Considerable difference between Firebase Performance and BigQuery export data

Question

For the same date range, Firebase console shows median slow rendering at around 30% and frozen frames at around 11.5% - however calculating the same logic in BigQuery gives considerably different results: slow rendering at around 26% and frozen frames around 0%. My query looks like this:

SELECT
  distinct 
  percentile_disc(round(trace_info.screen_info.frozen_frame_ratio, 2)*100, 0.5) over() as frozen_p50,
  percentile_disc(round(trace_info.screen_info.slow_frame_ratio, 2)*100, 0.5) over() as slow_p50
FROM
  `my-table-name`
WHERE
  event_type = "SCREEN_TRACE" and
  event_name = "_st_MainActivity" and
  DATE(_PARTITIONDATE) >= "2023-02-07" and
  DATE(_PARTITIONDATE) < "2023-02-08"

For given example date picker in Firebase console is set to Feb 7 - Feb 8 and percentile is set to 50% (median).

The questions that I'm trying to answer are these:

1. What's wrong with my query if anything?
2. Why such big difference assuming my query is ok?

---UPD:

I find it strange that for given set of records Google Sheet gives p90 as 7%, BigQuery's percentile_disc 0% and percentile_cont also 0%. The same timeframe, the same number of records, the same values of records.

---UPD2:

Went as far as getting data from smaller table for smaller period of time and compared percentile results from bigquery, same data percentiles from google sheet and same period data from firebase. All but firebase are approximately the same but firebase data is considerably different. I cannot explain that and would appreciate anyone with the knowledge to share ideas, if not answers.