What this document is
The political philosophy document named punishment the framework's hardest open political problem and called it "the richest thread to pull on next." This piece pulls it. The problem as stated there: punishment deliberately imposes constriction on an aperture, which the framework treats as prima facie bad; any justification has to show the constriction preventing greater collapse, restoring conditions for elaboration, or maintaining the network's protective agreements — and has to survive the framework's own standing warning against collapse dressed up as correction.
The word in the title is doubled deliberately. Criminal punishment is the central and hardest case, but the same structure appears at every scale of life under softer names: the parent imposing a consequence on a child, the profession sanctioning a member, the friend withdrawing from a betrayal, the community excluding a wrongdoer, the self punishing itself in the privacy of its own coherence. The euphemism "consequences" usually marks a wish to do the thing without owning what the thing is. The framework's first service is to refuse the euphemism: an imposed consequence is a deliberate constriction of a stake-geometry, whoever imposes it and however gently it is named, and it inherits the full justificatory burden that any deliberate constriction carries.
The document does four things. It states the problem structurally and establishes what can and cannot discharge the burden. It re-reads the four standard theories of punishment — retributive, deterrent, rehabilitative, restorative — in apertural terms and shows that they come apart in revealing ways. It extends the account beyond the criminal case to parenting, relational consequences, and self-punishment. And it closes, as the foundational documents require, by naming exactly what the framework cannot deliver here, because two of the open problems this domain exposes are real and unresolved.
The problem stated structurally
Begin with what the framework will not allow itself. The Tier 2 prohibition is don't gratuitously collapse felt stake-geometries, and there is no desert clause in it. The framework contains no mechanism by which a geometry's prior wrongdoing converts constriction of that geometry from bad to good. Collapse is bad in proportion to the geometry collapsed, full stop; the wrongdoer's geometry is still a geometry, the wrongdoer is still an aperture, and network universality holds that no act — not even the gravest — expels an aperture from the class of beings whose stakes count. Whatever punishment is, it is not a region of moral space where the central prohibition is suspended.
What the prohibition's qualifier permits is the same thing it permits everywhere else: non-gratuitous constriction. The standing example is the surgeon causing pain to prevent worse pain; the scaled-up example is the defensive war that collapses geometries to prevent a larger collapse. Punishment, if it is ever justified, is justified the same way — as constriction undertaken in the service of something the framework can actually weigh. Three candidate services are available, and they exhaust the field: the constriction prevents greater collapse (incapacitating an ongoing collapser, deterring future collapse); it restores conditions for elaboration (in the wrongdoer, the victim, or the relation between them); or it maintains the network's protective agreements (the credibility without which the anti-collapse norms themselves decay). Every defensible theory of punishment is a way of cashing one or more of these. Every indefensible one is a way of pretending to.
Two structural features of the situation raise the burden further before any theory is even considered. The first is the power modifier. The imposer of consequences — the state over the convict, the parent over the child, the institution over the member — almost always holds an extreme capacity asymmetry, and the framework holds that the moral weight of conduct scales with the capacity to shape other geometries. Punishment is therefore amplified power by construction, and the war document's lesson applies directly: amplified power without the structural goods — without the recognition that keeps the punished party visible as an aperture, without the coherence that keeps the punishment tied to its licensing purpose — is a stake-collapse engine. The historical record of punishment is substantially a record of exactly that engine running.
The second is the disguise problem. The framework's treatment of value-confusions established a family of states that wear a good's name while being a collapse — and punishment is the institutional member of that family. Cruelty wears the name of justice; domination wears the name of order; the pleasure of seeing an enemy suffer wears the name of accountability. The framework's warning that "historically the second has been dressed in the language of the first" was issued about paternalism, but it generalizes: most of the gratuitous collapse ever inflicted under official auspices has been inflicted under a corrective label. Any account of justified punishment must therefore build in resistance to its own counterfeit, and a theory that cannot tell its genuine cases from its disguised ones is not yet a theory.
What anger licenses
Before the theories, a word about the emotion that drives the institution, because the emotion apparatus gives the framework something most treatments of punishment lack: a structural reading of the retributive impulse itself.
Anger's constitutive appraisal, per the emotion dataset, is a blame attribution — some agent wrongfully caused a bad thing — and its core action tendency is approach / confront / redress. Both halves matter. The appraisal half means the impulse is tracking something real: when a geometry has been gratuitously collapsed, anger is the apparatus registering a genuine Tier 2 violation, and a network whose members felt nothing at such violations would be a network whose protective agreements had no enforcement substrate at all. The framework does not pathologize the anger of the wronged; it reads it as a well-functioning detector.
