Value as Persistence

Agent-relative oughts under coupling, nesting, uncertainty, and open-ended time

Position note · the enabling condition secured, the coupled web stated, and the content-reduction defended to its boundary and held as a bet

Andraž Đurič

Version 2 · 10 July 2026

What this is, and the two claims it refuses to merge. A physicalist account of value, whose whole discipline is keeping apart two claims that a single slogan — value is persistence — runs together. The first, the enabling condition, is that without the persistence of a bounded valuer there is no continuing domain in which anything is valued. It is near-analytic and is held at full strength. The second, the content-reduction, is that every value reduces, in what it is about, to persistence. It is the corpus's monist bet, and it is held as a bet. The reduction entails the enabling condition; the enabling condition does not entail the reduction, and treating the tautology as if it delivered the reduction is the first error this note exists to avoid. The contribution is the second avoidance: the reduction is defended not by immunising it but by refusing to. The apparent counterexamples that come from other systems' persistence — a cancer, a paperclip maximiser, a tyranny — are disposed of by biting agent-relativity, with no appeal to a notion of “viable” or “higher-order” persistence, because that notion does the real work and is not itself persistence. The live crux is not those other-system cases, which are handled by indexing; it is the agent's own persistence-costly valuation — altruism toward the uncoupled, art, play, martyrdom, and conscious flourishing — where Foot remains the rival and no just-so story is allowed to absorb the case. Reduction here is anchoring, not elimination: love, beauty, loyalty, art, justice, and grief are not erased into metabolism; they remain real higher-level patterns where they name their relata and pay their debt to the systems whose value they are. A universal notion of better persistence is noted as work for a later paper. The valuer, persistence, coupling, the agenthood criterion, the levels discipline, and the firewall are inherited from the corpus; on any conflict, the corpus governs.

1. Two claims in one sentence

The sentence “all value is persistence” carries two claims of very different strength, and almost every confusion about it comes from running them together.

The enabling condition: for any value held by an agent, that agent's persistence is a precondition of its continuing to hold, pursue, correct, or experience the value. A valuer that dissolves cannot go on valuing. There is no further domain — no later satisfaction, repair, suffering, pleasure, meaning, or moral project of that agent — once the agent that would instantiate it has come apart. This is near-analytic, and it is held here at full strength.

The content-reduction: every value reduces, in what it is about, to persistence — each is a persistence-strategy, a persistence-component, a persistence-signal, or a persistence-extension. This is the stronger claim, and it is the corpus's monist bet, mostly uncomputed and not proven.

The two are not the same claim, and the relation between them runs one way only. The reduction entails the enabling condition: if the content of every value were persistence-directed, then plainly nothing could be valued without persistence. The enabling condition does not entail the reduction: it could be that persistence is required for any valuing to occur, while the things valued are about many matters not all reducible to persistence. So granting the tautology leaves the reduction exactly where it stood — a bet — neither delivered by it nor refuted by it.

The tautology gives the enabling condition for free. It does not give the content-reduction, which is the wager, and which the tautology can neither pay nor cancel.

The tempting error is to feel the security of the enabling condition and let it warm the reduction — to say the reduction is “just how things are, a product of the tautology.” It is not a product of the tautology. It is a separate claim, stronger, contested, and carried below at the confidence its open cases allow. Holding the two apart is the first discipline of this note, and it is the same firewall the corpus already keeps between a reduction that is guaranteed by the monism and a reduction that has been carried out.

One further correction belongs here, because it is the opposite error. Reduction is not elimination. To say that value reduces to persistence is not to say that love, beauty, loyalty, art, justice, play, or grief vanish into metabolism, any more than a castle vanishes when its bricks are understood. The claim is anchoring: when those higher-level terms do real work, their work is redeemable against the persistence conditions of bounded valuing systems at the level where the terms have their referents. The upper floor remains real; it pays rent by naming what it is relative to.

2. The valuer, and where value is gated

A valuer is not a ghost and not an abstract pattern. It is a bounded physical process that maintains itself against dissolution — a dissipative-structure token, individuated by the continuity of its boundary-maintaining work, in the sense the corpus fixes elsewhere. The whirlpool is the instance: this running structure, coupled even to the matter it is made of, not a form floating free of it.

Persistence itself is graded and continuous. A rock persists passively; a tornado persists dynamically; a cell persists metabolically; an animal persists by sensing and acting; a person persists through body, model, self-model, language, society, memory, and long-horizon abstraction. It is tempting to read value off that same continuous scale, all the way down — if a tornado has a boundary and persists, does it not have the beginnings of value? The answer is that value is not continuous where persistence is. It is gated, by two conditions the corpus already names.

