With the same browser tabs open, Postgres CPU fell from — to — and the page header went from — to — per tab. Measured on the local stack with the market closed (Oct 9–10, 2026); market-hours fixes are simulated until Monday's check.
Median of 10 requests, same endpoints before and after; log scale. Endpoints under 10 ms are left out.
95th percentile of 10 requests while 3 extra page headers stream; log scale.
Old and new SQL on the same data in the API container, median of 9; identical rows. Log scale.
Old and new page code on real payloads (Node 24, and the browser for the pivots board); log scale.
Three report tabs and one board tab, market closed, 30-second window. "Tabs not reloaded" is the new server with the old page header still running in those tabs.
| Measure | Before | Tabs not reloaded | After | Change |
|---|
Read for 15 seconds each. The page header now asks for three fields instead of the whole dashboard.
| Stream | Before | After |
|---|
These paths only run while the market is open. The new numbers come from old and new code run on real data; the live column fills in after Monday's session.
| What | Before | After, simulated | After, live (Monday) | How to check |
|---|
Every change was required to leave outputs, ordering, errors and freshness unchanged.