I’m using local DynamoDB with the AWS SDK for Javascript 2.1062.0.
I’ve read in the documentation and other sources that an operation in a transaction consumes twice as many Write Capacity Units (WCUs) as a single equivalent operation: 1 to prepare the transaction + 1 to the validation.

However, I must be missing something with the API consumed capacity returns.

I have a table with 2 global secondary indexes.

If I push a very basic item (only the Partition Key PK and the sort Key SK) with a PutItem request, the API reports a consumption of 1 WCU. But if I do the exact same operation in a transaction with TransactWriteItems, the response mentions 4 WCUs instead of the expected 2.
If I also specify the keys of the secondary indexes (PK1/SK1 + PK2/SK2), the API reports a consumption of 3 WCU for a simple PutItem request and 12 (instead of the expected 6) for the same Put in TransactWriteItems.
However, if I add 2 additional Put operations (PK + SK) into the transaction, each consumes 2 WCU as expected.

To sum up:

1 PutItem for item1(PK + SK) => reported: 1 WCU, expected: 1
1 Put in TransactWriteItems for item1(PK + SK) => reported: 4, expected: 2
1 Put in TransactWriteItems for item1(PK + SK + PK1 + SK1 + PK2 + SK2) => reported: 12, expected: 6
3 Put in TransactWriteItems for item1(PK + SK + PK1 + SK1 + PK2 + SK2) + item2(PK + SK) + item3(PK + SK) => reported: 16, expected: 10

In the AWS document, example 1 is a transaction with an Update operation and a Delete operation that yields 10 WCU. Can anyone explain this result?
https://docs.aws.amazon.com/cli/latest/reference/dynamodb/transact-write-items.html

Thank you in advance for your feedback.

I tried to reduce the operations to the essentials as decribed in the question to see where the x4 factor could come from but I didn’t succeed.