Paradox of Self-Amendment by Peter Suber The formalist theory that legal rules are validated only by other legal rules requires either an infinite regress, an exceptional rule that is validated either by itself or by an extra-legal source of authority such as acceptance. On the simpler reading of Hart, he has chosen the latter path (validation by an extra-legal source of authority), and to that extent differs from Ross only on the permissibility of exceptions and on the authority of the ultimate rule of recognition. But in fact Hart does not adhere to a one-exception inference model of legal change. First, he denies that the rule of recognition is the only rule authorized directly by acceptance, and he gives other examples. For instance, the holdings, and even the jurisdictional basis, of some cases of first impression, and cases in which judges must exercise discretion are not validated by prior law but only (and contingently) by the acceptance they inspire. On this point Hart says,[Note 3] The truth may be that, when courts settle previously unenvisioned questions concerning the most fundamental constitutional rules, they get their authority to decide them accepted after the questions have arisen and the decision has been given. Here all that succeeds is success....[I]t will often in retrospect be said, and may genuinely appear, that there always was an 'inherent' power in the courts to do what they have done. Yet this may be a pious fiction, if the only evidence for it is the success of what has been done....Here the power acquires authority ex post facto from success....The statement that the court always had an inherent power to rule in this way would surely only be a way of making the situation look more tidy than it really is. If acceptance is an authority for law, and if it is a fact of social practice more than normative ideality, then it may be created in a society in the wake of a judicial ruling; hence it may come on the scene after the law it authorizes and have its effect retroactively.[Note 4] Similarly, Hart continues, judges in the United States with powers of judicial review accept certain standards limiting their discretion to nullify statutes, but these standards exist only in judicial usage and are validated only by acceptance.[Note 5] It is vital to an acceptance theory whether acceptance authorizes only one rule (the rule of recognition) or can authorize any rule. Hart holds the latter, although commentators who disregard the quoted passage have read him the other way. To avoid exegetical quarrels, however, I will call the acceptance theory I will most often contrast with Ross's inference model the direct acceptance theory. It holds that acceptance can authorize any rule of law directly, even if it rarely does so. Acceptance does not authorize only some master rule that validates other laws indirectly. Of course, for the purposes of this essay, it does not matter whether we agree or disagree that Hart himself held the direct acceptance theory. However I will use Hart's discussion of acceptance to show what issues must be raised in making such a theory comprehensive and plausible, since no other acceptance theorist has done nearly so thorough or thoughtful a job. The rule of recognition does not determine law with the certainty and finality of logical rules. That sort of formalism is repugnant to Hart's entire theory.[Note 6] The open texture or partial indeterminacy of legal rules means that "subsumption and the drawing of a syllogistic conclusion no longer characterize the nerve of the reasoning involved in determining what is the right thing [for a court] to do."[Note 7] In his later essay Hart directly addressed Ross's view that legal change is transacted by logical inferences, and replied,[Note 8] To this it might be objected that the exercise of legislative powers to introduce new norms is not a deductive inference and it is not clear how this logical principle applies to a legislative act. However, while Hart clearly intended acceptance to validate more than just the ultimate rule of recognition, he also intended that most rules be validated by conformity to the rules of recognition. The direct acceptance theory allows the acceptance of the people and the practice and usage of the officials to validate any rule not otherwise validated by the rule of recognition, and to invalidate any rule otherwise validated by the rule of recognition. Because acceptance is logically prior to any positive law, in conflicts between informal acceptance and formal recognition acceptance overrules, or at least amends, the rule of recognition. This supplements Hart's position insofar as he did not address the problem of conflicts between actual acceptance and literal recognition. There are defects in any inference model that admits no exceptions for self-validating rules or some alegal source of authority such as acceptance. The inference model allows formal logic to rule law, and to invalidate whatever fails the criteria of logical validity even while satisfying the courts, the people, and all legal criteria of legal validity. This is one of its most salient weaknesses. By contrast the acceptance model makes the conformity of law to logic contingent for each legal system at each moment; certainly the people and officials can accept a legal act that can only be described with self-contradiction, such as self-amendment, even if they will not accept other types of contradiction such as simultaneously binding inconsistent commands (see Section 21.B). The inference model conceives legal rules to be, as far as possible, as clean and determinate, and as immune to fading away and shading off as the rules of mathematics. Legal rules do differ from mathematical rules, in their origin at certain moments in time by human will, their mutability according to certain procedures and human will, and the spatial limitations on their validity. Sometimes reading Ross one detects a note of wistful regret that legal rules have material content and sully themselves with the complexities, contingencies, and uncertainties of human life. But once enacted, they have single meanings, even if judges fail to discover them, and they cannot validly contradict any other valid law in any way without one of them giving way according to the canons of interpretation and rules of implied repeal. The formal rules of formal deductive inference apply absolutely to all legal rules; whatever can be validly inferred from a valid law is valid law; whatever cannot be validly inferred from any valid law or laws cannot become valid law. This leads to one of the most radical absurdities of the exceptionless inference model. All change of law would be unlawful. Assuming we had a body of valid law to begin with, we could make no new law nor amend any existing law except by inference from the premises already embodied in 30