Linear memory rising monotonically across warm requests.
Trend, not threshold, and the distinction matters: wasm memory never shrinks, so an absolute
reading says nothing, while a monotonic rise across N warm requests is a leak. The lazy FS made
this reachable -- it converges on the union of every route ever served (up to ~52 MB against the
streaming mount's 39 MB) unless its LRU is holding.
Acts at the next quiet moment rather than the next request, because recycling the interpreter
mid-traffic trades a leak for a 4,019 ms boot.
Linear memory rising monotonically across warm requests.
Trend, not threshold, and the distinction matters: wasm memory never shrinks, so an absolute reading says nothing, while a monotonic rise across N warm requests is a leak. The lazy FS made this reachable -- it converges on the union of every route ever served (up to ~52 MB against the streaming mount's 39 MB) unless its LRU is holding.
Acts at the next quiet moment rather than the next request, because recycling the interpreter mid-traffic trades a leak for a 4,019 ms boot.