
Gathers consecutive messages of any stream t into fixed-size lists [t] of size elements, emitting one list every step arrivals. The count-based stream batcher and categorical dual of flatten, which maps [t] back to t.
A per-frame feature stream is gathered into non-overlapping batches of a fixed count before a stage that consumes a list, so each batch is processed once with no element reused (tumbling, step equals size).
A scalar measurement stream is gathered into overlapping windows that advance one element at a time, so a downstream smoother sees every trailing run of size values (sliding, step less than size).
[t] of exactly size elements; a downstream expecting a bare value needs a flatten or an index between this node and the consumer.size messages have arrived (warmup); if the whole stream is shorter than size, no list is ever emitted.step is greater than size (hopping), the step minus size messages between consecutive windows are dropped and never appear in any list; this is lossy by design.step equal to 0 is treated as step equal to size (tumbling); step less than size overlaps windows so an element reappears in multiple lists; step greater than size leaves gaps.Automated release