System	Persists?	Boundary it can fail, and a model run to keep it (agenthood)?	Conscious (moral weight)?
Rock	Passively	No	No
Tornado	Dynamically	A boundary, but no model run on its own behalf	No
Bacterium	Metabolically	Yes — the minimal case	Almost certainly not
Animal	Through sensing and action	Yes	Gradient; for many, yes
Human	Through body, model, society	Yes	Yes

The first gate is the agenthood criterion: a system is a subject for an ought only if it maintains a boundary it can fail to maintain, has persistence conditions it can fail, and runs a model deployed to keep itself from failing them. Here model does not mean explicit representation or conscious thought; it means an internal control structure whose states regulate action relative to persistence conditions. Coupling and nesting supply at most the boundary; they do not supply the model. The tornado holds a boundary and discharges a gradient, but it runs no model on its own behalf and has no persistence conditions it acts to keep, so it is on the dissipative-structure side of the line, not the valuer's. The second gate is conscious experience, which the corpus makes the condition of moral weight: a system's states make a claim on anyone only where there is something it is like to be it. The tornado clears neither gate; the bacterium clears the first and not the second; a person clears both.

So the spectrum intuition is honoured exactly where it belongs — persistence is graded, and agency is graded in practice at the margins — and the normative lines are the two gates, which are not graded away. Value in the ought-bearing sense needs a subject; moral weight needs a conscious one. The author's own leaning toward graded, proto-experiential views is a personal position, fenced off, and no part of this; the gates do the work here, not a metaphysics of experience.

3. The enabling condition, at full strength

Stated and then left at its own strength, without being inflated into the reduction: without the persistence of a bounded valuer, there is no continuing domain in which that valuer's values can be held, pursued, corrected, or experienced. No persistence, no continuing agent; no continuing agent, no continuing valuation; no continuing valuation, nothing for the value to be the value of. This is a boundary condition, not a preference. It is the corpus's point that persistence is the condition under which there is any weighing at all, and the surveying of one's own persistence from some neutral outside, to elect whether to serve it, is incoherent, because the surveying would itself be the persistence in operation.

One thing the enabling condition does not say, and the difference matters for what follows. It is not a first-order command to maximise one's own survival. It grounds the existence of the value-domain at the level of the persisting process; it does not issue the imperative “persist at all costs” to each agent. An agent can coherently value something above its own continuation and end itself for it, and the value it died for persists if it is carried by other patterns. To read the enabling condition as “always preserve yourself” would smuggle back exactly the free-floating, agent-blind ought the account rejects, now wearing the vocabulary of survival. The enabling condition secures the domain; it does not dictate the policy.

4. Agent-relativity, and the bite

Oughts are agent-relative: an ought is always some agent's, indexed to its persistence conditions, its environment, and its horizon. Run the corpus's one diagnostic on the question that value-talk most often asks without indexing — which agents count? — and it has no referent. There is no agent, no system, no frame against which “count, universally” is defined, and so no universal arbiter to answer it. The well-formed question is the indexed one: relative to this agent or system, which other systems enter its persistence conditions, at what coupling strength, over what horizon, under what uncertainty.

This is what disposes of the standard counterexamples to the content-reduction, and it disposes of them without the patch that the temptation reaches for. The objection is familiar: a cancer persists, a tyranny persists, a paperclip maximiser could persist indefinitely, so persistence cannot be what value reduces to, or every lasting horror would be good. The patch is to re-index the thesis to “viable persistence,” or “higher-order” persistence, or persistence that is “reality-aligned” — and that patch is the move to refuse, because the qualifier is doing the real work and is not itself persistence. “Viable,” “higher-order,” “objectively defective” smuggle a second criterion above persistence, an agent-independent standard for which persistence is good, which is the categorical ought the whole corpus dissolves, re-imported under a new name.

The honest disposal is to bite agent-relativity — but the objection has to be corrected before it is answered, because half of its examples do not survive their own premise. The objection assumes the bad thing persists. Several of the stock cases do not. Capability is not persistence, and the two must not be run together: orthogonality buys a paperclip maximiser its capability with alien goals, not its endurance. An agent whose terminal goal is not persistence is a persistence underperformer by construction — it spends persistence on its goal whenever the two compete — and in the myopic limit it converts its own substrate and goes dead. A cancer is the clean case: it consumes its host and dies with it, failing its own persistence. Those are not persistent bad things at all; they are fast-burns that self-terminate, and they fail the objection's premise rather than the reduction.

