Footnotes to the Text 1 I am greatly indebted to generous (in many cases extraordinarily generous) written comments from Matt Bedke, Selim Berker, Michael Bratman, John Broome, John Brunero, Fabrizio Cariani, Sinan Dogramaci, Aaron Elliot, Charlie Kurth, Ryan Millsap, Julia Nefsky, Anthony Price, Joseph Raz, Simon Rippon, Adam Ross, Mark Schroeder, Kieran Setiya, Jacob Stegenga, Mike Titelbaum, Peter Vranas, Jonathan Way, and two anonymous reviewers for Philosopher’s Imprint, as well as to comments from audiences at the Department of Philosophy at Texas Tech University, the Practical Philosophy Workshop at the University of Chicago, the Department of Philosophy at the University of California at San Diego, the Department of Philosophy at the University of Texas, the Department of Philosophy at Harvard University, the Workshop on Reasons and Rationality at the London School of Economics, and the 2011 Central Meeting of the APA. 2 Nagel (1970: 34) observes that it is a “perfectly general property” of reasons for action “that they transmit their influence over the relation between ends and means.” 3 As Kant claimed in the Groundwork, and as the idea has been developed (in markedly different ways) by Korsgaard (1997), Bratman (1987), and Broome (2002). 4 This formulation is agnostic on whether reasons for ends derive from our desiring those ends, or from the relation of those ends to things of independent value. However, desire-based theorists may deny, against Hubin (1999), that their theory is a combination of a principle of instrumental transmission and the principle that reasons for ends are provided by desires. Instead, they may say, there is just one principle, a principle of, if you will, instrumental transmutation: if one desires the end, then one has reason to take the means. See the discussion of General Production, in section 8, for a doubt about this. 5 See note 15 and the preceding text. 6 See, for example, Bedke (2009), Bratman (2009), Darwall (1983: 16, 46–48 and 2006), Hubin (1999), Raz (2005a and 2005b), Kolodny (2007 and 2008b), Price (2008: 7, 138–139), Schroeder (2009), Setiya (2007), Vranas (2011) (who distinguishes it from the kinds of support relevant to the validity of imperative inference), and Way (2010). Passages in Scanlon (2004) and Wallace (2006) assume or grant its correctness. 7 “If you should do E, all things considered, and doing M is a necessary means to doing E, you should do M, all things considered” (Setiya 2007: 660). “If X objectively ought to do A, and to do A X must do B, it follows that X objectively ought to do B” (Schroeder 2009: 239). The principle is often implicitly invoked: “because you promised to do A and cannot do A without doing B, you ought to do B” (Korsgaard 2009b: 38). Ought Necessity is also entailed by certain deontic logics and semantics for ‘ought.’ In broad outline, and expressed in the idiom of this paper, these views hold that “One ought to X” is true iff one X’s at every possible outcome that is “ideal.” (See Wedgwood 2006 for a particularly forceful defense of such views.) The argument in section 3 against Ought Necessity is, in effect, one objection to such views. (Interestingly, Horty 2001 sec. 3.4.2 presents a similar argument against such views, although not against Ought Necessity as such.) Moreover, since such views transmit ‘ought’s not only to necessary means, but also to necessary conditions, they may be subject to further objections. For instance, there’s Ross’s (1941) paradox: that “One ought (to post the letter)” entails “One ought (to post the letter or to burn it),” an entailment whose strangeness, as Cariani (2009) shows, stubbornly resists pragmatic explanation. There’s also our telegraphing boxer of section 1. If in every possible outcome in which he throws the punch he telegraphs it, then “He ought to throw it” entails, counterintuitively, “He ought to telegraph it.” One might think that the “end-relative” account of ‘ought’ presented by Finlay (2009) and (2010) is also committed to Ought Necessity, since that account implies that, whatever else one ought to do in a given context, one ought to take necessary means to the “end” specified by that context. However, the E in Ought Necessity need not be this contextually specified end itself. If the contextually specified end is, say, maximizing expected value (see Finlay 2009: 326 n. 26), then it might be the case for Unlucky, discussed in section 3, that he ought to save the antique (=E), but not the case that he ought to take the necessary means of taking the taxi. 8 “Reasons for me to make something my end are, owing to the hypothetical imperative, equally