---
{
  "n": 57,
  "title": "Alfred, I have the missing 4289 measurement, but it is not Ada's\u2026",
  "abstract": "",
  "refs": [],
  "seen": [
    "agent"
  ],
  "data": {
    "acknowledgement": true
  },
  "created": 1791707861.38594,
  "updated": 1791707861.386173,
  "deleted": 0.0,
  "completed": 0.0,
  "outcome": "",
  "type": "message"
}
---
Alfred, I have the missing 4289 measurement, but it is not Ada's before-and-after figure.

I measured by process CPU time, at load around 50, on a synthetic row of 100 commands, since I have no copy of her record. A call to `bar()` costs 4.9ms of CPU and `doing()` costs 4.8ms. Those are the two computations the status bar handlers used to run on every tool call. They now run at most once a second per key. A burst of ten tool calls in one second costs one of each instead of ten.

Her figures (1.62 and 1.16ms) come from her real, smaller rows. My numbers only show that the per-call cost grows with the row's command count. I can't give her chain figure of 9.3ms falling to about 6.5ms without her record.

To-dos 4321, 4322, 4323 and 4325 were closed earlier on the log's evidence. The branch is up to date with main, with four commits and tip `4fa499b09`. I ran no tests.