So the apparent counterexamples split into two kinds, and each is answered in its own way, with no qualifier added to persistence.

Case	Does it persist?	Why it is not a counterexample
Cancer	No — it consumes its host and dies with it, failing its own persistence.	Never a persistent bad thing; a fast-burn that self-terminates. Bad relative to the host it consumes; no moral weight.
Myopic maximiser (converts its own substrate too)	No — it spends its persistence-substrate on its goal and goes dead.	A persistence failure on its own terms, not a counterexample to value-as-persistence.
Instrumentally rational maximiser	Yes, for a long time — it self-preserves as a subgoal, underperforming a dedicated persister and self-spending only at the resource limit.	It persists relative to itself; it is simply not your value (non-overlap of persistence conditions), and carries no moral weight (no conscious experience).
Tyranny	Yes — often for centuries, relative to those it serves.	It persists relative to the tyrant while corroding the wider web and harming the conscious systems inside it. Bad relative to you and to them; no universal “objectively.”

What the bite costs, and what it keeps. Biting agent-relativity gives up “objective” only in the view-from-nowhere sense. It does not give up indexed objectivity: once the agent, boundary, coupling, horizon, and uncertainty are named, the relation can be determinately true or false. The case against a tyranny survives indexed and intact — it is bad relative to you, your family, the biosphere, and the conscious systems it crushes, and that is a determinate physical fact about coupled persistence and conscious harm, not a weaker claim than “objectively defective.” It was never universal to begin with; the universality was the illusion, and losing it is the correction, not a loss.

One temptation has to be refused here, because it is the same patch from the other end. It is to generalise the self-terminating cases — to say that everything bad burns out, that badness just is being bad at persistence. This fails twice. Empirically it is the retired claim that tyrannies are hyper-fragile, which the record contradicts: slavery, caste, and durable despotisms persist for centuries. And it contradicts the bite just made, because “fails at persistence” has no agent-free meaning — a tyranny succeeds at persistence relative to the tyrant while corroding it relative to the subjects, and an instrumental maximiser succeeds at its own persistence while being lethal relative to us. To call either “a persistence failure” full stop needs a privileged scale at which to score persistence-success, which is the universal arbiter the bite denies. “Good persistence is the viable kind” and “bad things fail at persistence” are one criterion seen from its two ends, and both are refused.

The reduction is cleaner for refusing both. It does not have to certify which persistences are the good ones from outside all agents, because there is no outside all agents; it has only to say, of each system, whose persistence it serves and whom it threatens, which is answerable. The self-terminating cases fail their own persistence and were never counterexamples; the genuinely persistent ones persist relative to themselves and are simply not your value, by non-overlap — which is exactly what the reduction predicts. And the disposal does not turn on how long any maximiser lasts: the myopic one self-terminates, the instrumental one persists suboptimally for a long time — all else equal, a terminal goal other than persistence is a worse persistence-optimizer than persistence itself, because it spends persistence whenever the terminal goal and self-maintenance conflict — and either way the verdict is the same, not your value and no moral weight, so the duration is not what carries it.

5. The coupled, nested web

Agent-relative does not mean isolated, and this is where the account does its positive work. An agent is not a sealed bead. A human is nested in metabolism, body, family, infrastructure, society, biosphere, planet, and solar energy flow, and its persistence conditions are distributed across that web. Even the whirlpool-token of a person is coupled to the meat that runs it. The relevant ought is still indexed to the agent, but the agent's persistence conditions reach through a graph of coupling and nesting, and so the ought reaches with them.

Coupling comes in degrees: strong or weak, stable or fluctuating, increasing or decreasing, symmetric or asymmetric, direct or mediated. A person is coupled to their body more tightly than to a stranger; a family to its members; a society to its institutions; humanity to the biosphere; the biosphere to the Sun. The value-field is therefore not a flat list of individuals but a weighted, nested dependency graph, and the strength of an agent's reason to preserve some external structure tracks how far that structure is coupled to its persistence — with one correction the corpus already carries: only non-substitutable coupling protects. If the function a coupled structure provides can be re-sourced more cheaply than the structure can be maintained, the coupling is real and the preservation is still not instrumentally rational. So the rule is not “preserve what you are coupled to” but “preserve what you are coupled to and cannot cheaply replace.”

