layout B: loaded 304 docs into 19 namespaces (one per section) in 8.2s, 111469 embedding tokens
sum of the 19 section namespaces right after the last upsert: 304
sum after 0.1s of polling: 304

A) one namespace 'docs', no filter: 1 search per question
B) 19 namespaces, search all of them and merge by score: 19 searches per question

question                                                   A rank A calls   A ms B rank B calls    B ms A+filter B 1 ns
How do I limit how many pages a crawl fetches?               None       1    134   None      19    2495        2      2
How do I get notified when a crawl job finishes?                4       1    119      4      19    3049        1      1
How long are scrape results cached and how do I disable         1       1    160      1      19    2940        1      1
How do I cancel a running crawl job?                            1       1    123      1      19    2741        1      1
How do I check my organization's remaining balance throu        1       1    124      1      19    2978        1      1
How do I use WebCrawlerAPI from an MCP client like Claud        2       1    146      2      19    3041        2      2

reported read_units for 6 questions: A=6, B=114
query embedding tokens for 6 questions: A=93, B=1767
mean latency per question: A=134 ms, B=2874 ms (sequential)

cleanup of layout B: delete the 19 section namespaces
deleted 19 namespaces in 3.9s
namespaces left: ['docs', 'integrations', 'root']