reasons for me to take the necessary means to it” (Darwall 1983: 16). “If one has conclusive reason to believe that one will E only if one Fs, then one has reason to F that is at least as strong as one’s reason to E” (Kolodny 2007: 251). “If X has an objective reason to do A and to do A X must do B, then X has an objective reason to do B of equal weight to X’s objective reason to do A” (Schroeder 2009: 245). “If R is a practical reason in favor of X, X is attainable by the agent, and M is a necessary means to or necessary constitutive element of X, then R is a practical reason in favor of M” (Bratman 2009: 424). “If you have a reason to do A and doing B is a necessary means to doing A, you have a reason to do B which is at least as strong as your reason to do A” (Way 2010: 225). See also Millsap (ms a). 9 Way (2010: 224) and Bedke (2009: 687 n. 10) endorse Weak Sufficiency. The “(M/E)” principle of Dreier (1997: 38) has a similar structure. The “logic of satisfactoriness” of Kenny (1966) validates an analogue of Ought Sufficiency for “fiats.” 10 Anscombe (1957: 36) anticipates these problems when she observes that the intention expressed by, “I do P, so that Q,” can be “contradicted” either by saying, “But Q won’t happen, even if you do P” or by saying, “But it will happen whether you do P or not.” In my terms, P can fail to inherit the reason to Q because P fails to “probabilize” Q—Q won’t happen even if you do P—or because P is “superfluous” with respect to Q—roughly, Q will happen whether or not you do P. 11 Alternatively, one might suggest that the weight of the reason depends on: (1) the difference of the probability, conditional on one’s M-ing, that one E’s less the probability that one E’s; or: (2) the difference of the probability, conditional on one’s M-ing, that one E’s less the probability, conditional on one’s not M-ing, that one E’s; or: (3) the difference of the probability, conditional on one’s E-ing, that one M’s less the probability, conditional on one’s not-E-ing, that one M’s (i.e., (2), with E and M changing places); or: (4) the difference of the probability, conditional on one’s M-ing, that one E’s less the probability, conditional on the worst one could do with respect to E-ing, that one E’s. I reject (1) on the grounds that it implies that the probability of one’s M-ing can affect one’s reason to M. At the limit, if one is sure to M, then one has no reason to M, since the conditional probability equals the corresponding unconditional probability. (2) also gives the wrong results. Suppose the only relevant end is keeping the patient alive. Giving drug A alone improves the patient’s chances of living by 98 percentage points, and giving drug B in addition improves them by a further percentage point. Intuitively, the doctor has no less reason to give both A and B, which has a 99% chance exactly of keeping the patient alive, than to give the patient at least A, which has at most a 99% chance. However, giving both A and B rather than not giving both A and B (assuming the doctor is very likely to give A but not B, if he does not give A and B) raises the probability by something approaching only one percentage point, whereas giving at least A rather than not giving at least A raises the probability by something approaching 98 percentage points. So the current proposal implies, counterintuitively, that the doctor has less reason to give both A and B than to give at least A. Stegenga (2012) proposes option (3), along with a variant structurally analogous to a ratio of likelihood functions. While Stegenga’s application of confirmation theory to these issues is welcome, I have doubts that his proposal suits the practical context. We don’t seem to practically deliberate in terms anything like (3). And some of the implications that Stegenga suggests intuitively support (3) seem to me to tell against it. For example, no (as opposed to little) reason to buy a lottery ticket transmits from reason to get rich. And, to take an extreme version of a case that he considers, no reason transmits to studying from reason to pass the exam, if one is sure to study, as was the case with (1). Bedke (ms.) pursues option (4). It has an appealing simplicity, but I worry that it sacrifices relevant structure. To anticipate (and to vary slightly) an example from section 4, suppose that the end is my entering, and no matter what, either I enter by the front or I enter by the back. Then the worst I can do by way of entering is just as good as the best I can do. (4) implies, counterintuitively to my mind, that I have no reason to enter by the front. (Note that (1)–(3) also have this implication, with