> Juxtapose - if you have a group of records that aren't easy to tell they're > associated, make it so they are.
Sure, and lookup :-) I don't have a picture of your expected output yet. Some generai ideas: You can group the records per device into a chunk with something like 'joincont leading / / x15' and still be able to reconstruct the lines. If you use 'fanout' to have a copy of the original records. Rather than identify the good one,s you might be able to identify the bad ones. If nothing else you might split the streams by (PIM) etc tags and use a 'combine' stage to recognize a mix (e.g. all '-' or '+' stays like that, but a mix would result in '>'). If necessary use 'xlate' to pick elegant code points for the '+' and '-' signs. Something I like is to use 'predselect' to throw out or select records when you may have more than one trigger. So you'd have 'predselect' pass or reject the "chunked" records based on whether the detailed analysis gave a reason to reject. If nothing else, you can certainly write yourself a sipping pipeline in REXX and process each chunk on its own. Sir Rob the Plumber