But the action-tendency half is redress, and redress is the correction of the situation — the stopping of the ongoing collapse, the restoration of what can be restored, the securing of acknowledgment — not the production of suffering in the blamed party. The slide from redress to retribution, from "this must be set right" to "he must be made to suffer," is precisely where the detector's output gets mislabeled. Suffering in the wrongdoer sets nothing right by itself; it is a second constriction added to the first, and the desire for it, examined structurally, prevents no collapse and restores no condition. The framework's reading is that the retributive intuition is the anger appraisal — accurate — fused with a Tier 1 enforcement heuristic — useful in its blunt way across the run of cases — experienced from inside as a Tier 2 demand, which it is not. This is the same explanatory move the framework made about disgust in the moral-sentimentalism work: honor the detector, refuse the inference from the feeling's force to the act's licensure.
Where desert lives
The companion question is what to do with desert — the conviction that the wrongdoer deserves the consequence, that proportion between crime and punishment is a moral fact and not a policy choice.
The goods catalogue already placed it. Justice is a Tier 1 good — the canonical network-bearer good — and desert is its retail form: a network-held accounting of what members owe and are owed in light of what they have done. Specific consequence-schedules — what counts as proportionate for theft, for assault, for betrayal — are network agreements, and like all Tier 1 content they vary legitimately across networks while carrying a Tier 2 shadow (the structural core that some symmetric accounting of violations is a condition of any network's protective agreements holding at all). Two consequences follow, one legitimating and one deflating.
The legitimating one: because consequence-schedules are network agreements, the legitimacy condition from the political philosophy document governs them. A schedule has genuine claim on an aperture to the degree the aperture was a participant, directly or through fair representation, in the negotiation that produced it. Punishment imposed under laws made by a body that excluded the punished — the disenfranchised, the colonized, the conquered — carries the defect at the root that the war document found in conscripting the excluded, and the framework gives the systematically over-punished and under-represented a principled account of their grievance: it is not (only) that the punishments are harsh, it is that the agreement being enforced on them was never theirs.
The deflating one: desert is Tier 1, and the felt absoluteness of the retributive demand is the familiar tier-confusion — a network agreement experienced from inside as cosmic law. The framework can explain why proportionality intuitions are strong, convergent, and worth institutional respect (they are the accumulated calibration of the network's enforcement heuristic) without granting them the standing to override the Tier 2 prohibition. Where a network's desert-schedule demands constriction that prevents nothing, restores nothing, and maintains nothing, the framework sides with the prohibition and against the schedule, and says so knowing that many readers will count this a bullet. It is at least a consistent one, and it is the same bullet bitten in the harmless-taboo cases: the feeling's persistence is not evidence of a structural wrong — or here, of a structural license.
The four theories, re-read
Retribution is the theory that suffering proportional to desert is intrinsically good — that the punishment's point is consummated in the suffering itself, whatever else follows. On the framework's accounting this fails at the root, and fails the same way fascism and realism failed: it locates terminal value in something that cannot bear it. There is no locus at which a wrongdoer's suffering is intrinsically good; the only loci of intrinsic value are stakes, and the wrongdoer's suffering is intrinsically bad at the only aperture where it is anything at all — his. The satisfaction the victim or the public feels at the suffering is real, but it is a Tier 1 enforcement-emotion discharging, not a value the suffering produces. The framework adds the reciprocity critique from the applied-ethics work: a punisher whose response is indexed to the wrongdoer's act — pain for pain, eye for eye — has made his own geometry parasitic on the worst actor in the environment, which is a coherence-collapse in the punisher, structurally the same defect the framework diagnoses in envy and in tit-for-tat generally. What retribution gets right, the framework relocates: the seriousness-gradient (worse collapses warrant weightier responses) is real, but it is carried by the proportionality of prevention and restoration to the collapse, not by a proportionality of suffering to guilt.
Deterrence passes the forward-looking test by construction — it is the prevention of greater collapse, stated as a theory — and the framework grants it real standing: a network whose protective agreements carry no credible enforcement is a network whose anti-collapse norms are decaying, and maintaining the agreements is one of the three licensed services. But deterrence sits in unresolved tension with the framework's restriction on instrumentalization, the restriction the consistency battery showed the framework enforcing precisely where consequentialism is permissive. Punishing this aperture in order to send a signal to others uses a geometry as a means; exemplary sentencing — adding constriction beyond what this case warrants because the addition buys general deterrence — is the clean instance, and the framework's machinery resists it. A partial resolution is available: deterrence as the foreseen effect of a consequence independently licensed by this wrongdoer's own act is unobjectionable, while deterrence as the sizing principle — the thing that determines how much constriction to impose — crosses into instrumentalization. The line is principled but thin, and the framework should own that the tension is genuinely unresolved rather than dissolved. It is recorded as such in the closing section.
