1 Introduction
Constitutivism is a thesis about the metaphysics of normativity: what a thing is fixes how it ought to be. The norms and standards governing an entity hold in virtue of its nature—of what is constitutive of being an entity of that kind.Footnote 1
Most famously, constitutivism has been applied to the nature of agency, to derive the norms of practical reason or morality. But the strategy is much more general in application. A knife is a straightforward illustration of the constitutivist strategy. On account of its nature—its being an entity with the constitutive function to cut—norms of sharpness apply to it. A knife can be evaluated as better or worse by how well it discharges its constitutive function, and activities that deal with knives as knives fall under standards that derive from that nature. A knife-maker, for instance, is criticizable for turning out dull blades.
Artifacts are easy cases, because they wear their constitutive standards on their sleeves. But constitutivism can also be applied to a variety of capacities and powers and especially to entities that carry a whole suite of distinctive powers and capacities, such as organisms and agents.
Constitutivism foregrounds the internal character of these norms. It is because one handles knives as knives, rather than as mere paperweights, say, that one is by default under a pro tanto pressure to keep them sharp. One can escape this default pressure only insofar as one can legitimately be indifferent to the entities in question as the entities that they are.Footnote 2
For many entities, failure to meet the constitutive standards does not amount to a loss of existence. A knife remains a knife even when maximally dull.Footnote 3 True, an organism that fails massively at self-maintenance dies; but its standards can go unmet in many ways without any threat to its existence—the organism might chug along as an imperfect or unhealthy specimen. Likewise for engagements with the entities in question. Someone can be correctly described as genuinely engaged in bladesmithing even if she is so incompetent that no knife ever results. Even the maximally failed bladesmith is, by the very nature of that activity, directly subject to the internal standards of bladesmithing, which in turn reflect the internal standards of knives.Footnote 4 So far, so good.
There are, however, kinds of entities whose existence seems tied not merely to the application of their constitutive standards but to full compliance with them.
Take chess. To move a bishop along a column rather than a diagonal is not to play chess worse; it is not to play chess at all. One is either playing some other game—shmess, say—or not playing at all. Both the existence of the entity (a particular game of chess) and the activity of engaging with it (the playing of that game) are immediately at stake in the full and perfect compliance with chess’s constitutive standards. Other entities, such as certain practices, institutions, rituals, and languages, seem to have a similar structure. Consider promising, where an utterance that fails the conditions for making a promise yields not a defective promise but no promise at all—and so no obligation, rather than a weak one. A statute enacted without proper authority is not a bad law but no law at all (it is void ab initio). In a language, a badly formed sentence is not bad speech but gibberish. In logic, the analogue of a failed move is not weak inference but a failure to infer altogether.
For entities of this sort, the constitutive standards come with an existential threat: full compliance is a necessary condition of the entity’s very existence. This reshapes the force of the standards. The internal demands generated by the nature of these entities are not about securing better rather than worse ways of being, performing, or engaging with them. The demands are about whether the entity exists, or can be engaged with, in the first place. The basic concern is existential rather than evaluative.
The threat is at its most powerful when combined with a claim of necessity—when engaging with the entity is categorically required or, in some suitable sense, inescapable. But this is not required in order to appreciate the distinctive force of the existential threat. To anyone who doubts the authority of a constitutive standard, one can reply that, as long as she has reason to engage with the entity as the entity that it is, she must fully abide by the standard on pain of not engaging with the entity at all. This is a straightforward, if only conditional, way to see why constitutive standards, in Korsgaard’s phrase, “meet skeptical challenges to their authority with ease.”Footnote 5
Arguments that combine inescapability with the existential threat lie at the heart of the earliest and most ambitious constitutivist programs—especially Korsgaard’s and Velleman’s—programs that reach as far as the grounding of categorical morality or at least proto-morality.Footnote 6 But they carry two major vulnerabilities. The first concerns the alleged inescapability of agency: what its nature is, what kind of normative necessity it can ground, and whether agency is inescapable in the relevant sense. I set these questions aside here.Footnote 7 I want to focus instead on the second vulnerability: the existential threat appears to leave no room for defective or bad instances of agency. On this picture, agents must fully comply with the constitutive standards to be agents at all. Exercises of agency are either perfect, relative to their standards, or not exercises of agency at all.
This flies in the face of the evident existence of defective actions and agents, what is known as the “bad action” problem.Footnote 8 It also sits badly with the “error constraint”: to be genuinely governed by a norm, one must be able to violate it, at least in principle—precisely what the existential threat seems to rule out.Footnote 9 As Peter Railton put the worry in one of the earliest critiques, constitutivism threatens to “pull the claws of the very criticisms” it was meant to ground: if a defective x is no x at all, nothing can be criticized for failing the standards of x. Footnote 10 Call this the “bad-x problem.”
One can dissolve the bad-x problem by giving up the existential threat. On this approach, agency is better modeled not on chess but on the operation of powers and capacities, especially in the combinations found in organisms and functional entities. Being an agent then requires only that the constitutive standards apply to one, not that one comply—let alone comply perfectly—with them. The standards of sharpness apply to knives in virtue of their function; a maximally dull knife is criticizable but does not thereby cease to be a knife. Likewise, the norms of agency might apply to an agent through her endorsement or commitment to them, even where she fails to abide by them.
This seems to be the current mainstream interpretation of constitutivism. Explicit appeals to the existential threat, which were common among the first constitutivists, are hard to find.Footnote 11 And the move looks well motivated, since guidance by violable norms and the existence of defective instances are general desiderata, not special pleadings about agency. Their generality makes the early constitutivists’ embrace of the existential threat somewhat ill-conceived.
In this paper, against the mainstream, I want to argue that, for some kinds of entities—including a central aspect of rational agency—the existential threat does real work in grounding the structure and force of the constitutive standards. But to do so, I need to show how this constitutivism-cum-existential-threat can address the bad-x problem. That is my task here.
I will begin by sharpening the two strategies illustrated above into a contrast between two models of constitutivism: the “legal/grammatical” and the “functional.” By “model,” I mean a basic orientation—one that fixes which questions are worth asking, what counts as an acceptable answer, and which examples are paradigmatic—rather than a fully articulated theory. Particular accounts need not be explicit about, or even aware of, the orientation that animates them, and some may combine both.
