Some USDT-margined futures bookDepth data entries significantly deviate from mark prices

[daeseoklee]

Hey guys, first of all, I'd like to say that I am really grateful about your efforts to make this detailed market data public.
Upon looking at the bookDepth data of USDT-margined futures (which are available only in "daily" section), I think I detected some signs of data misalignment.
I want to understand what is happening.
Also, please correct me if there's any misunderstanding on my part.

So take the following lines extraced from
BTCUSDT-bookDepth-2025-05-19.zip of https://data.binance.vision/?prefix=data/futures/um/daily/bookDepth/BTCUSDT
"""
timestamp,percentage,depth,notional
2025-05-19 11:07:31,-5,7885.27500000,647368074.36920000
2025-05-19 11:07:31,-4,7178.45300000,590768470.27090000
2025-05-19 11:07:31,-3,5443.14900000,450287742.99870000
2025-05-19 11:07:31,-2,3989.69900000,331452387.75030000
2025-05-19 11:07:31,-1,2294.32400000,191369991.21780000
2025-05-19 11:07:31,1,2097.41000000,176607765.20950000
2025-05-19 11:07:31,2,3485.14100000,294629128.47080000
2025-05-19 11:07:31,3,4813.86500000,408742768.20550000
2025-05-19 11:07:31,4,5699.71200000,485628858.29740000
2025-05-19 11:07:31,5,6491.51500000,555012399.28440000
"""

If you compute notional/depth and list them in the percentage order, it becomes something like:
82098.35, 82297.46, 82725.60, 83077.04, 83410.19, 84202.79, 84538.65, 84909.48, 85202.35, 85498.13

In my understanding, these values must be within +-5% the mark price.
However, the mark price at that point in history is about 103k, which is clearly inconsistent with this result.

The following image shows a more detailed analysis across timeline.

Essentially, this shows 5%, 25%, 50%, 75% and 95% quantile values of (notional at 5%) / (depth at 5%) / (recent tick price) - 1 in monthly intervals.
As you may see, the figure at 2025-04 and 2025-05 seems alarming, while the other values seem to be in the valid range.

Here are dates in which at least one negtive value was found:

['2023-01-09', '2023-01-16', '2023-01-21', '2023-02-08', '2023-03-17', '2023-03-18', '2023-08-29', '2023-10-01', '2023-10-16', '2023-11-15', '2024-02-26', '2024-02-27', '2024-03-02', '2024-03-03', '2024-03-05', '2024-03-09', '2024-03-11', '2024-03-13', '2024-04-18', '2024-04-19', '2024-06-12', '2024-08-05', '2024-08-08', '2024-12-05', '2025-04-19', '2025-04-21', '2025-04-22', '2025-04-23', '2025-04-24', '2025-04-25', '2025-04-26', '2025-04-27', '2025-04-28', '2025-04-29', '2025-04-30', '2025-05-01', '2025-05-02', '2025-05-03', '2025-05-04', '2025-05-05', '2025-05-06', '2025-05-07', '2025-05-08', '2025-05-09', '2025-05-10', '2025-05-11', '2025-05-12', '2025-05-13', '2025-05-14', '2025-05-15', '2025-05-16', '2025-05-17', '2025-05-18', '2025-05-19']

Another concrete evidence that something went wrong:

Note that the flat line includes more than 1000 datapoints.

Thank you for reading this, and have a good day.