Thanks for the clarifications.
> it's not limited only to string parsing, but can be applied to any other series-like value and code (i.e. blocks) itself
Sounds somewhat similar to OMeta (http://www.tinlizzie.org/ometa/) which can pattern match streams of objects.
With contexts it sounds like almost every word can be late bound? Also it seems blocks and contexts are decoupled which means you could evaluate multiple blocks from different parts of your program in the same context, or give each a unique context, etc?