Parse a streaming reply again only from its last settled block
A reply that is still streaming is re-read from its last finished block instead of from the top, so long replies cost less to keep rendering.
While an answer streams in, Napdock used to parse the whole reply again on every update, so the work grew with the length of the reply. It now keeps everything before the last finished block and parses only the text after it.
In a measured 7 KB reply streamed 40 bytes at a time, each step went from 79 µs to 16 µs. Cases that can reach back across a blank line, such as lists that keep growing or link reference definitions, still fall back to a whole parse, so what you see does not change.
Evidence
No screenshot: Replies render exactly as before; the change is in how much work it takes to render them.