undefined probabilities in the case of (3).) There seems to me an intuitive difference between my reason to enter by the front in this case and my reason, say, to immobilize myself by tying my ankle to a stake in the front lawn in an altered case in which that is possible. If you ask me, “Why did you do that? What reason had you to?” I could give an intelligible answer in the first case, but not in the second. However, I suspect that my approach and Bedke’s (4) can be made to deliver the same ‘ought’-judgments, at least relative to a given Reasons-Ought principle of the kind introduced in section 3. Moreover, the choice between these approaches does not affect the criticism of the Necessity and Sufficiency principles in sections 2, 3, and 5, or the implications discussed in sections 7 and 8. 12 See Hall (2004) for examples of preventers of preventers and for an argument that, on one conception of causation, they are not causes. It might seem that “helping to bring about” as I am understanding it here is just the other conception of cause that Hall distinguishes: roughly, counterfactual dependence among distinct items. But this would not cover cases of, say, constitutive means, which are not suitably distinct from the end. 13 Bedke (ms.) and Stegenga (2012) consider cases like the telegraphing boxer, but resist adding a “causal” condition. I believe that the counterexamples that, according to Bedke, tell against a “causal” condition can be addressed by the point that preventing preventers can help to bring about and by the nonsuperfluity condition of section 4. Stegenga suggests that the “causal” condition is not necessary, so long as we specify the relevant “population” appropriately: in this case, not this boxer, but all boxers. But this seems to me misguided. Why shouldn’t our advice by tailored to this boxer given what we know about his idiosyncracies? 14 Insofar as we use the epistemic framework, we may need some response to Newcomb’s problem. Adding the phrase “helps to bring about” is, in itself, no response, since both oneboxing and two-boxing help to bring about the relevant payoffs. One response, which would not radically affect the discussion, would be to screen off whatever information one’s M-ing gives about the probability of factors that (i) bear on whether one’s M-ing helps to bring about one’s Eing, but that (ii) one’s M-ing itself does not help to bring about. One method for doing this would be to take a partition of the possible states, Si, that one’s M-ing does not help to bring about. (These are not quite states of “nature,” since they may depend on other things, besides Ming, that one may independently do. We are not assuming that M-ing is an “exhaustive” action in the sense described in section 3.) Then we sum, over i, the products of the probability, unconditional on one’s M-ing, of Si, and the probability, conditional on M-ing in Si, that one’s M- ing (or a part thereof) helps to bring it about that one E’s in Si. (Note that this formula still needs the phrase “helps to bring it about that.” Without it, the formula might deliver a positive probability, on telegraphing, of connecting, and so imply, wrongly, that there is reason to telegraph.) Other responses to Newcomb’s problem, however, might more radically affect our discussion. 15 As for the question of how context determines the relevant body of information, I favor the view of Kolodny and MacFarlane (ms.), that the truth of an occurrence of an ‘ought’ or ‘reason’ sentence depends on the information relevant at a context of assessment: to a first approximation, the information of the person—agent, advisor, onlooker, etc.—who is considering the occurrence. 16 See Belnap, Perloff, Xu (2001) and Horty (2001). On such a framework, “outcomes” would be more naturally understood in terms of histories, instead of worlds. If agents are conceived as choosing “freely” in a strong sense, probabilities will be unavailable whenever the occurrence of the end depends some further “free” choice. Nevertheless, “sure-thing” reasoning may still be possible, as Horty shows. 17 Raz (2005a: 7) makes essentially this point with the case of buying a ticket to visit one’s grandmother on Easter Island when the airline workers are on strike. 18 One might also naturally describe the cost as a “reason not to taxi” or a “reason against taxi- ing.” For simplicity, I assume that reasons not to X and reasons against X-ing are just equivalent to reasons to refrain from X-ing. This assumption may well be false, but I don’t think it affects the argument in any substantive way. See Bedke (ms.) for discussion. 