I first introduce a purely procedural game I call “Legal Chess,” which lays bare the structure of the legal/grammatical model in its simplest form. I then show how the legal model can accommodate defective instances and genuine violations by pushing them somewhat outside the entity, yet without making them wholly extraneous to it. I do so via the distinction between Strict and Broad play: between an activity unfolding in full compliance with the constitutive standards and the activity of trying to generate such compliance, which leaves room for defects of strict play, rather than within it.
I then extend the legal model to more complex entities—such as sports and institutions. The added complexity earns three new resources: a category of internal violations I call “intelligible fouls,” a form of internal self-correction I call “Strict+ play,” and a dynamic cohesion that lets such enterprises survive local breaches. These imperfections amount to a kind of delayed existential threat. But the spirit of the simple model is retained in the need for a core—a kernel—of perfect compliance.
I then argue that structural rational agency—the backbone of rational agency—fits the complex legal model better than the functional one. Hence, structural rational agency falls under a kind of existential threat. This threat is more indirect than the one that looms over the simpler legal entities but allows for forms of internal violations, thereby avoiding the bad-x problem. Or so I argue.
My conclusion vindicates the original constitutivists’ emphasis on the existential threat. This does not show, however, that the legal model can deliver the grounding of substantive normative pressures. The legal/grammatical model, when properly understood, might bring to the fore certain basic limitations of constitutivism. As I briefly mention in closing, the basic constitutivist story about structural rational agency might need to be supplemented by appeal to general circumstances of constitution—such as scarcity, competition, and mutuality.
Even if my specific conjecture about structural rational agency fails, the contrast between the two models carries a methodological lesson for the constitutivist project at large. Any constitutivist account must recognize that there are two distinct models, apply them appropriately and separately, and keep track of the different kinds of rational pressure they generate—especially for complex entities, where distinct models may govern distinct dimensions of one and the same kind of entity.
2 Two Models
2.1 The legal-grammatical model
The first model I call legal or grammatical. Games, rituals, institutions, languages, and codes often exhibit the following pattern: a piece of conduct either satisfies the constitutive rules of the enterprise and counts as a “move” within it, or fails and falls outside it altogether. Imperfections do not grade performance from within; they threaten the standing of the move as such. This is the existential threat: the existence of the enterprise (or portions thereof) is at stake in meeting the constitutive norms. And if the enterprise—a game, say—is not present, an agent cannot be genuinely engaged in it.
The second model is the functional one. It applies most clearly to capacities, organisms, and artifacts: an item may possess an X-function and exercise it badly, yet remain an x even when defective—possibly maximally so. Here imperfection grades performance from within without necessarily threatening the existence of the entity as an x.
Chess is often used as a standard illustration of constitutivist claims. But there are different ways to think about the game, its constitutive standards, and associated conditions of success. First, constitutive standards of legality: for example, a bishop can only move along diagonals. Second, a constitutive aim: for chess as a competitive game, the aim is that of winning by checkmating. Third, standards of competence that determine skillful performance, such as strategic norms and heuristics that promote checkmating. A standard game of chess involves all these norms and the associated conditions of success and evaluation.
There is an order of dependency: to play skillfully, one must play competitively, and to play competitively, one must play legally. A complete failure of skillful play does not necessarily threaten competitive and legal play. A chess novice might lose in two moves by the Fool’s mate, even if wholeheartedly committed to checkmating with a full understanding of what it involves. But a failure to aim at checkmating puts the existence of the competitive game in jeopardy. Someone might be correctly chastised for not really playing competitive chess if she makes only legal moves with no interest in checkmating. A fortiori, there is no game of chess at all (whether competitive or not), if one is not even making legal moves.
Standards of skillful chess play fit the functional model—one can be more or less proficient at the game without threatening the game’s existence. Standards of legality fit the legal-grammatical model, instead. The checkmating aim also fits the legal model but with a caveat. Having the aim is constitutive of competitive play in the legal mode, whereas pursuing the aim falls under the functional standards of competence. That is why one cannot play the competitive game without the checkmating aim, yet can be utterly incompetent at pursuing it without falling outside the game.
To avoid any confusion raised by the constitutive competitive aim, I will focus on a stripped-down version of chess—the purely procedural enterprise of making alternating legal moves, with no competitive or strategic goals, which I will call “Legal Chess” (hereafter, unless otherwise noted, I will use “chess” to refer to the purely procedural game). Although this might seem a strange kind of game, we are familiar with purely procedural enterprises such as ceremonial protocols, liturgies, or the legal execution of formal instruments (such as rules of order).Footnote 12
In any such enterprise, an illegal “move” is no move at all: any violation falls outside the game. Perfection is built in. Every constitutive standard is satisfied by necessity whenever genuine play occurs. Playing in full compliance with a game’s own constitutive standards is what I call Strict play, or S-play. Alongside it is Broad play, or B-play: the activity of the realization of S-play. In chess, B-play includes at a minimum the player’s actual physical moves with material chess pieces but can be extended to preparatory steps (e.g., setting up the chessboard). In addition to successful realizations of S-play (such as the physical motions that realize the abstract moves of Legal Chess), B-play might include failed attempts at S-play (e.g., false starts, a hand that hovers over a piece and then withdraws, etc.) and rectifications of illegal moves (e.g., repositioning a chess piece that was not properly moved).
S-play is necessarily perfect as long as it exists. It is what is described in listing moves in chess notation: an illegal move would be chess-gibberish, barred from the record. The illegal “moves” that cannot belong to S-play may instead appear in surrounding B-play. In this extended sense, one might be said to be playing Legal Chess even while failing to engage properly in the Strict version. Extending an arm to move a piece might count as bona fide B-play even if the move ends up illegal from the point of view of S-play. The standards of B-play make reference to, but do not reduce to, the standards of S-play.
B-play allows for internal defects, such as teleological failures to bring about S-play—defects that do not necessarily affect the existence of B-play. My present focus is on the legal-grammatical model as exemplified by S-play. Yet, how S-play and B-play relate matters, given that “play” and “engagement” admit both a strict and a broad reading.
