What are the requirements of an immune effector response that can rid a pathogen without debilitating the host? This is the question under discussion here. I have suggested that two sets of decisions are required, namely Decision 1, the sorting of the paratopic (combining site) repertoire, and Decision 2, the regulation of the magnitude and class of the effector response. In order to discuss Dembic’s integrity model as a solution to this question, I will recast it in my two-decision framework. I view his model as dealing with Decision 2, the regulation of the magnitude and class of the effector response. It is not a model of Decision 1, the sorting of the repertoire. Dembic questions the need for two decisions by pointing out correctly that ‘magnitude’ is also a property of the sorting of the repertoire. If so, he argues, Decision 2 is all that is required to achieve an effector response that would rid harmful nonself (NS) without attacking the host [self (S)]. Decision 1, the sorting of the repertoire, in his view, becomes a gratuitous assumption because evolution is, after all, selecting at the level of the effector output that must be specific enough to rid NS without debilitating S. The precise distinction between our two positions becomes the answer to ‘What does it take to accomplish this?’ Using, for illustration, the immune system of mammals, antigen-responsive cells [referred to here as initial-state cells (i-cells)] are generated continuously throughout life. These i-cells express anti-S or anti-NS specificities. Decision 1 purges anti-S leaving the residue, anti-NS, as the functional repertoire. Activation of an i-cell whether it be an anti-S or an anti-NS, puts it under the control of Decision 2, the determination of effector class. ‘Magnitude’, at the level of Decision 2, makes no distinction between S and NS. Decision 2 is optimized by evolutionary selection to rid the target. Once a cell is activated, it follows this optimization pathway for ridding whether it be an anti-S or an anti-NS. Now let us explore what it would take to solve the S–NS discrimination in the absence of Decision 1, at the level of Decision 2? Consider two i-cells interacting with proximal antigens, an anti-NS and an anti-S. What germline-selected signal would distinguish them? Because recognition of S cannot be selected in the germline, the ‘innate’ system is unresponsive to the S of the species. Whether it is agreed that a significant portion of the NS-antigenic universe is also not recognized by the ‘innate’ defence system, no signal from that source could distinguish these two i-cells. If the NSantigen is harmful, both cells would be driven to effectors, a recipe for autoimmunity because the effector output ‘inflammation’ is not antigen specific. If the two cells are reactive, respectively, with an Sand an NS-epitope linked on an antigen, there would clearly be no way to distinguish them. Any associated property of that antigen, danger, pathogenicity (PRR), harm, integrity, etc. is also a property affecting responsiveness of i-cells recognizing epitopes linked on that antigen. Roughly 10% of NSantigens share epitopes with S-antigens. In the absence of Decision 1, the sorting of paratopes, any signal at the level of Decision 2 will either initiate an effector response to S or result in unresponsiveness to NS, both lethal in consequence. Does the antigen-presenting cells (APCs) solve the problem? Under the presently accepted view, the APC converts by processing every NS-antigen into one that shares ‘epitopes’ with S because the APC presents both Sand NSpeptides. As the APC is postulated under ‘costimulation’ models to be either tolerigenic-only or inductive-only dependent on a signal derived from an associated property of the antigen, it cannot discriminate between the two i-cells, an anti-NS and an anti-S, interacting with it. What would it take to make a S–NS discrimination uniquely at the level of Decision 2? The regulation would have to be such that the response to S would be in an ineffective effector class and to NS in an effective effector class. Unfortunately, there is no way to do this, because the selection pressure on Decision 2 is to optimize the response so that it is of appropriate magnitude and always in an effective class (i.e. one that rids the target). Consequently, Decision 1, the sorting of the repertoire, is obligatory for a sufficiently specific anti-NS effector response. If germline-encoded recognition revealed all that is harmful, protected all that is useful and neglected the remainder, what would be the selective advantage of a somatically generated, large and random immune repertoire? A reasonable This work was supported by a grant (RR07716) from the National Center for Research Resources at the National Institutes of Health. DISCUSSION FORUM