In this blog post, Jack Vanlightly introduces an alternate scenario for TPL Dataflow pipeline with different requirements around latency and data loss. The new scenario is that the producer can be slowed down as much as required, and the pipeline must process every message. Latency is not an issue, and we need back-pressure that can flow from leaves all the way back to the producer. The article explains how we can perform a broadcast and keep the back-pressure that we want using TransformManyBlocks and a couple of wrapper classes for our DecodedMessage class. The post also shares code on Github for a better understanding.