The central feature of S-play in Legal Chess is that, when it obtains, it is in full compliance with its constitutive standards. A lack of compliance does not yield a defective S-game but the absence of one. There cannot be fouls within S-play; there can only be fouls of S-play—putative but failed strict legal moves, only visible as bona fide moves within the larger B-play. In S-play, the existential threat is preserved without compromise.
S-play has both axiological and ontological primacy over the corresponding B-play. Start with axiology. Certain goods and values are available only within S-play. The elegance of certain configurations or sequences of moves, for example, can only be instantiated and appreciated within the game. Players might come to appreciate kinds of striving and skill that can only be cultivated within the game. The value of B-play, instead, is only derivative: it is instrumental to realizing goods that can only be secured and appreciated within S-play.
Likewise for conceptual-ontological primacy. B-play is only contingently necessary for S-play; its role is to realize it. This is clearest with an abstract game like chess, which admits many material realizations. Each game is a sequence of well-formed moves within an abstract logical space of possibilities. For material and temporal beings like us, this space must be physically realized—on chessboards, on paper, in computer code, and so on. B-play is the enterprise of securing some such material realization. Hence, S-play is conceptually and ontologically primary: S-games and S-players exist as such only insofar as they meet the S-play standards, whereas the messiness of B-play shows up only in the demands of the causal-material realization of the legal enterprise.Footnote 13
The existential threat built into the legal-grammatical model pushes defects and violations outside of S-play but into B-play. Although from the point of view of S-play, there cannot possibly be genuinely internal defects and violations, these do not completely drop out of play or engagement. They can stay within B-play. But given the subordination of B-play to S-play, it is important that the failures do, in a very real sense, fall outside of the primary sense of “playing” and “play.” Likewise, players can make errors and violate norms of S-play, but those are visible only from inside B-play. Again, given the primacy of S-play, recall that there is an activity of playing chess in the strict mode that is indeed metaphysically immune to internal errors. This is an asymmetry that cannot be captured by the functional model—where the failures and errors are by default internal to the entity or activity.Footnote 14
2.2 The functional model
Consider now the alternative model—the functional. Take a knife. Sharpness, balance, and durability are constitutive of being a good knife. A dull, unbalanced, or non-durable knife invites an internal negative appraisal—possibly as severe as useless—without threatening its loss as a knife.
More generally, a functional entity e with a function or capacity F is subject to standards governing F’s proper exercise. The entity e may exercise F badly, if at all, without losing its function F or its identity as an F-functional entity. Internal defects of functional entities are common and gradable, and perfection is often unattainable, operating only as a regulative ideal. Crucially, defects need not be existential threats. A blunt knife is a worse knife but not necessarily a non-knife. Only the most catastrophic defects (the knife melting into a blob of metal, say) destroy the knife as such.
What is primarily at stake in a functional entity’s proper operation is not its existence but our engagement with it. A blunt knife remains a knife, but it is likely to be abandoned for a sharper one. Yet, a maximally defective knife is still a knife, even if no one would ever use it as one.Footnote 15
2.3 Comparing the two models
The legal model vindicates full compliance under existential threat but admits no internal defects. The functional model, by contrast, admits possibly maximal internal defects but carries no necessary existential threat.
This reflects a difference in the direction of perfection. In the functional model, perfection is aspirational and possibly unattainable; the entity normally falls short of it, along a gradient. In the legal model, perfection is presumed as the starting point; an entity can only fall from it, and the fall is necessarily catastrophic.
The difference also shows in how we describe the distinctive pressures, failures, and corrections. In the functional model, the pressure is to make the entity work and to work better. The defect is a fumble: a misfire within the distinctive operation. A fumble calls for adjustment, not restoration of status: the system has misfired but is still the system it was. The correction vocabulary is one of repair, improvement, and course adjustment.
In the legal model, by contrast, the pressure is to make the entity whole, to make it be. The defect is a foul—a breach of order rather than a misfire within it. The failure is in the coming to be, not in the performance. A foul calls for restoration of proper form: something has breached the order; the question is whether and how it can be reinstated. The correction vocabulary is one of rectification, restoration, and reinstatement.
The fumble/foul contrast marks where the standards get hold, and the two directions of perfection correspond to two different kinds of normative criticism.
Consider how the two models relate to Jeremy Fix (Reference Fix2021)’s distinction between “norm-first” and “nature-first” constitutivism. In norm-first accounts, an item belongs to a kind because it is subject to certain standards; in nature-first accounts, an item is subject to a standard because it belongs to a kind. At first pass, the two taxonomies seem to overlap: the legal-grammatical model’s paradigms, such as games and institutions, fit the norm-first mold; the functional model’s, such as organisms, fit the nature-first mold.
The two taxonomies, however, track different explanatory questions. My distinction concerns kinds of norms/standards and the associated defects. Fix’s distinction concerns the direction of metaphysical explanation between kind-membership and governance by norms. This allows a nature-first constitutivism for entities whose norms are legal-grammatical. Their nature would explain why legal standards apply, without itself being characterized primarily by those standards. Some institutions might be individuated by roles (coordination or adjudication, say) they can perform only in the legal-grammatical mode. Likewise, languages and codes are by their very nature governed by logical-grammatical standards—conditions for well-formed formulas, validity, and the like—without being individuated by them in a norm-first way. Later I suggest the same of rational structural agency, a capacity whose nature could be characterized independently of a specific set of norms but whose core operation fits the legal-grammatical model.
The two models capture recognizable regions of the normative landscape: the functional model fits artifacts, capacities, and organisms; the legal-grammatical model is a better fit for games, rituals, institutions, and languages. The two models are not necessarily incompatible. As previously indicated, a game might display features of both models, legal constitutive standards together with standards of skillful competence, for instance. But different aspects of the same entity might fall under different models. And in the case of such entities as games, rituals, institutions, and languages, the legal model plays a central and crucial role in understanding their nature and operation.Footnote 16
2.4 The models and agency
Which model fits the agency better? Agency in the broadest sense, as the capacity for teleological regulation, aligns naturally with the functional model.Footnote 17 This regulation comes in degrees of success, and the capacity might be attributable even when it cannot be exercised, possibly beyond repair. Massive failures of teleological regulation need not threaten the existence of this capacity or its carrier. Violation of constitutive standards of minimal agency need not raise an existential threat.Footnote 18
What about rational agency? If we think of it broadly in terms of the image of dwelling and navigating in the space of reasons, rational agency appears to be a regulative capacity: a responsiveness to a normative landscape of reasons and normative pressures. This capacity looks akin to teleological regulation and hence a good fit for the functional model.
