Pal 1.0 (concentrated, higher risk).
What we publish is the part most copy-trade marketers skip. Each strategy card carries its full 2022–2024 backtest — per-year returns against the S&P 500, the drawdown, and every backtested trade — labelled hypothetical, and the live record will sit next to it, gap included. Each strategy also carries pre-registered kill criteria: rules fixed in advance that halt it if live performance breaks down. Written before the fact, so they cannot be moved after it.
All three are in paper trading. The decisions and the prices are real; the account is not. No client capital is at risk, no strategy is open to copy, and the numbers on the site are still backtests. That is deliberate — the track record has to exist before it can be sold, and building it takes months of calendar time that nothing can compress.
This blog is where changes get announced: new model versions, new strategy versions, kill-criteria events. Nothing changes silently under a name that stays the same.