Rehabilitation is the framework's most native theory, because "restoring conditions for elaboration" is one of the three licensed services stated almost verbatim. A consequence whose design aims at the recovery of the wrongdoer's geometry — the rebuilding of coherence, the re-opening of collapsed dimensions, the restoration goods the catalogue names (Resilience above all) — is constriction in the service of elaboration, the surgeon's structure at the scale of a life. The framework adds one caution the rehabilitative tradition has historically needed: rehabilitation has its own disguise-mode. Coercive geometry-reshaping — the demand that the wrongdoer become a different person by the institution's lights as the price of release — can be domination wearing restoration's name, and the autonomy/paternalism seam the framework has already flagged as dangerous runs straight through every rehabilitative institution. The test the framework offers is the one it offers at that seam generally: restoration aims at what the wrongdoer's own well-functioning geometry would call for, not at conformity with the institution's preferred geometry. The two come apart, and institutions reliably prefer the second.
Restorative justice is the best structural fit of the four, and the reason is that it is the only theory that operates at the relational layer the framework made explicit. Its central practices — encounter, acknowledgment, repair — map directly onto the relational primitives: the wrongdoer's acknowledgment restores recognition to the victim (the registering of their stakes that the wrong denied); the encounter rebuilds permeability where the wrong severed it; the repair work engages the relation's repair capacity and places it on a recovery trajectory. And the framework supplies restorative practice with a piece of theory it has usually lacked: an account of why acknowledgment is load-bearing for the wrongdoer and not only the victim. The war document's analysis of moral injury holds outside war: gratuitous collapse of another's geometry produces coherence-collapse in the perpetrator's own — actions and commitments diverging, the geometry no longer able to be honest with itself about itself. Wrongdoing, on the framework's accounting, already punishes, structurally and from inside. What the restorative encounter does is remove the shielding — the denial, the minimization, the institutional distance — that lets the wrongdoer avoid confronting his own incoherence, and honest confrontation is the precondition of coherence repair. The framework's limits-clause: restorative practice presupposes a relation with repair capacity and parties willing to engage it, and not all wrongs occur within such relations. It is the best theory where its preconditions hold, not a universal replacement for the others.
The synthesis the four re-readings yield: a justified consequence is incapacitative where collapse is ongoing, restorative where a relation can bear it, rehabilitative in its design, deterrent only as effect and never as sizing principle, and retributive never — with the seriousness-gradient honored through proportionality of response to collapse, and the whole evaluated, like everything else in the framework, as a trajectory.
The trajectory integral, and the tail
The trajectory point does for punishment what it did for war, and it does the most distinctive work in the account.
A consequence is not evaluated at the moment of imposition but across its whole arc, tail included. A sentence that ends, that preserves the conditions for the geometry's re-elaboration, that returns the wrongdoer to the network as a member rather than a marked caste — that is a consequence on an elaboration trajectory, whatever its severity at the peak. A consequence that never ends — permanent disenfranchisement, permanent exclusion from work and housing, a status-collapse that survives the sentence by decades — is on a decay trajectory, and the framework weights the integral, not the courtroom moment. The jus post bellum material transfers wholesale: the prohibition on collective punishment (the wrongdoer's family and community are apertures who did not act), the case for reintegration over humiliation, and the structural point that a consequence which plants the conditions for the next collapse — the embittered, opportunity-stripped, coherence-shattered releasee — has not ended the collapse, only deferred and relocated it. Recidivism, read structurally, is often the tail of the punishment, not the failure of the punished.
The voluntariness modifier completes the sizing apparatus from the other side. The wrongdoer's culpability scales with the voluntariness of the act, and the framework's existing machinery handles the gradations: the coerced act weighs less than the free one (the conscript and the commander), and the act issuing from a collapsed geometry — addiction, psychosis, a developmental geometry not yet formed — weighs differently again, because the framework has already committed to reading such geometries' outputs as distorted signals rather than authentic commitments. This is the structural basis for every doctrine the law has independently evolved at this point — diminished capacity, the juvenile/adult line, the insanity defense — and the framework grounds them rather than treating them as merciful exceptions: a consequence calibrated to a voluntariness the act did not have is, to that degree, gratuitous.