But two distinctive aspects of rational agency are a much better fit for the legal/grammatical model. First, the normative landscape, unlike a purely physical, causal one, has contours and features with an unmistakably legal/grammatical flavor: concerns for correctness and validity, deontic roles and statuses, entitlement and commitment, warrant and standing, score-keeping, admissibility, and grammaticality. Responding to these features appears more akin to the playing of strict Legal Chess than mere teleological regulation. There might be a regulation at the level of the corresponding B-play, but the point of this responsiveness and navigation is to meet the legal-grammatical standards in the Strict mode of S-play.
Second, there is the so-called structural dimension of rationality: securing the structural consistency and coherence of some of the agent’s attitudes and actions. Here too, I want to suggest, the legal/grammatical model is the better fit—at least if we view the structural demands through the lens of intelligibility, understood as the attributability of attitudes and actions to a unified rational subject. Unlike failures of teleological regulation, failures of intelligibility are not mere performance defects. They threaten the attributability of attitudes and actions to the rational subject: an unintelligible psychological or bodily “move” is not the move of a rational subject—the single locus in which attitudes, decisions, and actions hang together as the doings of one and the same agent.Footnote 19
As an illustration, consider three clear-eyed violations of basic norms of structural rationality. First, holding a patent contradiction: an agent clear-headedly affirms both p and not-p. Second, a failure of inference: an agent who, at the same time, clearly believes p and p implies q but refuses to draw the conclusion q. Footnote 20 Third, an agent who intends an end while refusing to intend the clearly known necessary means to it. In all these examples, the subject simultaneously holds together, clearly and distinctly, a set of seemingly inconsistent attitudes.Footnote 21 A clear-headed subject who insists on these combinations turns out to be unintelligible. The putative attitudes or actions cannot be made sense of as belonging to a single unified rational agent. Either (a) they devolve into psychological gibberish, linguistic noise, or mere motions; or (b) they remain locally well-formed but attach only to a fragmented subject (or to different subjects altogether). The trouble is not irrationality or inconsistency as faults of a unified subject, but the very attributability of the attitudes and actions to a single such subject (hereafter, talk of attitudes might often be interpreted as encompassing, mutatis mutandis, actions as well).
As Eric Marcus (Reference Marcus2026) aptly puts it, this is a metaphysical impossibility grounded in the nature of rationality. A failure to meet the demands of rational necessity makes the attribution of attitudes inconceivable.
Indifference to […] fundamental logical principles ultimately makes understanding someone impossible—not because we view it as an obstruction in the expression of their beliefs—but because we have lost any reason for thinking that the relevant noises are serious and sincere expressions of belief.
(Marcus Reference Marcus2026, p. 289)By analogy with Legal Chess, a failure to meet the standards of intelligibility looks like a foul in what might be called the Be-a-Unified-Rational-Subject game or enterprise. When such a failure occurs, that conduct appears to drop outside the unified subject. Such a failure does not count as the exercise of rational structural agency by a single unified agent.
The existential threat is not just to the attitude or action but also to the agent.Footnote 22 In two ways: the agent loses a part of themselves—something that should have been theirs is not; and a systematic violation of these norms jeopardizes the agent’s existence altogether.
If so, however, the bad-x problem returns. If being a rational agent amounts to the S-play of the Be-a-Unified-Rational-Subject enterprise, there is no space for defective exercises of rationality within the subject in the strict sense, akin to the impossibility of illegal moves within S-play of Legal Chess. The rational agent is always in perfect compliance with its own constitutive standards: what does not comply falls outside the subject altogether. But how can one deny that rational subjects often hold contradictory beliefs and inconsistent attitudes? Structural irrationality appears pervasive.
Here is where the Strict versus Broad play distinction might help: the irrationality drops out of S-play but still counts as the subject’s Broad-play: as a (failed) attempt at intelligibility and attributability. The distinction would preserve the primacy of S-play and the asymmetry of compliance versus violations, making violations external to S-play and only indirectly visible to it (for instance, in a “third-personal” rather than “first-personal” mode). This is supported by common ways of describing attitudes and actions that fail to meet standards of structural rationality in terms of “alienation” or lack of “identification.” Likewise, the mechanisms invoked to explain the violations are cast in terms of “interferences” to operation and attribution, such as ignorance, obfuscation, distraction, repression, dissociation, denial, and masking.Footnote 23 Finally, that irrational attitudes drop outside a fully integrated subject is suggested by talk of divided or fragmented subjects, at least when larger sets of attitudes or actions are involved in the failure of integration.Footnote 24
Hence, by analogy with the game of Legal Chess, one might suggest preserving the purity and priority of the unified subject in the strict mode by “dumping” errors and violations into B-play. But there is a worry that the notion of B-play may not carry over to structural rational agency. To see why, let us return to Legal Chess. When playing such a game (and, more generally, engaging in any of the legal/grammatical enterprises), both S-play and B-play take place within rational agency. This is why the violations of constitutive standards are possible for the agent “at large.” What is ruled out is only violations for the agent in its role as a strict-player of Legal Chess.Footnote 25
When playing chess, the agent at large can, more or less seamlessly, move in and out of S-play by falling back into the related B-play; a fact that explains the ambiguity in our everyday talk of “playing” and related notions of “success” or “failure.”
But the option to fall back into B-play appears unavailable for structural rationality itself. As I argued in Ferrero (Reference Ferrero2009, Reference Ferrero2018), rational agency is the only—and the biggest—game in town. All deliberation, evaluation, justification, and critique—the distinctive moves of rational agency—must proceed from and answer to the constitutive standards of rational agency itself. There is no coherent external vantage point from which those standards could be assessed or supplanted. Rational agency is, in my terminology, dialectically inescapable (Ferrero Reference Ferrero2018, p. 128): closed under its own distinctive operation. Any reflective assessment of a rational failure is itself an exercise of rationality, answerable only to its own standards (see Ferrero, Reference Ferrero2009, sec. 3).
