Palmer (1995) drew attention to a distinction between 'programmatic' and 'non-programmatic' aspects of criminal justice interventions. While a considerable amount of research has accumulated on the former, the latter by comparison remains under-researched. Nevertheless some advances have been made and the present article identifies the key components of this. Following analysis of the concepts forwarded by Palmer, a methodical comparison is made between his findings on programmatic elements and those of two other major groups of meta-analytic findings from this area. This provides further opportunity for testing of the Risk-Needs-Responsivity (RNR) model and an evaluation is offered of its current status in synthesizing relevant knowledge. A parallel set of comparisons is then drawn with respect to non-programmatic factors and the paper considers the level of agreement between separate reviews of that knowledge base. This directs attention to a number of instances of intervention 'failure' which can be explained by insufficient attention to non-programmatic issues. There is a generally high level of agreement between the three sets of data surveyed. There is not a complete consensus however, caused not by disagreement between datasets but by gaps in the types and range of evidence assembled. There are larger gaps remaining on non-programmatic factors and the nature and extent of those is described. There is also discussion of some objections and proposed alternatives to RNR, and to some conceptual confusions arising from them. The present state of knowledge on criminal justice interventions is a 'work in progress' but nevertheless can provide firm guidance on the design of such interventions, highlighting areas in which much further work is needed. Tables and references (Published Abstract)