I'm somewhat skeptical of the analysis when it revised the share of criminal activity in 2019 from 1.1% to 2.1%--in other words, more than doubling that value. That revision is on the basis of "the identification of more addresses associated with criminal activity that were active in 2019."

Your citation that it's "only 0.34%" is a somewhat dishonest distillation, because the report itself advises that the number is only for known criminal activity, and that the number is expected to rise over time as well.

What I would instead like to see, and that this report does not illuminate at all, is a breakdown of how all transactions looks. How many transactions are related to cryptocurrency-specific transactions, such as mixing coins or moving between hot wallets and cold wallets? How many transactions are related to speculation (e.g., trading one cryptocurrency for another)? How many transactions are associated with known entities, e.g. the Bitcoin address of an online (non-darknet) marketplace? How many transactions are unclassified?

Even if 90% of illicit financing transactions are being missed by Chainalysis, the original claim is that "many are used [...] mainly for illicit financing" is outright false.

By 'transactions sense' I think Yellen means using cryptocurrency to buy and sell goods and services. The Chainalysis figure seems to include people simply swapping USD for Bitcoin. So in that sense Yellen may be correct

She's using it as a pretext for when they eventually ban or regulate crypto to death, not to genuinely inform anyone.

I sincerely doubt she cares what the data says in this case. She's building a narrative.