Type inclusion for records changes CSV output

Consider the following valid Ballerina program that maps two identical data structures to CSV with io:fileWriteCsv:

import ballerina/io;

type Base record {|
    int base1;
    int base2;
|};

type Derived1 record {|
    *Base;
    int derived1;
    int derived2;
|};

type Derived2 record {|
    int base1;
    int base2;
    int derived1;
    int derived2;
|};

public function main() returns error? {
    Derived1[] d1 = [{ base1: 1, base2: 2, derived1: 3, derived2: 4 }];
    Derived2[] d2 = [{ base1: 1, base2: 2, derived1: 3, derived2: 4 }];

    check io:fileWriteCsv("derived1.csv", d1);
    check io:fileWriteCsv("derived2.csv", d2);
}

I'm expecting the content of the both output CSV files to be identical to:

1,2,3,4

But instead I'm getting:

$ cat derived1.csv 
3,4,1,2
$ cat derived2.csv 
1,2,3,4

I'm surprised that type inclusion changes the order of the columns in CSV file. Is this a bug or by-design? If by-design what are the column ordering rules?

I'm using:

$ bal --version
Ballerina 2201.2.0 (Swan Lake Update 2)
Language specification 2022R3
Update Tool 1.3.10