If this is correct, there is no broader external place whence a rational agent could try to fix violations of the constitutive standards of rational agency. Piecemeal failures can be addressed and possibly fixed, but only from within what is still Strict-play of structural rational agency. The rational agent cannot step out of agency’s S-play to try to fix it, as she instead can do for ordinary enterprises.
One solution would be to treat B-play as the arena for non-rational agency—a non-reflective (though possibly self-maintaining) teleological regulation that sustains rational agency. After all, we need our underlying biological agency to embody and support our rational agency. We often fail as rational agents because of breakdowns in the biological substrate (fainting, say) that interrupt rational agency from below.Footnote 26 But this still leaves agency’s S-play internally immune to failure: defective B-play, in the mode of mere biological regulation, only shrinks the region of perfect S-play, without introducing any genuine internal defect of rationality.
We are at an impasse. If there is no possible B-play for structural rational agency, the bad-x problem returns: there are no genuine internal violations of the constitutive standards of agency, only the brute causal dropping in (and out) of a unified rational subject who is always in full compliance with its own constitutive standards (even if she might be smaller in extension because of losses of attitudes and actions that are unintelligible and unattributable to her).
What next? I suggest that we set aside rational agency for a while and revisit the legal model. I want to consider enterprises more complex than simple Legal Chess. These enterprises require richer forms of integration that, as I will suggest, provide additional tools that make room for some kinds of genuinely internal failures of structural rational agency. Or so I will argue.
3 Legality and Complex Integration
On the picture of legality modeled on Legal Chess, any illegality produces immediate and complete collapse: an illegal “move” interrupts the entire sequence then and there. The collapse is not irremediable—another legal move can always be added—but the violation can be undone only by closing the gap at the point of collapse.
The fragility of Legal Chess may blind us to the complexity of other enterprises that still fit the legal-grammatical model. Many ordinary games, practices, institutions, and languages exhibit three features that Legal Chess lacks. First, instead of a single sequence of discrete moves, they admit multiple concurrent moves, often extended and continuous. Second, they display varied scopes of integration—local, regional, and global. Third, they have degrees of robustness: parts can fail without bringing down the whole.
Soccer and a political institution illustrate all three features. In soccer, 22 players act at once, in continuous, overlapping motion; its norms apply at different scopes—a single tackle, the offside line across a stretch of field, the global structure of two teams in a match—and a handball or a player sent off need not end the game. A constitutional order—legislatures, courts, agencies, local authorities—has a similar shape: many concurrent moves (statutes, rulings, ordinances, often in parallel), integration running from a minor official’s act up to the order as a whole, and a robustness that lets a defective ordinance, an ultra vires act, or even a rogue agency be isolated, corrected, or absorbed while the rest carries on.
For such enterprises, a violation need not produce immediate, complete collapse. It may instead force a resizing or reshaping—a contraction or redrawing of the enterprise’s expanse, including its division or fission—with no internal imperfection; or it may pose a threat of delayed collapse, a breach that propagates through the structure over time.
That complex enterprises tolerate local breaches raises a worry: is the threat now functional rather than legal-grammatical? I maintain that it is still legal-grammatical. Even at its most accommodating, the threat is to the form rather than the function of the structure, to its coming to be rather than its performance.
What changes is not the nature of the threat but its tempo. The threat no longer comes in the immediate, catastrophic mode of Legal Chess or a logical system. A Legal Chess sequence interrupted by an illegal move has nowhere to go. Under ex falso quodlibet, a lone contradiction trivializes a logical system, every sentence becoming provable. But soccer and political institutions are not abstract structures. They may rely on logical operations and aspire to consistency, yet they are material in constitution and operation—physically realized, with finite signal speeds and finite means of propagation. A breach of legality need not spread at once.
For these enterprises, proper form is not the static integration of a logical structure or a one-dimensional sequence of moves. It is a cohesion-in-motion under material constraints of locality, causality, and resource scarcity—maintained by avoiding practical or operational inconsistency across the enterprise’s spatiotemporal stretch, not logical inconsistency in an abstract structure. This drag buys time and space: a local failure can be isolated and corrected before it propagates, and proper form persists as a dynamic pattern of spatiotemporal legal-grammatical cohesion.Footnote 27
Operational integrity is thus more accommodating of violations, at least temporarily. At its simplest, the enterprise resizes: it externalizes what it cannot currently hold—declaring it out of scope, suspending it, and treating it as not-yet-incorporated. For instance, a political institution might invalidate an ultra vires act, effectively expunging it from the constitutional order. In other cases, the enterprise might manage a breach dynamically by “absorption.” Consider the advantage rule in soccer, where the referee acknowledges a foul but lets play continue in the service of the continuation of the game’s central goals. Or cases in which a violation can no longer be expunged but is instead incorporated by retrospective ratification—as when a legislature passes an act of indemnity, retroactively conferring the authority an official lacked and validating what was already done. Such absorptions are possible because cohesion is dynamic: the enterprise can partly reset or modify its own terms.Footnote 28
3.1 Conjecture: a dynamic kernel of perfection
These observations motivate a conjecture. Proper form, for any materially realized, temporally extended, legal-grammatical enterprise, requires a local kernel of full compliance. The kernel is the locus of basic individuation and re-identification, and the seed of extension and re-integration: the place from which repair or absorption of peripheral breaches begins.
Why a non-trivial locus of full compliance? If violations are widespread and no region is locally compliant, on what basis could one claim the enterprise is present rather than merely apparent? The locus is needed for individuation, attribution, and correction of attitudes and actions. First, individuation: any extended enterprise needs synchronic unity and diachronic continuity, and an enterprise defective everywhere has nothing to anchor either. Second, attribution: to identify a move as a correct move of the enterprise rather than an infraction, one needs a determinate locus relative to which moves can be sorted. Third, correction: re-integrating a peripheral breach must radiate from somewhere already recognizable as having proper authority.
