EXPLAIN BUFFERS and I/O timing accounting questions

Hello,

I ran across an EXPLAIN plan and had some questions about some of its
details. The BUFFERS docs say

>The number of blocks shown for an upper-level node includes those used by
all its child nodes.

I initially assumed this would be cumulative, but I realized it's probably
not because some of the blocks affected by each child will actually
overlap. But this particular plan has a Shared Hit Blocks at the root (an
Aggregate) that is smaller than some of its children (three ModifyTables
and a CTE Scan). This seems to contradict the documentation (since if
children overlap fully in their buffers usage, the parent should still have
a cost equal to the costliest child)--any idea what's up? I can send the
whole plan (attached? inline? it's ~15kb) if that helps.

I also noticed the I/O Read Time (from track_io_timing) of two children in
this plan is equal to the I/O Read Time in the root. Is I/O time
potentially fully parallelized across children? There are no parallel
workers according to the plan, so I'm surprised at this and would like to
understand better.

Also, a tangential question: why is the top-level structure of a JSON plan
an array? I've only ever seen one root node with a Plan key there.

Thanks,
Maciek