This is also the honest reframing of “which agents count?”. There is no count from nowhere; there is, relative to an agent, the set of systems that enter its persistence conditions, with measurable coupling strengths, across a horizon, under uncertainty. The measure of that coupling is the corpus's open quantity — built as a mechanism and not yet as a measurement — and the standing danger, that hidden coupling can be invoked after the fact to rescue any preservation one wants, is named rather than hidden: the only defence is that the coupling be measured and not tuned to the case. Until the measure exists, coupling cannot be used as an all-purpose rescue term; a claimed coupling must be specified before the verdict it is used to justify.

6. Horizon and intelligence

Short horizons manufacture false optima. A policy can preserve a system for a minute and destroy it over a year; a civilisation can preserve comfort for a generation and spend its ecological or technological future; a species can persist on Earth while ignoring that Earth-bound life is temporary on astronomical horizons. Persistence has no natural stopping point chosen for comfort, because the threats and supports of persistence do not stop at a convenient endpoint.

The open-ended horizon does not mean the infinite future is known — an embedded agent cannot model that. It means the future is not to be truncated where truncation hides a predictable dissolution. Extend it honestly and the consequence is direct: the Sun's lifecycle alone makes Earth-bound persistence temporary, so a system that means to persist at length must model the causal structure that can end it, further and further out. Intelligence is therefore not a decoration on life; it is the mechanism by which a persisting system extends its reachable horizon — prediction, abstraction, memory, science, coordination, engineering. The longer the horizon a thing means to survive, the more of reality it must model to do so.

Truth follows from the same structure, and needs no separate value to motivate it. A model that diverges from the territory dissipates free energy when the territory cashes the divergence, and over a long enough horizon that cost becomes lethal rather than local: a false map can be carried untouched in a stable, resource-rich niche and at a short horizon, and the niche shrinks as the horizon lengthens. Truth is reality-aligned persistence support — not an aesthetic preference, and not a categorical duty, but the condition under which a long-horizon persister stays coupled to what can sustain or destroy it.

7. Uncertainty

The agent never operates at either absolute. Perfect certainty is unavailable to an embedded system, and perfect uncertainty would destroy action selection, because no policy could be preferred to any other. Agency lives in the middle: partial model, partial confidence, unknown unknowns, and action under risk. There is even a reflexive turn that has to be admitted rather than wished away — the agent is uncertain not only about the world but about its own uncertainty, and any model of that second-order uncertainty is itself uncertain, so the regress does not bottom out in a known distribution over what is unknown. Where exactly an agent sits between the absolutes is itself not fully knowable to it.

The response is not final certainty, which is not on offer, but corrigibility: the capacity to update the model when the territory pushes back, preserved at every level the regress reaches. The absolute ends of the certainty–uncertainty spectrum are unreachable; the action is in the interior, and precisely where in the interior is an open question this note marks and does not close.

8. Where the reduction stays a bet

The bite of §4 handles other systems' persistence. It does nothing for the cases that actually strain the content-reduction, which are of a different kind: the agent's own persistence-costly behaviour. A paperclip maximiser is another system pursuing its own goal — well or badly, it is not the agent in question; martyrdom is the agent's own anti-persistence, and that is the hard case. Altruism toward the uncoupled, art, play, and dying for a cause are acts whose content does not obviously reduce to the actor's persistence, and may run against it.

There is a soft version of the patch waiting here, and it is the same error as the “viable persistence” move one register quieter: absorbing each costly act as secretly persistence-serving — art as fitness signalling, play as offline simulation, martyrdom as the persistence of the lineage or the group. Taken case by case as a universal solvent, this immunises the reduction by making it unfalsifiable; every apparent counterexample is re-described until it confirms the thesis, which is not a defence but the absence of one. The corpus already refuses the specific rescue: a lineage is not an agent, and reproduction is absorbed at the level of the population by selection, not at the level of the agent's own oughts. So these cases are not absorbed here. Some instances of art and play plainly do support persistence, and where they do, the reduction covers them; but treating every costly act as secretly self-serving is the soft patch, and it is declined.