Consider an ultra vires exercise of power by a local authority within a larger political institution. It issues acts that exceed its competence. Locally, these may go unrecognized and operate smoothly, but their downstream effects accumulate and ultimately clash with other portions of the institution, threatening practical inconsistency and collapse. At that point a response is called for. The hitherto unaffected kernel can correct, shrink, or absorb the past infractions and their consequences in selective and varied ways—just as, in soccer, the still-compliant run of play is what allows a handball to be answered rather than to dissolve the match. But all such responses presuppose a kernel with constitutive competence and authority: a locus that operates as the singular standpoint of a dynamically integrated enterprise.Footnote 29
The integrity at issue is not centralization in a single spatial location. Likewise, the kernel need not be fixed over time: it can shift and be scattered across the enterprise. What is required is a kernel in the operational sense—a unified, identifiable, traceable core of correct legal operations that handles infractions as such, by externalization, correction, or absorption.Footnote 30 But within its extent, the kernel must be coherent, non-defective, and non-trivial: even if small, it must preserve enough structure to operate at the relevant level of integrated complexity.Footnote 31
By contrast, a purely functional system needs no kernel of perfection. Its components can perform more or less well, more or less defectively, without a core of perfect operation. The operation need not be anchored in a core from which authority or legality originates. In this sense, functional operation may be more holistic than legal-grammatical operation. (This is not to deny that the overall integration of a legal-grammatical enterprise can be assessed in a gradable way—but the function in question is a function of the legal-grammatical form.) The kernel’s perfection is required not for the enterprise’s performance but for the proper form on which identification and standing depend. This is also why the loss of the kernel is an existential threat in a way that even complete functional breakdown is not. Sudden, complete functional breakdown need not impugn the identity of the entity to which the function is attributed; but for a legal-grammatical enterprise, absent any kernel of correct operation, there is no integrated enterprise at all—at most the ghost of one.
3.2 Internal repairs: strict+ play
A second dimension of complexity is the kernel’s ability to self-adjust. Some enterprises engage in internal repair: they have procedures for detecting, adjudicating, and mending violations.
Strict Legal Chess does not. Any violation is external to it, and any correction takes place within B-play. There is no procedure within S-play for handling an illegal move: if a piece is moved illegally, legality must be restored externally—by revoking the move, resetting the position, or abandoning the game.
By contrast, consider a sport that handles some violations internally. In soccer, a handball is a foul: a player who stops the ball with a hand has not made a legal move. But the game does not end. The referee awards a free kick and play resumes. The foul is absorbed—its disruptive effect registered and answered, without the players stepping out of the game to reinstate it from outside. The rectification is a move within the game, specified by its own constitutive rules and drawing on some of the same capacities that produce ordinary play.
A rectification restores proper form “from within.” It, thus, differs from Broad play, such as a kid who runs onto the field to kick the ball while a match is in progress. The kid attempts to enter the game from outside, but constitutively fails: the entrance is not a foul but a genuine interruption, and the match must be resumed from outside once the disturbance is removed. The kid’s attempt to enter is a failed B-play; the players’ resumption is a successful B-play.Footnote 32
The foul called by a referee, by contrast, is an instance of self-correcting S-play: what I will call Strict+ play. It adds to S-play the capacity to absorb some infractions from within, through conduct that is itself a move within the game rather than an excursion into B-play. Rectification moves are second-order: they target violations of first-order norms, and so require an explicit understanding of what a first-order violation is, plus whatever capacities are needed to detect and correct it. (First-order soccer does not require the capacity to issue yellow or red cards—it might not even require a referee; this capacity belongs to Strict+ soccer, instead.)
Any rectification, however, presupposes a kernel of perfection: a sufficiently large, locally integrated, infraction-free portion of the game from which the handling of a breach can proceed. A handball is absorbed as a foul, rather than registered as a mere interruption, only because the surrounding play still operates under the game’s constitutive standards: the whistle, the free kick, the resumption draw their status from a portion of the enterprise that is, at that moment, fully compliant. If violations were everywhere—players ignoring offside, the referee awarding random penalties, the goalkeeper wielding a tennis racket—there would be no locus from which touching the ball with the hands could be identified as a handball, and no authoritative standpoint from which to call the foul.Footnote 33 As in the political case, this kernel need not be central or fixed: it is distributed across field, players, officials, and the run of play and can shift as the match unfolds—through substitutions, expulsions, or changes of officials. What it must preserve is enough integrated structure to sustain Strict+ play’s second-order operations: recognizing violations, issuing authoritative rulings, and reintegrating play under shared terms. Strict+ play thus instantiates the dynamic kernel. The same locus of full compliance that individuates the enterprise provides the standpoint from which local breaches are answered without external repair.Footnote 34
4 Structural Rational Agency: Kernel and Strict+ Play
Let us return to structural rational agency. Strict+ play offers a better model for rational agency as the Be-a-Unified-Rational-Subject enterprise. One plays it by securing a dynamic, partially self-repairing, locally compliant kernel of intelligibility—vulnerable to disintegration, but often only of the delayed kind, propagated from the distal effects of peripheral infractions. Structural rational agency exhibits capaciousness, resilience, and self-correction resembling those of soccer or political institutions.
4.1 Capaciousness and resilience
Rational psychologies like ours are more capacious and resilient than the fragility of simple Legal Chess would suggest. A rational psychology is a kind of hyper-chess: it unfolds on multidimensional boards, with many pieces moving in parallel, never fully transparent at the global level of the agent. Attitudes, inferences, evaluations, decisions, and actions run concurrently, at varying levels of explicitness and integration, and what happens at one location need not affect every other in the same way or at the same time. Logical consistency may operate as a regulative ideal, but operational holism is limited by the locality and drag of material propagation: a local defect need not propagate at once and can be quarantined, however imperfectly, while the kernel conducts repair.
This capaciousness is tied to multidimensionality. A defective attitude or action may still play partially correct local roles—guiding action here, prompting reflection there—even as its faulty features are diagnosed and repaired. A belief held in self-deception may still function correctly in most roles: guiding inference in ordinary cases, anchoring practical reasoning outside the self-deceptive zone, remaining available for further use. Its defective aspect—imperfect responsiveness to reasons where the stakes trigger the self-deception—does not nullify these correct roles. The rational subject does not simply “have” a defective belief the way a dull knife is dull: the belief plays a complex partial role in a psychology that, as a whole, remains committed to the norms it violates.
