Looker Studio inconsistent in summing metrics? - google-analytics

Question

I've come across what appear to be random inconsistencies in how Looker Studio is aggregating data from the same Analytics source. I can't tell if it's an issue on the software's end, or a result of my incorrectly summing and joining the data in question (or possibly misunderstanding how the aggregation is supposed to work).
Would appreciate any information that could confirm if it's an issue with my process (especially regarding the choice of database join), or alternatively whether my expectations are correct and thus it's an issue on Looker Studio's end.
Data
I have three data sources: A UA Analytics profile, a GA4 Analytics profile, and a blended source that is a full outer join of the two profiles on the Date dimension (UA left, GA4 right).
The two profiles were added as data sources using the default DS Google Analytics connector. There aren't any filters or source-level manipulation of data, though for the purposes of this report I'm only looking at some basic metrics for the month of April 2022, via the standard date range selectors in the report.
The UA Analytics profile:
Sample of the UA metrics being used:
Date
Users
Sessions
Bounce Rate
Goal Completions
Apr 1, 2022
870
955
0.756020942408377
85
Apr 2, 2022
849
923
0.782231852654388
82
Apr 3, 2022
1023
1105
0.774660633484163
100
Apr 4, 2022
1007
1095
0.74337899543379
121
Apr 5, 2022
1111
1210
0.765289256198347
130
Apr 6, 2022
1010
1111
0.756975697569757
92
Apr 7, 2022
1007
1141
0.743207712532866
100
Apr 8, 2022
928
1009
0.77205153617443
80
Apr 9, 2022
941
1054
0.80550284629981
72
Apr 10, 2022
1002
1113
0.761006289308176
85
...
...
...
...
...
Apr 30, 2022
854
931
0.767991407089151
75
The GA4 profile:
Sample of the GA4 data being used (it was only implemented on the domain halfway through the month, so metrics are empty for the first few weeks):
Date
Total users
Sessions
Engagement rate
Conversions
Apr 18, 2022
766
791
0.378002528445006
0
Apr 19, 2022
890
930
0.394623655913978
0
Apr 20, 2022
849
884
0.39027149321267
0
Apr 21, 2022
844
891
0.354657687991021
0
Apr 22, 2022
745
780
0.33974358974359
0
Apr 23, 2022
833
871
0.330654420206659
0
Apr 24, 2022
878
910
0.306593406593407
0
Apr 25, 2022
904
949
0.355110642781876
0
Apr 26, 2022
932
982
0.346232179226069
0
Apr 27, 2022
910
963
0.349948078920042
0
Apr 28, 2022
878
911
0.354555433589462
0
Apr 29, 2022
809
850
0.342352941176471
0
Apr 30, 2022
782
832
0.278846153846154
0
The data blend (date range is auto for both, no filters or segments applied):
Expected Output
Because they're originating from the same data source, I would expect the metrics from the data blend to be equal to those of their relative individual profiles. That is, I would expect the Total Users metric from the GA4 data source, and the Total Users metric from the blended data source (which is referencing the same GA4 data source), to show the same numbers, with the same true of the UA data source.
Instead, I'm getting the following results on Chart Scorecards:
Source
Total users
Sessions
Engagement Rate
Conversions
User engagement
GA4 Profile
9874
11,711
34.20%
0
57:30:07
Blended GA4
11,020
11,544
34.78%
0
57:30:07
Source
Users
Sessions
Bounce Rate
Goal Completions
Avg. Session Duration
UA Profile
25,528
31,370
76.34%
2808
00:00:35
Blended UA
28,709
31,370
76.26%
2808
00:17:44
I expect all of the GA4 metrics to match each other, but while Conversions and User Engagement match, the Total Users, Sessions, and Engagement Rate don't.
For the UA data, Sessions and Goal Completions match, but the Users, Bounce Rate, and Avg. Session Duration don't.
There doesn't appear to be a pattern as to which metrics have a discrepancy or by how much. They aren't multiples of each other, so I don't think it's a simple issue of rows being re-counted in the join.
Even weirder, if I make Chart Tables out of the same data, the tables' summary rows don't always show the same results as the scorecards, even though they're referencing the exact same data.
In addition, if I manually sum up all the metrics in each column by spreadsheet, I get results different from those in the tables' summary rows:
Source
Users
UA Sessions
Bounce Rate
Goal Completions
UA summary row
25,528
31,370
76.34%
2808
UA manual calculation
28,709
31,370
76.26%
2808
Source
Total users
GA4 Sessions
Engagement Rate
Conversions
GA4 summary row
9874
11,711
34.2%
0
GA4 manual calculation
11,020
11,544
34.78%
0
Source
Users
Total users
UA Sessions
GA4 Sessions
Bounce Rate
Engagement Rate
Goal Completions
Conversions
Blended summary row
28,709
9874
31,370
11,711
76.26%
34.2%
2808
0
Blended manual calculation
28,709
11,020
31,370
11,544
76.26%
34.78%
2808
0
These discrepancies happen even when sampling only two rows of data at a time, and never by the same amount.
For this specific report, the GA4 data for Total Users, Sessions, and Engagement rate have discrepancies of 1146, -167, and 0.58 respectively for the entirety of April (or 111%, 99%, and 102% as a percentage of the GA4 data source).
Populating the same report with Analytics' default demo data (see link at bottom), the GA4 data for Total Users, Sessions, and Engagement rate have discrepancies of 51732, 2432, and -2.1 for the entirety of April (or 135%, 102%, and 97% as a percentage of the GA4 data source).
Looking at just April 1 & April 2 with that same demo data, there are discrepancies of 160, 93, and -0.71 (or 102%, 101%, and 99% as a percentage of the GA4 data source).
Applying CAST to the metrics (e.g. CAST(Total Users as number)) has no effect and results in the same metrics and sums.
Chart configuration
All scorecards have the same default configuration, with the only difference being the data source that's referenced:
Same for the tables, with the exception of relabeling Sessions and Sessions as UA Sessions and GA4 Sessions for clarity:
Issue
Why are these discrepancies happening, since to my understanding these charts should be pulling the same data and summing in the same manner, and therefore should have the same results? Am I misunderstanding the output I should be receiving?
I've triple checked all my charts to make sure they're using the same aggregation functions, date range, source metrics, etc. etc. so I don't know why so many of these summed metrics are inconsistent. I thought it might be a matter of date range processing, but I don't think that would explain the different sums in different chart types for the exact same data source, or why the UA sessions match up with each other but the GA4 sessions don't.
I've also tried different join methods for the blended data, but all return the same results except for cross joining (which multiplies all resulting values, as expected).
Am I missing something? My manual calculation results make me feel like the blended and spreadsheet metrics are more "trustworthy", but the blended data still showed incorrect summary row results in the report table, so I'm genuinely unsure if I'm getting correct results for any given data source.
Report
Made a copy populated with demo Analytics data, so the numbers aren't the same, but similar discrepancies are happening: https://datastudio.google.com/reporting/40bab31a-a0d0-4b79-8dcf-25c11279f229
Spreadsheet with manual summing of exports of the tables from the same report (note that they don't match said report): https://docs.google.com/spreadsheets/d/1CvM-4PqPNfBqNIlzJEe9QQVQ5tWNVkOdaZOVLfqnRRU/edit?usp=sharing
Even more simplified report that specifically highlights the discrepancy between default data source aggregation, and aggregation via spreadsheet: https://datastudio.google.com/reporting/a4f989ed-474e-4f04-955d-5ffb6339fc3a