[rwkeyes]

Problem and result

The comparison table shows 1, 2, 4 and 8 streams and never says which of those a reader will actually be. Without that
mapping the 22 % headline is abstract, and the equal result at eight streams reads as a weakness rather than as a
property of the test rig.

Related discussion

none — documentation only.

Approach

A short block under the table maps the rows to situations: one stream = one person or one agent loop, where the 22 %
applies and which is most desktop use; two to four = a few agents or tools on one box, 17 % and 11 %; eight = a queue
(batch job, evaluation harness, team), where both servers ran --batch 0 — one request at a time — so eight clients
mostly wait and the lines meet.

It also says what to do if you need concurrency: the engine takes --batch/--slots (2–8 in the usual case, or
--parallel in the Terminal), which is a different experiment from the one measured, linked to docs/BATCHING.md.

The "what it does not show" line now carries the real reason for the eight-stream tie — --batch 0, one request at a
time — instead of the vaguer "queue-bound".

Validation

Passed:

• The configuration claim is from the files, not the prose: both 0.2.13 configs carry --batch 0.
• The slot options are from docs/BATCHING.md ("By default Strata serves one request at a time … with parallel:
  N (the engine's --batch N, also spelled --slots N)") and docs/AI_SETUP.md (--parallel N, 1–8).
• The percentages are unchanged (22.1 / 17.1 / 10.8 / equal).
• Links and rendering: no broken relative link, the BATCHING.md link resolves, GitHub's Markdown API renders
  the new list.

Failed:

none observed.

Not run:

• No multi-slot measurement. The page explicitly says the --batch 2–8 numbers would be a different experiment.

Limits

• The mapping is about client streams against one server, which is what GuideLLM sends; a multi-slot configuration
  changes the picture and is out of scope here.