This explains why violations of rationality, when they do not turn putative attitudes into psychological nonsense or push them into another subject, take one of the two forms noted earlier: the temporary isolation of piecemeal attitudes through mechanisms of “alienation” or “interference” (such as distraction, obfuscation, and masking), or the more sustained, structured separation of larger portions of a psychology, in the mode of fragmentation or division.
4.2 Self-correction of rational agency
Structural rational agency also exhibits self-correcting powers in a non-accidental way. A standard legal-grammatical enterprise need not be self-correcting, even when necessarily “guided” by its standard: an infraction may simply drop the alleged move from the enterprise, and the players step outside into B-play to steer their conduct back into the S-play, without the enterprise itself adjusting in the way that teleological regulation would do. Teleological regulation self-corrects a course of action, within a range, for deviations from a target; B-play may regulate toward correct S-play as its target, but S-play need not regulate for itself.Footnote 35 The capacities exercised in B-play need not match those of S-play: the B-play teleological ability to put pieces back after knocking over the chessboard is not the S-play “grammatical” ability to make a legal move—which need not even require a physical board.
Structural rational agency, by contrast, is necessarily liable to internal self-correction. To be a rational agent is, in part, to be an entity whose attitudes and actions are, at least implicitly, positioned as correctable upon recognition of a violation of rationality standards. This is not merely teleological self-correction. A teleological system corrects itself because a mismatch between progress and target causes a corrective signal—the deviation pushes the system back on track through a feedback loop. Rational self-correction responds to the mismatch as a reason to correct: the agent apprehends the deviation as failing a standard and is moved by its normative force, not its causal pressure. The correction is something the agent does—attributable to her as her own doing—not something that happens within her. It is not a different set of powers brought to bear, nor compensation through feedback, but a move within the same Be-a-Unified-Rational-Subject enterprise: a further exercise of the capacity for intelligibility that brings defective conduct into line with the recognition of reasons. (Explicit self-correction may require a reflective capacity, but this added power determines the explicitness of the correction, not its origin or direction.)Footnote 36
4.3 Psychological kernel
According to my conjecture, the dynamic kernel of perfection for a rational psychology plays three roles. First, it is the starting point of attributability—the primary locus of the rational subject, anchoring both the attributability of the psychology and the identity of the subject. Second, it is the source of corrections as restorations of proper form: the standpoint from which the subject diagnoses and repairs peripheral breaches. Third, it offers the model of rectification—the shape toward which corrections steer. The kernel is dynamic in size and shape: it may expand, contract, and migrate, but at every moment, it is in full local compliance.
Consider the formation of a belief that is not fully responsive to epistemic reasons—through self-deception, say. Once formed, the belief can operate smoothly for a time, but it stands in tension with the kernel of the subject’s rational psychology, including the subject’s commitment to epistemic standards. When a decisive confrontation arrives—overwhelming contrary evidence, say—the subject faces the question of what to do with the belief and its consequences, from self-correction to assimilation. These responses must proceed from the kernel. Without a locus of rational psychology in full local compliance, there is no standpoint from which to call the defective belief back, no model for correction, no subject to whom the violation and its responses can be attributed.
4.4 Imperfections of rational agency
A more fine-grained taxonomy of failures is now available: alongside fumbles and nonsensical “moves,” we have intelligible fouls.
First, a fumble is a teleological or functional failure that is nonetheless legal: a move that meets the conditions of intelligibility but fails its substantive end. In chess: moving a rook correctly, along a rank or file, but weakening one’s position. In rational agency: forming attitudes through valid reasoning and in response to reasons, but failing to attend fully to relevant considerations or to exercise one’s reflective powers. Attention and reflection operate on the functional model against a background of basic structural intelligibility; their defects, therefore, register as (functional) fumbles rather than (illegal) fouls.
Second, a nonsensical “move” falls outside the rational subject altogether: causally explainable but rationally unintelligible, not even recognizable as an attempt at a rational move. In chess: someone who tries to eat a piece rather than move it. In the psychological case: stretches of behavior so disorderly that no rational interpretation applies. Attribution withdraws, and brute causal explanation takes over.
Third, and most important here, are intelligible fouls. An intelligible foul is an interpretable but failed attempt at a legal move: an attempt at a rational move, made from within rational agency, that nonetheless fails the constitutive standards. A seeming chess analogue is moving a rook diagonally, as if it were a bishop. The move is illegal but recognizable as an attempt at a chess move (unlike trying to eat the piece). In chess, however, this is a move within B-play. It is not internal to chess—which is why the bad-x problem seems to arise for chess. Strict Legal Chess has no bad moves; B-play accounts for defectiveness, but only as external to S-play.
In structural rational agency, by contrast, a defective move, once recognized, can be partially excused—not as exempt from the demand for correction, but as a faulty exercise of rational powers rather than a clear-eyed violation of constitutive standards. An excuse locates the defect within the agent’s rational economy, showing how the foul came about while keeping the agent recognizably within the Be-a-Unified-Rational-Subject enterprise.
Consider the self-deceived believer. Imagine a researcher who, driven by career stakes, discounts reviewers’ criticisms and clings to the merits of her submission. Her belief remains explainable: she consults evidence, draws inferences, marshals reasons. But it is not fully intelligible: under sustained scrutiny, she could not endorse the link between the data and her confidence in the manuscript. The same rational capacities that enable epistemic self-assessment here serve to subvert it. The belief is only half-owned. It puts pressure on attributability without collapsing into obsession or nonsense: she is neither a different subject from the one who reads her reviewers fairly, nor an entity whose belief-formation has fallen outside rational attribution. In my terminology, she is, at least for the moment, a subject committing an intelligible foul.
Intelligible fouls are possible because of the capaciousness and self-correction of Strict+ play. Capaciousness under material drag allows partial structural integration: even ultra vires acts can play an intelligible role, at least locally and for a time. And when internal conflict ultimately arises, the violation can still be handled internally, by self-correction or by accommodation and absorption. What the continued existence of the entity rules out is only the wholesale clear-eyed rejection of the constitutive standards as such.
