Pull queue performance proof
Linux x86_64; Node v26.13.1; same machine, sequential base then head.
Real server; fresh isolated database per side; fresh local repos per repetition.
Two-second delay applies only to standalone background status fetches, not git pull.
Five runs per N. Duration is request-to-response. Every result status=pulled; HEAD and file verified.

base N=0: [2084, 2161, 2089, 2116, 2171] ms; median 2116 ms; range 2084–2171 ms; spread 87 ms
base N=3: [2172, 2105, 2096, 2141, 2234] ms; median 2141 ms; range 2096–2234 ms; spread 138 ms
base N=10: [5344, 5286, 5319, 5242, 5335] ms; median 5319 ms; range 5242–5344 ms; spread 102 ms
head N=0: [2137, 2157, 2111, 2205, 2119] ms; median 2137 ms; range 2111–2205 ms; spread 94 ms
head N=3: [2166, 2145, 2107, 2140, 2139] ms; median 2140 ms; range 2107–2166 ms; spread 59 ms
head N=10: [359, 400, 379, 391, 371] ms; median 379 ms; range 359–400 ms; spread 41 ms