Bring Your Own Historical Market Data: A New VisualHFT Guide
Most microstructure tooling has a hard split: one stack for the live feed, another for research on files. The two never agree, because they are two different code paths with two different bugs.
VisualHFT does not have that split. Historical data is re-injected through the same path a live exchange connector uses, reaching every study through the same dispatch. The only per-message difference is the provider identity stamped on the data: ProviderID = 98, "Historical File".
We have published a full guide to working with your own captures:
Use your own historical market data
What it covers
Four ways in. The Replay Engine reads NASDAQ TotalView-ITCH 5.0 pcap (a whole directory as one continuous session), FIX 4.2 text logs, and VisualHFT’s own .vhftbin and .vhftsess capture formats. Anything else goes through a converter.
A real ITCH walkthrough. Loading a 37 GB, 103-file trading day, including the two traps that make a capture silently produce nothing: the StockDirectory segment has to sort first, and most of a capture day is empty. The guide shows how to compose a minimal directory that jumps to the window you actually care about.
A converter you can write in an afternoon. The complete VHFT-BIN-1 byte specification, header offsets through the CRC window, plus a working dependency-free Python converter that turns ordinary book and trade CSVs into a capture the dashboard can play.
What replaying actually buys you. The guide ends at the NASDAQ opening auction. In the first 38 seconds after the bell, AAPL alone carries 41,013 ITCH messages: 33,303 adds, 4,862 deletes, 1,507 replaces, 1,011 executions, and exactly one cross print, 587,385 shares at $177.00. Being able to watch the queue form around that print, at a speed your eyes can follow, on data where you already know what happened next, is the entire point.
Why it is worth trusting
Every behavioural claim in the guide is cited to shipping source rather than to documentation, which drifts. Every screenshot and animation is a real run of the application driven through its real UI, not a mockup. The numbers are measured from the capture files themselves, and the guide states its limits in their own section, including where the pipeline is slower than you might hope and why.
That last part matters. The transport in the screenshot above reports an achieved 7.2x against a requested 1000x, and the guide explains exactly why: on a single-symbol ITCH replay the parser decodes every instrument on the tape before discarding the symbols you did not ask for, so decoding, not order-book maintenance, is the binding constraint.
Read the guide: Use your own historical market data
The Replay Engine is included with Pro. Session Recorder and Event Capture Recorder are separate add-ons; the guide labels which tier every surface needs, at the point where you would need it.