19 Insofar as Schroeder (2009: 246) takes his “General Reason Transmit” (“If X has… reason to do A and X’s doing B would facilitate her doing A, then X has… reason to do B of weight at least proportional to X’s… reason to do A, and to how well her doing B would facilitate her doing A”) to entail Strong Necessity, it also seems to face this problem. 20 Way (2010: 225 n. 32) reports an objection to Ought Necessity along these lines from John Broome. 21 This is similar to the principle “Negative” in Millsap (ms. c). 22 This might be a special case of General Production of section 8. 23 Compare the “Combo” example of Millsap (ms. c). 24 One might reply that the instrumental reason derives directly from the value of the state of affairs of the antique’s being saved, instead of from any reason for the end of one’s saving it. To anticipate just one of two problems with this reply, which will be discussed in section 8, there may not be any such valuable state of affairs. It might be the case, to alter the scenario, that the reason to save the antique is not that one’s saving it would itself be a valuable state of affairs, but instead that one’s saving it would honor something, or someone, of value, e.g., the memory of Unlucky’s grandmother, who schlepped the antique from the old country. (Hence the awkward example of saving an antique.) 25 Independent reasons for accepting Strong Availability might seem to be provided by the following puzzle. Suppose ten miners are trapped in one of two shafts, we know not which, with the floodwaters rising. If we block shaft A, all ten will survive if they are in A, whereas all will die if they are in B. If we block shaft B, the reverse. If we block neither shaft, nine will survive and one will die wherever they are. It seems that we ought to block neither, which will save nine wherever they are. But why ought we not instead block whichever shaft the miners are in? After all, that is sure to save all ten, wherever they are. As Kolodny and McFarlane (2010) observe, something like Strong Availability would give us an answer. Because there is nothing under our control, that we can knowingly do, that is sure to bring it about that we block the shaft the miners are in. So doing so is not available to us. However, we do not need Strong Availability to give us the result that we ought to block neither shaft, at least relative to the set of default alternatives that includes blocking neither, if we accept Coeval Default. Saving all ten does not belong to that set, since it is not among the things under our control, that we can knowingly do, at the time of blocking neither shaft. 26 Alternatively, there is a proposal suggested by remarks of Kiesewetter (ms.) and Simon Rippon in correspondence: X is available to one only if: if one does everything one ought to do up until the time, t, when one would X, then at t knowingly X-ing is fully under one’s control. In many cases, this collapses into Super-Strong Availability. For example, in the case of Buridan’s ass, before the ass sets out, neither eating from the left bale, nor eating from the right bale is available to him (at least if retracing steps is not allowed). In heading for the left bale, the ass does everything that he ought to do, but he cannot then eat from the right bale. Same if he heads for the right bale. In any event, in the Mountaineer case, below, stepping while not looking down still counts as available by this proposal. 27 It may sound better to conclude, in the third person, “Procrastinate ought to accept.” If so, I suspect, it is because it is easier to hear the third-person claim as being not in the mode of advice or deliberation, but instead in the mode of criticism or assessment: as equivalent to, “If Procrastinate were ideal, he would accept.” 28 Here is where I feel the force of the complaint of Kiesewetter ms. that if two pieces of advice cannot be jointly satisfied, then one of the pieces must be bad advice. 29 It is true that when one enters by the front, one “also” (enters by the back or by the front) in some determinate, by-the-front, way. However, in violation of clause (iii), it is not true that if one did not enter by the front, one would still (enter by the back or by the front) in that determinate, by-the-front, way. Of course, this raises a question about how “ways” should be individuated. But I see no reason to doubt that they can be individuated so as to lead to plausible results. 