Split
Split nodes allow you to branch a flow probabilistically.
Split nodes allow you to branch to different paths based on relative probability. Split nodes can have one or more output handles, with each having a name and relative weighting out of 100%.
Weights
Each time a split node is executed an output is chosen based on the relative probability. The sum of all weights must be less than 100%, any remaining weights will be assigned to an internal “unspecified” pseudo weight. Any runs that select the unspecified option will end the flow, as it by definition has nothing attached to continue to.
Overwrite value
By default split nodes pass the original value through unmodified, allowing you to simply route control. A secondary use case of splits is as a weighted string generator. When checking the “overwrite output value” checkbox, the output value of the node will be the chosen output label value.
Overwriting the output value can be useful for selecting different filters with different likelihoods (like “hot”, “new”, “top”, etc) and then passing that to a subsequent API call.