Exchange inflows as an early-warning signal for Bitcoin drawdowns
On-chain flow shocks compared with volatility regimes as a risk overlay, tested walk-forward.
Alpha Lab: research in the open.
Research notes, the signal library and model leaderboard — with methodology published alongside every result.
Research notes
Latest from the lab.
Common sources of bias in crypto backtests — and how point-in-time data helps address them
Survivorship, look-ahead and listing bias, examined across hundreds of cryptoassets.
LLM sentiment embeddings vs. keyword scores
How each may relate to next-day market behaviour, and how that relationship may change over time.
Rates, liquidity and the crypto beta
How global liquidity conditions shift the correlation between digital assets and equities.
Volatility analysis for 24/7 markets
Applying volatility analytics to assets that never close.
Anatomy of a coordinated market spike
Research into market patterns around coordinated spikes.
No notes in this category yet — try another filter.
Signal library
Signals, with their receipts.
Each signal includes a model card and out-of-sample statistics from walk-forward testing. Signals stay “In research” until they pass our promotion criteria.
| Signal | Universe | Horizon | Published at release | Status |
|---|---|---|---|---|
| ETH 7d mean-reversion | ETH, SOL, AVAX | 7 days | Published at release | In research |
| BTC exchange-inflow risk overlay | BTC | 1–5 days | Published at release | In research |
| Funding-rate analysis | Selected cryptoassets | 1 day | Published at release | In research |
| Sentiment momentum | Top 50 by volume | 3 days | Published at release | In research |
| Volatility regime classifier | BTC, ETH | Daily | Published at release | In research |
| Listing anomaly alert | New listings | Intraday | Published at release | In research |
Model leaderboard
Comparing models on out-of-sample results.
Transformers, gradient-boosted trees, recurrent networks and ensembles are ranked on the same out-of-sample folds, using error, calibration and directional accuracy — never on in-sample fit.
How we evaluate modelsTrainTest (scored)
Get research notes first.
Waitlist members may receive research notes, signal releases and model cards before publication.