30 More precisely: M-ing is an excessively costly means to one’s E-ing iff there is some M* such that: (i) it is possible for one to M* without M-ing, (ii) the probability, conditional on one’s M*-ing without M-ing, that one’s M*-ing, or some part of one’s M*-ing, helps to bring about one’s E-ing nonsuperfluously is at least as high as the probability, conditional on one’s M-ing, that one’s M-ing, or some part of one’s M-ing, helps to bring about E-ing nonsuperfluously (which is positive), (iii) there is stronger reason to refrain from M-ing than from M*-ing. 31 Here’s another example, due to Aaron Bronfman. Suppose on my way home, I could leave the route to help stop a boulder from destroying my house, then returning to my route and finishing the trip home. However, there are already enough people there to stop the boulder; my additional help won’t make it any more likely that my house is saved. Taking the detour is not superfluous. I do have reason to do it: namely, that it will move my body to locations from which I will then be likely to proceed home. But it is presumably excessively costly. Just staying on route is at least as good as means to getting me home and takes less effort. 32 Broome (ms.) argues against Weak Sufficiency with an example that, I believe, exploits superfluity. An itchy donkey has the end of scratching its itch. If it walks to the scratching-post, which it hates to do, that will help to bring it about that it scratches its itch. However, the donkey can also easily, effortlessly scratch the itch with its hind leg. In that case, Broome suggests, it has no reason to walk to the scratching-post. But this judgment seems so natural, I think, only because we implicitly assume that, because the donkey can so easily scratch its itch with its hind leg, it is almost sure to do so. In that case, its walking to the scratching-post, in addition to its hind-legging it, is superfluous (as well as excessively costly). Alternatively, we might imagine that the donkey will hind-leg it only if it does not walk. In that case, walking is not superfluous (but still excessively costly). It has just as much (itch-alleviation) reason to walk as to hind-leg it (although more reason to refrain from walking than from hind-legging it). 33 For rather subtle reasons, Joseph Raz’s: Facilitative Principle: When we have an undefeated reason to take an action, we have reason to perform any one (but only one) of the possible (for us) alternative plans that facilitate its performance (Raz 2005a: 6) also has trouble with superfluity. This is so, despite the fact that his “but only one” restriction seems intended to ban transmission to superfluous means. So understood, the restriction would presumably say (in the idiom of this paper) that if in all possible outcomes, one takes some facilitating plan A, then no reason is transmitted to taking a second facilitating plan B. (Alternatively, the restriction might be understood as a ban not on superfluous means, but instead on excessively costly means. It would then say that if A and B are each facilitating plans, no reason is transmitted to the facilitating plan consisting of the conjunction of A and B, at least if that conjunction is more costly than either alone. But this restriction would be misplaced, since, as Raz 2005b: 2–4 rightly stresses, reason transmits to excessively costly means.) The problem is that this restriction bans more than merely superfluous means. The condition might be met even though B is not superfluous. B might be a failsafe, in case A doesn’t come to fruition. Intuitively, in such a case, it seems that at least some reason to take B is transmitted from the end. So, unless more is said about what a facilitating plan is, the “but only one” restriction seems too strong. One might reply that by “facilitating plan,” Raz means “sufficient means.” Taking a second sufficient means, when one has already taken one sufficient means, is always superfluous. However, Raz writes that the Facilitative Principle is only “roughly speaking” about sufficient means to an end or “what we may crudely and inaccurately describe as means sufficient for its realization” (2005b: 9). Alternatively, one might reply that by “facilitating plan,” Raz means “unimprovable means.” Unimprovable means need not be sufficient; they may fail to achieve the end. But they fail to achieve the end only when this couldn’t have been helped: only when taking additional means would not have achieved it either. More rigorously, a means, M-ing, to E is unimprovable iff there is no means M* such that if one M’s in every possible outcome, there is some possible outcome which one’s M*-ing helps to bring about one’s E-ing in a nonsuperfluous way (and the way in which one M’s at that outcome is not the way in which one M*’s at that outcome). This would validate the “but only one” restriction. If one is sure