BlackForge vs Tardis.dev
Updated
BlackForge and Tardis.dev sit on opposite sides of the same order book. Tardis is the deep raw historical archive: tick-level level 2 and level 3 books built up over many years, which you download and reconstruct yourself. BlackForge is the live computed feed: 120 order-book and trade columns per pair per five-minute window, delivered ready to query. One is built for replaying the past in full detail; the other is built for reading the present already measured.
What is the difference between BlackForge and Tardis.dev?
Tardis.dev is the deep raw historical archive: tick-level level 2 and level 3 order books built up over years that you download and reconstruct yourself. BlackForge is the live computed feed: 120 order-book and trade columns per pair per five-minute window, delivered ready to query. Raw historical reconstruction versus live computed delivery.
What each one actually is
Tardis.dev is a historical market-data archive built for backtesting. It ships tick-level level 2 and level 3 order books with full limit-order-book reconstruction, in both exchange-native and normalized formats, across more than 50 venues and over 200,000 instruments, with history reaching back years. Spot access starts at $450/mo, with All Exchanges plans from $650 up to $6,000/mo (tardis.dev/#pricing, read 2026-08-12). It computes zero derived metrics: it hands you the raw tape and book.
BlackForge is a live feed. It runs its own collectors against nine spot venues, aligns every venue onto one five-minute clock, and publishes 120 computed columns per (exchange, symbol) per closed window across roughly 11,800 pairs. The most recent row is at most five minutes old, and the columns inside it are already computed. You query a row; you do not download an archive and rebuild the book.
| Tardis.dev | BlackForge | |
|---|---|---|
| Unit | Raw ticks you reconstruct | One computed row per window |
| Delivery | Download archives, rebuild the book | Live query, ready to read |
| History | A multi-year raw archive | Collection began mid-2026 |
| Computed metrics | None; you build them | 120 columns per pair per window |
| Granularity | Tick-level L2/L3 | 5-minute computed bar |
| Entry price | $450/mo Spot | Free tier $0; PAYG $0.005/call |
Historical reconstruction versus live delivery
Tardis is historical-first. You download the archives for the venues and dates you want, then reconstruct the book from the tick stream to whatever state you need at whatever moment you choose. It is a backtesting substrate, not a live five-minute delivery, and it is exactly what a quant needs to replay a long window across a regime change.
BlackForge delivers a live computed five-minute bar. The most recent window is at most five minutes old, and the columns are already computed, so you do not reconstruct the book and you do not handle any per-venue protocol yourself. The two providers answer different questions: Tardis answers what the book did over a long history you rebuild, BlackForge answers what the book is doing now, measured, without a reconstruction step.
Reconstruction is real work, and it is worth being honest about its size. A Tardis archive is the raw event stream, so to get a usable book at a given moment you replay every message from the last snapshot forward, apply each delta in order, and carry the state. Do that across 50-plus venues and their exchange-native quirks, over years of dates, and you are running a nontrivial storage and compute job before a single number comes out. That job is the price of total control over the raw data, and for a serious backtest it is often worth paying. It is also exactly the job BlackForge has already run when it hands you a computed row.
| Name | Raw L2/L3 ticks | Deep history | Live delivery | Computed columns | Normalized |
|---|---|---|---|---|---|
| Tardis.dev | Yes | Yes | No | No | Yes |
| BlackForge | No | No | Yes | Yes | Yes |
The columns Tardis leaves you to build
Tardis hands you the raw tape and book and computes nothing on top: every metric is yours to build. BlackForge computes 120 columns, including the behavioural block that describes how the book moved within a window. Two of those columns come straight out of the same book Tardis would hand you raw: the median time a price level stood before it cleared, and how many levels were rebuilt after clearing. On Tardis you would reconstruct the book tick by tick and write that measurement yourself; here it arrives in the row.
The behavioural block is wider than those two columns. It covers gross liquidity added and removed on each side, the depth resting in bands away from the top of book, seven-rung order ladders, and large trades bucketed by aggressor side. Each of those is a defined statistic over the window, and each is derivable from a fully reconstructed book. The difference is not that Tardis cannot get there; it is that Tardis gives you the raw material and BlackForge gives you the finished measurement. If your work is the measurement rather than the reconstruction, the row is the shorter path.
BTCUSDT on binance
- Median level lifetime
- 1,800ms
- Level flicker count
- 7,970
Where Tardis.dev wins
Tardis.dev
- It holds a tick-level L2/L3 archive built up over years, far deeper than BlackForge, whose collection began mid-2026.
- It ships true order-by-order, level 3, market-by-order data.
- It offers exchange-native formats alongside normalized ones.
- It is exactly what a quant backtesting across a long window and a regime change needs.
Choose them if you need raw tick-level historical order book for backtesting, true L3 market-by-order data, or a history window BlackForge does not yet have.
Which one should you reach for?
If you are backtesting across a long window and you want to reconstruct the book exactly as it stood, tick by tick, Tardis is the right tool and its deep multi-year archive is the reason. If you want the current state of the spot market already computed, delivered live within a five-minute window, with the depth bands and liquidity flows and resting-level measurements already in the row, that is what BlackForge is for. Some teams use both: Tardis for the deep raw history, BlackForge for the live computed feed they would otherwise have to build and run themselves.
Is Tardis.dev real-time?
Primarily it is historical. Tardis is built around archives you download and reconstruct, so its core product is deep raw history rather than a live delivery. BlackForge is the live computed feed: the most recent five-minute window is at most five minutes old, and the columns inside the row are already computed, with no reconstruction step on your side.
Does Tardis compute order-book metrics?
No. Tardis ships raw level 2 and level 3 order books and full limit-order-book reconstruction, and computes zero derived metrics. Every depth band, liquidity flow, and resting-level measurement is code you write on top of the raw tape and book it provides. BlackForge publishes 120 such columns already computed per pair per five-minute window.
Does BlackForge have tick-level history?
No. BlackForge publishes a five-minute computed bar, not tick-level data, and its collection began in mid-2026, so it does not hold deep raw history. Tardis owns that ground, with tick-level level 2 and level 3 archives built up over years. If you need years of raw reconstructable ticks, that is Tardis, not BlackForge.
Which is better for backtesting?
Tardis, if backtesting means replaying deep raw history and reconstructing the book tick by tick across a long window and a regime change: its deep multi-year archive and level 3 data are built for exactly that. BlackForge suits testing against a ready computed feed within its window, where you want the columns already measured rather than the raw tape to rebuild them from.
See the full metric reference, or browse all comparisons.