Return now to B-play. Earlier I suggested that structural rational agency does not admit of it. This does not mean one cannot exit rational agency. Exiting is easy: behavior may fall into mere purposiveness, fragmented psychological gibberish, or a stream of brute causal happenings. But in none of these cases is there any longer a place from which a subject can handle violations as internal intelligible fouls. Whatever remains might regain entry, but not by exercising rational agency: re-entry would be like regaining consciousness or acquiring rational powers through training and early psychological development.
The Be-a-Unified-Rational-Subject enterprise is thus closed under its distinctive operation—not because there is absolutely no space outside it but because that operation cannot be exercised from outside itself. There is space for the agent to go upon exiting, but no space outside structural rational agency for its own operation. Unlike other legal-grammatical enterprises, which admit a nested series of Strict, Strict+, and Broad play, rational agency admits only Strict+: a single arena, the same distinctive powers, under a default demand and capacity for self-correction. The same powers are responsible both for correct rational moves and for redressing violations.
What this preserves from the legal model is not fragility but a distinctive kind of normative authority. Rational agency remains a matter of legality rather than mere function: failures still come in the mode of the foul rather than the fumble, in the being or coming to be of the subject rather than in the performance of a function. The axiological and ontological primacy of Strict+ play over any auxiliary accommodations is preserved—it is only within the Strict+ game of intelligibility that the distinctive goods of being a rational subject become available. What is added is the capacity for internal self-correction: a kernel that can re-form as well as preserve.
4.5 Primacy of the legal
With Strict+ play and the dynamic kernel in view, a rational psychology supports gradable assessments. A belief held in self-deception is more or less integrated; an intelligible foul is more or less excusable; the kernel itself is more or less extended. These assessments locate conduct on a continuum that S-play did not admit. One might thus be tempted to claim that the legal model has been quietly abandoned for a functional one.
This temptation should be resisted. The gradable assessments are still primarily about form, not performance: they measure the scope and cohesion of the enterprise and the dynamic distance from the kernel. The scalar vocabulary describes these gradients; it does not capture the structure of the norms of dynamic cohesion.
The existential threat persists, even where the concern with violations takes the seemingly weaker form of a worry about dynamic inconsistency rather than the immediate loss of intelligibility and attributability. Such inconsistencies matter because, left to propagate unchecked, they force the rational subject to shrink, fragment, or collapse—eventually undermining the form of the Be-a-Unified-Rational-Subject enterprise and the attributability of conduct to a single subject.Footnote 37
5 The Limits of Constitutivism
This picture of the legal-grammatical operation of structural rational agency meets two main desiderata but exacts a price. First, it vindicates the existential threat—the existence, size, and shape of a rational subject are at stake in meeting the constitutive standards. Second, it leaves room for genuine defectiveness—the intelligible foul—as an internal failure rather than a falling out of the enterprise. In my view, these two claims address the bad-x problem for structural rational agency.
But the existential threat now bites less hard than some constitutivists might have hoped. The constitutive standards exercise full authority only over the kernel, which may be tiny in extension. If rational psychology is capacious enough to excuse many intelligible violations, the agent may shrink to a small core of perfection surrounded by a large, defective remainder that limps along—the standards still applying throughout, but too easy to fudge outside the kernel. Such constitutivism thus risks being thin in content and tiny in authority: it polices the kernel’s integrity but has only a more limited, indirect role beyond it. This is not nothing—the preserved existential threat anchors the distinctive status of rational agency, keeps normative pressure in the mode of being or coming to be rather than doing, in line with the legal-grammatical model. But some constitutivists might want the existential grounding to underwrite a more substantial body of requirements, up to the demands of morality or practical rationality, perhaps, and the kernel alone does not seem to deliver this.
Whether the gap can be closed without abandoning the existential threat is a question I cannot settle here. The route I find most promising is to augment constitutivism with structural constraints arising not from structural rational agency as such but from the general conditions under which this agency is materially realized for beings like us—what might be called the circumstances of constitution. Two seem central: the scarcity that binds the pursuit of ends, and the mutuality through which intelligibility is sustained among rational subjects, across persons and across time. These circumstances might press rational agents to be larger, more complex, more stable, and more integrated than the bare kernel requires—without necessarily prescribing any unique configuration. They establish a trade-off rather than a mandate: a subject may remain smaller, simpler, and closer to the kernel, easier to keep in compliance, or extend itself into something larger and internally richer, commanding the goods of integration, mutual correction, and reliance but at the cost of exposing itself to more intelligible fouls and the attendant higher risks of ultimate disintegration or fragmentation. How far such pressures reach, and whether they can ground more than the bare maintenance of attributability, I leave for further work.
In this paper, I have offered an account of how constitutivism about structural rational agency can accommodate imperfect subjects while preserving the existential threat’s appeal as a ground for normativity according to the legal-grammatical model. The bad-x problem is not dissolved by making defectiveness cheap; it is answered by showing that legality, suitably enriched, has room for a distinctive kind of internal defectiveness—the intelligible foul—that is neither a mere fumble nor a fall out of rational agency.
My conclusion depends on a conjecture: that any materially realized rational psychology must possess a kernel as locus of perfect intelligibility, a dynamic pattern of cohesion-in-motion from which self-correction proceeds. Around this kernel, the Strict+ play of the Be-a-Unified-Rational-Subject enterprise gives structural rational agency closure under its distinctive operation, making it dialectically inescapable. Whether anything beyond the kernel’s minimum can be grounded—by the circumstances of constitution under which rational agency is materially realized—is a further question to be explored elsewhere.Footnote 38
Imperfect rational agents are not defective versions of impossible perfect ones. They are the kind of subjects that materially realized rational agency permits, and, under the pressures that the circumstances of constitution bring to bear, the kind of subjects that materially realized rational agency shapes. What the constitutivist can say about them is less than the most ambitious forms wanted, and more than the functional model ever could. Between the impossibility of perfection and the dissipation of engagement lies the space of the intelligible foul: a failed move made from within the Be-a-Unified-Rational-Subject game, by a kernel of agency still in play and still answerable for it.
Luca Ferrero is Professor of Philosophy at the University of California-Riverside. He works on the philosophy of agency, practical rationality, and personal identity, with particular attention to their temporal dimensions.
Open access