to take an unimprovable means, then any further (putative) means are (at best) superfluous at every possible outcome. The problem is that if facilitative plans are identified with unimprovable means, then the Facilitative Principle transmits reason to facilitating plans that are superfluous. Suppose that, in every possible outcome, one takes some means, M*-ing, that is improvable and so not a facilitating plan. M*ing makes taking some further means, M-ing, which is unimprovable and so a facilitating plan, superfluous in every possible outcome. If one additionally M’s, one will still be taking only one facilitative plan overall. So the Facilitative Principle allows reason to transmit to additionally Ming, even though it is superfluous in every possible outcome. (For an illustration, suppose that in every possible outcome, one gives the patient Drug A (our M*). Drug A will cure the patient if the patient has Disease 1, as is likely. However, administering Drug A is not unimprovable, and so not a facilitating plan. If one additionally gives the patient Drug B, this will not interfere with the effect of Drug A and will cure the patient on the off chance that the patient has Disease 2. If one additionally gives the patient Drug C (our M), this will combine with Drug A to make Drug D, which has exactly the same effect as Drug A. However, giving Drug C is unimprovable, let us suppose, since it precludes Drug B from having any effect. So giving Drug C is a facilitating plan. According to the Facilitative Principle, under the present interpretation, one has reason to give Drug C, in addition to A. But this seems wrong, since giving Drug C is superfluous in every possible outcome.) 34 This is in contrast to the Weak principles, Bedke’s (2009: 678) “Instrumental Principle” (“One has reason to take the means to what one has ultimate reason to do”), and Raz’s Facilitative Principle. The Strong principles and Schroeder’s General Reason Transmit are more informative. However, while Schroeder says that the strength of the reason depends on “how well” the means facilitates the end, he leaves it largely open how “how well” is to be understood. General Transmission might be seen as offering one understanding. 35 To illustrate, suppose an aimless teenager has not adopted any plan that contains taking the SAT as a step. Nevertheless, a parent may know that the teenager has reason to take the SAT exam as a means to a college education (perhaps partly because the parent knows that, in time, the teenager will begin to take his future seriously and take other steps, such as filling out application forms). Taking the SAT is neither a necessary means (since some schools permit the ACT instead), nor a sufficient means (since other partial means, such as submitting the application are necessary). By contrast, Bedke’s Instrumental Principle and Schroeder’s General Reason Transmit apply to such means. Raz’s Facilitative Principle, however, seems not to. Taking the SAT does not fit the description of a “facilitating plan”: “Facilitating steps can come in ordered sequences, each constituting one way of bringing us to the point where we can take the action we have reason to take. We could call each a (possible) plan of how to get to a point at which we can take that action” (2005a: 5). (Moreover, if taking the SAT were a facilitating plan, then other partial means, such as submitting the application, would also count as facilitating plans to that end. But then the “but only one” restriction of the Facilitative Principle would imply, incorrectly, that the teenager does not both have reason to take the SAT and have reason to mail the application.) Nor is it covered by Raz’s supplemental claim that “we have only conditional reason to take [steps within a plan], the condition being that we have adopted and are pursuing the plan” (2005a: 6). The teenager has not adopted, and is not pursuing, any plan that has taking the SAT as a step. To be sure, Raz observes that reason can transmit to such means, which “keep options open” for not yet adopted plans (2005b: 8). The problem, to be pedantic, is that letter of his formulations doesn’t actually allow for this. Furthermore, General Transmission applies to means, such as taking the left bale as a means to taking either the left or the right bale, that cannot be steps even within adopted facilitative plans, since such means do not put the agent in a position to perform the action, but instead are ways of performing the action. Again, the point is pedantic, since Raz (2005a: 13 n. 18) clearly recognizes transmission to such “means,” governed by the more basic and general principle that, as I recount in section 7, he uses to argue for Weak Detachment. 