How many hosts to probe at once, PER DOMAIN. Separate from the source concurrency, which throttles free third-party APIs that need to be treated gently. When multiple domains run concurrently (see "Domains enumerated at once"), each gets its own probe pool of this size, so the total probing parallelism across the whole run is this value multiplied by "Domains enumerated at once". The maximum (30) is capped well below what the field alone might suggest, and the reason is memory rather than rate-limiting: at 30 combined with 10 concurrent domains the run holds 300 probes in flight, each peaking at roughly 0.66 MB of response body, favicon and decoded-text buffers (~198 MB), plus about 43 MB of TLS connection buffers for the 650 pooled connections that configuration opens, on top of a ~95 MB baseline — roughly 336 MB of the container's 512 MB, leaving about a third of it free for estimation error. At the defaults (25 x 3 domains) the same arithmetic lands near 155 MB. Note that this figure covers the probing buffers only: the memory held by the results themselves scales with how many subdomains you find and is governed separately by "Max subdomains per domain", where that arithmetic is written out.
Minimum:1
Maximum:30
Default:25
