B-79MV-modeswitch2kms-is-116-9-ms-against-a-100-ms-budget-all-of-it
The miss decomposes entirely to source -> read: live -> source 14.5 ms, read -> live 8.9 ms,
source -> read 116.9 ms of a 116.9 ms total.
That is why the next reader should start in the renderer, not at the mode boundary. The whole miss
is mounting MdViewer and running the unified/rehype pipeline over the 2 000-line note for real — a
genuine render, not discarded work.
Measured on a production build in assert mode at load 9.4/11.8/10.1. Pre-existing: 117.8 ms on
main before #1722, unchanged by it. Budget not
loosened.