36 This is more or less Bedke’s solution. Millsap (ms. a) observes that caution about what we count as necessary or sufficient means will solve many versions of the problem. It is also worth noting that, if we add the Intransitivity Proviso, then we need not worry about the problem of “explosion”—that if there is reason for something, then there is reason for anything—that Millsap (ms. b) notes is a consequence of certain transmission principles, and that similarly appears to worry Raz (2005a: 12 n. 17). 37 For discussion of silencing, see McDowell (1998a, b, and c); and Price (2008: 12 n. 9, 184 n. 72). Setiya (2005) raises this silencing objection against something like Weak Necessity, whereas Broome (2005) appears to raise it against Weak Sufficiency and Raz’s Facilitative Principle. Broome (2005) argues that no reason transmits from the end of alleviating one’s hunger to the overwhelmingly (prudentially, if not also morally) objectionable, but not superfluous, means of killing oneself. In Broome (ms.), he claims to make the same point with a different example: the itchy donkey of note 32. But the donkey makes a different point. Its walking is superfluous, but not overwhelmingly objectionable. What may lead Broome to assimilate the two examples is that, in both, the means are excessively costly. But that may be a red herring. 38 Kieran Setiya suggests another response: namely, that even if there are intuitive cases of silencing, the Silencing Proviso is unnecessary, because such cases are already handled by the Intransitivity Proviso. We resist the conclusion that I have reason to poison the don, for example, only because we think that the ultimate end for which I have reason is not changing policy, but instead changing policy in a morally permissible way, which poisoning the don (like baking the mega-loaf) does nothing to further. Whether this ingenious suggestion will handle all the intuitive cases of silencing, however, I am not sure (in part because I don’t share many of the relevant intuitions). 39 While sympathetic to Broome (1999), Bratman (2009) avoids this result by arguing that in such exceptional cases, the general reason to be instrumentally rational lapses, because it derives from reason to be “self-governing,” which is not possible in such cases. See Brunero (2010) for criticism. 40 Against Weak Detachment, Bratman (2009: 422) suggests that reason in favor of one’s F-ing “involves an implicit comparison of [one’s F-ing] with its available alternatives,” including refraining from E-ing. Thus, it might be that refraining from E-ing, rather than F-ing, is “what there is reason to do.” But this reply seems to the point only if “there is reason to X” means, in the idiom of this paper, “at least as much reason for X-ing as for any alternative to which we are implicitly comparing X-ing” (a usage which makes “overridden reason” an oxymoron). That isn’t the usage at issue here. 41 Compare Yalcin (2007) and Kolodny and MacFarlane (2010). 42 General Production may differ from standard decision theories, however, in at least two respects. Decision theory is typically presented as a theory of rationality, and so focuses on epistemic probabilities fixed by the agent’s information at the time of acting. By contrast, judgments about reasons often depend on other probabilities (e.g., such as those fixed by the information of an advisor, or the non-epistemic historical situation). Second, decision theory often focuses on “exhaustive” actions of the sense described in section 3. By contrast, ordinary judgments about reasons rarely, if ever, involve exhaustive actions. Moreover, focusing on exhaustive actions threatens to crowd out the very phenomenon of instrumental transmission itself. For it ignores other actions from which to transmit reason to, or to which to transmit reason from, exhaustive actions. (See Wedgwood’s insightful 2011 for an analogous point about instrumental rationality.) 43 Compare Anderson (1993: Ch. 2); Scanlon (1999: Ch. 2); and Raz (2012: Ch. 12). 44 This isn’t to deny that one can treat the state of affairs of honoring the thing of value as itself a valuable state of affairs and still get many of the right “results.” But it is still a distortion, even if useful for certain modeling purposes. 45 Insofar as decision theory cares only about ranking actions, it does not need to distinguish this sort of case, in which two actions are equally ranked because there is the reason of equal strength for either, from a case in which two actions are ranked equally because there is no reason for either.