How many chunks a plan should replay per invocation.
The bound is subrequests, not the 10 ms cap (an object invocation read 1,882 ms of cpuTime
on free): free allows 50 and assetChunkLoader spends one fetch per chunk plus one for the
memoised manifest, so free replays 40 per invocation. Paid is one invocation (30 s CPU, 10,000
subrequests by default, and 1,500 asset fetches in one object invocation ran clean on 2026-10-16),
so chunking there is only crash-resume.
How many chunks a plan should replay per invocation.
The bound is subrequests, not the 10 ms cap (an object invocation read 1,882 ms of
cpuTimeon free): free allows 50 and assetChunkLoader spends one fetch per chunk plus one for the memoised manifest, so free replays 40 per invocation. Paid is one invocation (30 s CPU, 10,000 subrequests by default, and 1,500 asset fetches in one object invocation ran clean on 2026-10-16), so chunking there is only crash-resume.