spread bps
0.06
Cortazar Labs is a market-microstructure desk for crypto perpetuals. We capture every order-book update from five exchanges into an Iceberg lakehouse and surface a small set of intraday signals — the same ones we use to trade. No sentiment, no tip sheet, no performance claims.
| asset | mid (USDT) | 1h spark | 24h Δ | fund 1h | basis bps | OI Δ 1h |
|---|---|---|---|---|---|---|
BitcoinBTC-PERP · binance | 108,245.30 | +1.82% | +0.94 bps | +12.4 | +1.2% | |
EthereumETH-PERP · binance | 3,512.18 | +1.41% | +0.81 bps | +09.1 | +0.4% | |
SolanaSOL-PERP · binance | 224.07 | −2.34% | −0.42 bps | −04.3 | −0.8% | |
BNBBNB-PERP · binance | 684.20 | +0.62% | +0.21 bps | +02.1 | +0.1% | |
XRPXRP-PERP · binance | 2.6814 | +0.18% | +0.04 bps | +00.4 | +0.3% | |
DogecoinDOGE-PERP · binance | 0.18421 | −0.74% | −0.12 bps | −00.8 | +0.7% |
Layered cancellation pressure within 120ms of a same-side aggressive print, levels 2–6, normalized to median depth. Reading is elevated — past the 6m median — but well below the 90th-percentile threshold.
Forward signed adverse selection on resting top-of-book quotes over a 400ms horizon. Refresh-driven cancels excluded. Currently above the 6-month median; not extreme.
Latency, queue position, quote churn. We surface the measurements that decide whether your colo box is competitive — not the strategies that run on top of them.
Order-flow imbalance, aggressor balance, sweep detection. Slower than HFT, faster than session — for traders who hold positions in minutes, not microseconds.
For session traders: basis decay, cross-venue funding dispersion, CVD across the four busiest perpetuals. Values move; methodology rarely does.
The 365-day tape is for slower questions: venue migration, realized volatility distributions, the shape of funding-curve regimes. Same Iceberg tables, longer windows.
Five exchanges, WebSocket + REST snapshots, written to Redpanda then Flink, then Iceberg on a single on-prem K3s rack. No vendor middle layer.
Every signal is a parameterized SQL or PyArrow definition in cortazar/march. The same code runs intraday and on the 365-day backfill.
A signal appears on the board only after replication on at least 60 days of unseen tape. We update values, not the definitions, on-the-fly.
The full terminal — every symbol, every signal, every minute — is one step away. Access is invite-only while infrastructure stabilizes; sign in if you have credentials, or request access and we'll route you a key.
A market-microstructure desk for crypto perpetuals. Operated by one engineer who trades, on infrastructure that fits in one closet.