v0.11 · ITCH Replay
Did nanobook reconstruct a real exchange day correctly and quickly?
One full NASDAQ TotalView-ITCH 5.0 trading day (2019-01-30), replayed through nanobook's limit order book. Two questions: does it process real cancel-heavy flow fast enough, and does the reconstructed book stay self-consistent at every event?
Latency, end-to-end
Three pipeline stages measured independently: ITCH parse → book update → strategy-to-order. Long tails come from cancel-heavy bursts.
| Stage | p50 | p95 | p99 |
|---|---|---|---|
| Parse | 51 ns | 89 ns | 124 ns |
| Book update | 152 ns | 287 ns | 431 ns |
| Strategy to order | 318 ns | 712 ns | 1,204 ns |
Message rate through the session
Per-second ITCH message throughput. The report should make open/close bursts visible without hiding midday calm.
Spread distribution
Histogram of top-of-book spread in cents after each book-changing event. A healthy reconstruction clusters at one tick.
Book reconstruction (sample)
Top-5 levels at one random timestamp during the replay. Self-consistent: monotonic prices, no crossed book, conserved volume.
| Bid level | Price | Shares | Orders |
|---|---|---|---|
| 1 | 104.97 | 2,400 | 8 |
| 2 | 104.96 | 5,100 | 14 |
| 3 | 104.95 | 3,800 | 11 |
| 4 | 104.94 | 7,200 | 23 |
| 5 | 104.93 | 4,600 | 15 |
(0.95 bps)
| Ask level | Price | Shares | Orders |
|---|---|---|---|
| 1 | 104.98 | 1,900 | 6 |
| 2 | 104.99 | 4,300 | 12 |
| 3 | 105.00 | 6,700 | 19 |
| 4 | 105.01 | 3,200 | 10 |
| 5 | 105.02 | 8,500 | 28 |