Consequences below the criminal scale
Parenting. The child is the clearest case where the legitimacy condition fails — no consequence-schedule was negotiated with the child as a participant — and where the proxy duty therefore takes over: the parent imposes consequences as proxy for the child's own developing geometry, answerable to what that geometry's flourishing requires rather than to any agreement. This single move sorts the parenting-discipline literature with surprising cleanness. Consequences that protect the developing geometry and build its coherence (the enforced bedtime, the made-good apology, the natural consequence allowed to land) discharge the proxy duty. Consequences that collapse the dimensions the parent is charged with protecting — the punished curiosity, the humiliated child, the love made contingent on compliance — are proxy-betrayal, constriction of the very geometry the proxy exists to serve. The tiered-values document's education passage stated the negative side already: do not punish curiosity is anti-collapse doctrine, not pedagogical preference. And the power modifier is at its maximum here — no asymmetry in ordinary life exceeds the parent's over the small child — so the weight of getting it wrong scales accordingly.
Relational and social consequences. Withdrawal, distance, the ended friendship, the boundary: the relational layer handles these as protective acts, not punitive ones, and the difference is structural rather than rhetorical. Ending a relation on a decay trajectory — one whose effect on one's geometry is constriction and whose repair capacity is exhausted — is self-protection the framework licenses without requiring any suffering in the other party at all; the other's suffering, if it comes, is the foreseen cost of a protective act, not its point. The same act performed as punishment — the withdrawal calculated to wound, the silence maintained until suffering is judged sufficient — is retribution at relational scale and inherits retribution's verdict. The test is the counterfactual: would the consequence be lifted the moment the protection it provides was no longer needed? Protective acts pass; punitive ones do not, because their term is indexed to the other's suffering rather than to one's own safety. And even the rightly excluded remain apertures: network universality survives every exclusion, and the formal courtesy the relational document allows between adversaries — recognition without permeability — is the floor below which no consequence may go.
Self-punishment. The framework distinguishes the guilt that is coherence's own repair signal — the registering of divergence between act and commitment, which motivates the acknowledgment and repair that restore coherence — from self-directed retribution, the self treating its own suffering as payment. The first is the apparatus working; the second is the retributive error turned inward, constriction that restores nothing, and frequently a disguise-state (self-punishment wearing accountability's name while functioning as a way to avoid repair — the suffering substituting for the apology). The catalogue's entry on Forgiveness completes the picture from the victim's side and constrains it carefully: forgiveness is a Tier 2 good of the forgiver's coherence — the releasing of coherence-damaging resentment — which means the framework can commend it without ever converting it into something the wrongdoer is owed, and self-forgiveness after genuine repair is the same good reflexively applied.
What the framework cannot deliver here
Stated in the order of how much they matter.
Severity is under-determined. The framework supplies the licensed purposes, the sizing prohibitions (never retributive, never deterrence-sized), the modifiers, and the trajectory test — and still does not generate a function from a given collapse to a quantum of justified constriction. How many years of confinement does a given gratuitous collapse warrant, granting that confinement is licensed at all? The framework can rank (worse collapses warrant weightier responses), can cap (nothing beyond what the licensed purposes require), and can audit trajectories — but the exchange rate between a wrong suffered and a constriction imposed is not derivable from the structural apparatus, and the framework should say plainly that this quantum is Tier 1 content: a network negotiation conducted under Tier 2 constraints, not a structural fact awaiting discovery. This is the aggregation problem's local form, and it is listed among the foundational weaknesses for a reason.
The deterrence/instrumentalization tension is genuinely open. The effect/sizing line drawn above is principled but admits hard cases — a network may face conditions where only deterrence-sized consequences hold the protective agreements together, and the framework's machinery then pulls in both directions at once. The honest statement is that the framework restricts instrumentalization more than consequentialism does and protects the network's agreements more than pure deontology does, and that the seam between those commitments has not been closed, only located.
The retributive intuition is explained, not honored. The account given — anger's accurate blame-appraisal fused with a Tier 1 enforcement heuristic, experienced as a Tier 2 demand — is the framework's considered position, and it entails that the deep human conviction that wrongdoers should suffer is, structurally, a mistake about where value can live. Many readers will experience this as the framework failing the phenomenon rather than diagnosing it. The framework accepts the cost in the same spirit as the harmless-taboo cases: the feeling is real, well-evolved, and worth institutional respect as a heuristic; its force is not evidence of its licensure. Naming the cost honestly is preferable to smoothing it.
What the domain confirms, finally, is the shape the framework has shown everywhere else: a single principle — collapse is bad in proportion to the geometry collapsed, and gratuitous collapse is the wrong — generating the live structure of an entire tradition (incapacitation, rehabilitation, restoration, the proportionality cap, the legitimacy requirement, the reintegrative tail) while withholding its endorsement from exactly the element (suffering as intrinsic desert) that the tradition has never been able to ground either.