MBO, MBP and BBO: order book data, level by level

The levels, side by side

The difference is easiest to see on one real instant. This is the BTCUSDT perpetual book on Bybit, rebuilt from the free archive, as an L1 feed sends it and as L2 shows it:

Which level does a backtest need?

It depends on what the strategy does with the market, not on how fast it trades. A strategy that reads closes and holds for hours needs bars, and nothing on this page. One that crosses the spread many times a day needs L1, because a backtest that fills at the close rather than at the bid or the ask hides the cost that decides whether it pays.

A strategy that reads the book, the size waiting on each side or how it leans over the top five levels, needs L2: in Manifold-BT that is bt.book.imbalance(5) or bt.book.bid_size(3) in the same expression language as any other signal.

A strategy that rests passive orders, a market maker or a limit entry, needs the queue. A resting order is filled because everything ahead of it at its price went first, not because the price touched it. L2 lets a backtest estimate that queue; L3 lets it read it.

Why L2 can only estimate a queue

A level-2 book shows each level's total size. When the size at your price drops from ten to eight, the book does not say whether the two left ahead of you, which moves you up, or behind you, which does not. So on L2 Manifold-BT places a new order behind the whole depth shown at its level when it is posted, lets prints at that price work through it, and takes a rate of cancellations ahead as a declared assumption. When the book shows nothing at that level, the queue is unknown: the run uses a declared fallback or refuses by name, and never reads it as empty, which would put the order at the front of a queue nobody saw.

On L3 the guess goes away. Every order has an identifier, so an execution or a cancellation of an order ahead takes its size out of the volume ahead, and an execution of an order behind yours means the market went through your place and yours was filled.

A maker that reads its place in the queue

The strategy below joins the best bid and ask with a round lot each and stands aside on a side where more than 20,000 shares wait ahead of it. On a day stored order by order, queue_ahead is that size exactly. The full run, on the Nasdaq sample day of 9 December 2025, is example 32 in the repository.

import manifoldbt as bt
from manifoldbt import book

# A day of Nasdaq TotalView-ITCH 5.0, stored order by order (L3)
store = bt.ingest_mbo("S120925-v50.txt.gz", {"AAPL": 1}, "2025-12-09")

# Join the best bid and ask, but stand aside behind a crowded queue.
# On this market, queue_ahead is the size ahead of the quote exactly.
inventory = bt.position()
book_ok = (book.bid_levels() > 0) & (book.ask_levels() > 0)
crowded_bid = bt.queue_ahead("bid") > 20_000
crowded_ask = bt.queue_ahead("ask") > 20_000

strategy = (
    bt.Strategy.create("queue_aware_maker")
    .quote("buy", bt.when(book_ok, book.bid_price_at(1)), 100, tif="GTX",
           enabled=(inventory < 500) & ~crowded_bid)
    .quote("sell", bt.when(book_ok, book.ask_price_at(1)), 100, tif="GTX",
           enabled=(inventory > -500) & ~crowded_ask)
)

Reading the tape, the book and order-by-order days into the engine is part of the Researcher plan.

Where each level comes from

For crypto, trades and L2 are free: Bybit publishes its book by price level, up to 500 levels deep, and its trades, one file per day, and Binance publishes its spot trades. L1 can be read off that same book, so it costs nothing more. The archives, their sizes and how far back they go are listed in free crypto tick data.

The free crypto archives stop at L2: none of them is order by order. For L3, Nasdaq publishes TotalView-ITCH 5.0 sample days, a whole session of every order on its book, which is what the example above reads; Databento sells MBO and MBP-10 for CME futures and US equities.

Frequently asked questions