Federated DPFTRL and Adaptive Clipping noise injection.

Dear developers of the privacy framework. I was checking the implementation of quantile_adaptive_clip_tree_query.py

I can see that a new tree is required to inject noise after aggregating the norm bit as the traditional QuantileEstimatorQuery
is wrapped from the TreeQuantileEstimatorQuery

In this paper Federated Learning of Gboard Language Models with Differential Privacy I see the paragraph highlighted in the image below.
Question: Does this mean that I can just keep the logic of restarting the tree and then updating the clipping norm of the DPFTRL skipping the second tree that injects noise on norm bits and just replace the whole noise of Federated DP-FTRL with Adaptive clipping setup with just a Tree based Gaussian noise with Z as stated in the figure? I am confused here.

Thanks you in advance.

Read tensorflow_privacy/privacy/dp_query/quantile_adaptive_clip_tree_query.py and quantile_estimator_query.py, then compare their tree and norm-bit handling with the cited Federated Learning of Gboard Language Models with Differential Privacy paper. Trace whether the proposed DPFTRL and adaptive-clipping noise arrangement is supported by the current implementation; done means a maintainer-confirmed design decision or a clearly scoped implementation plan.