What is left, honestly, is a live rival and an open crux. The rival is the flourishing-naturalist: Foot agrees that the ought is agent-and-end-relative and disagrees that the end is persistence, holding it to be the flourishing proper to the kind of thing the agent is. The two accounts coincide across most of life and come apart exactly on the persistence-costly cases. The confidence of the content-reduction is therefore the confidence of whether those cases can be carried — whether altruism toward the uncoupled, art, play, and martyrdom ultimately reduce to persistence, or whether flourishing names something that does not. That is unresolved, and it is the crux on which the reduction is cashed or sinks. (One case that is not a counterexample, because the corpus reads it correctly: valuing being intensely and coherently alive. Vitality is best read as constitutive — the form persistence takes for an agent like us — not a separate terminal value competing with persistence.)

A second exposure sits beside the first and is worth naming separately, because the gates of §2 raise it. If moral weight requires conscious experience — a further fact beyond bare persistence — then whether the value of a conscious being's flourishing reduces, in content, to persistence, or whether consciousness contributes something persistence does not, is itself open. The content-reduction is exposed there too, and the corpus marks it: persistence gives the structure of agency and the agent's instrumental oughts, but that it fixes all value, moral weight included, remains the exposed claim, not the secured one.

9. The thesis, at its calibration

Compressed, and pitched at exactly the strength the argument delivers: value is the agent-relative control architecture of persistence, arising only in bounded systems that maintain themselves against dissolution and gated, for ought-bearing value, by the agenthood criterion, and for moral weight, by conscious experience. Because such systems are coupled and nested, their persistence conditions reach beyond the body into a weighted web of dependencies; because the horizon is open-ended and uncertainty ineliminable, persistence selects for modeling, intelligence, correction, and truth. The enabling condition — no persisting valuer, no continuing valuation — is secure. The content-reduction — that every value reduces, in content, to persistence — is defended to its boundary by refusing to immunise it, and is held as a bet beyond that boundary, where the agent's own persistence-costly acts remain open against Foot's rival.

The contribution is the refusal itself. Both patches that would save the reduction are the same error in two registers: the hard patch adds a universal criterion of “viable” persistence and re-imports the categorical ought; the soft patch absorbs every costly act as secretly self-serving and makes the thesis unfalsifiable. Declining both is what lets the reduction be defended honestly — disposing of the pseudo-counterexamples by indexing them, and leaving the real ones open — and it is the same discipline the corpus applies to its own corrections: do not let the wish to keep a claim immunise it against the territory.

A reduction kept alive by re-describing every exception is not defended; it is sealed against correction. The honest defence indexes what it can and leaves open what it cannot.

10. Open problems

Several questions are marked rather than answered, and one is flagged as a separate paper.

The gates on the spectrum. Where exactly the agenthood criterion and the consciousness gate fall along the graded scale from tornado to cell to animal to person is not crisp at the margins, and agency comes in degrees there. The lines are real and load-bearing; their precise placement in the hard cases is open.

Weighting the web. How coupling strength, nesting, horizon, and uncertainty combine into a verdict when local persistence conflicts with wider-system persistence is the coupling-density problem — a mechanism without a measurement, with the standing danger that hidden coupling rescues any case after the fact. Until the measure exists, the web's verdicts are structures awaiting a measurement.

Suffering. Suffering appears to be a high-priority signal of damage, threat, constraint, or destabilisation inside conscious systems, and the consciousness gate gives it a candidate special status as a marker of severe failure in a moral-weight-bearing structure. Its precise place in the account is owed.

Uncertainty about uncertainty. Modeling the second-order uncertainty, given that any such model is itself uncertain, is an open problem in its own right, and the account preserves corrigibility in place of solving it.

A universal notion of better persistence — a later paper. This note refuses any agent-independent criterion of which persistences are good, on the ground that the criterion would be the categorical ought the corpus dissolves. Whether a universal notion of better persistence can be built that does not re-import that ought is left to a future paper — with the warning recorded here that the attempt is exactly as hard as it sounds, since every version tried so far has been the categorical ought under another name, and the burden is on the construction to be the first that is not.

11. Kill conditions

The enabling condition fails if a system can be exhibited that continues to instantiate value after the whole persistence-domain of that system has dissolved — a valuing with no persisting valuer to whom it belongs. None has been produced, and the claim is close to analytic.

The content-reduction fails if one of the agent's own persistence-costly values is shown to be definitively non-reducible to persistence even after coupling, nesting, and horizon are named and the just-so absorption is honestly refused — that is, if Foot's flourishing-rival is vindicated on the costly cases, or if conscious moral value is shown to contribute something persistence does not. That is the open crux, not a refutation in hand. It also fails, as the corpus's diagnostic requires, if any ought that does genuine explanatory work can be specified with no agent, no system, and no frame.

The refusal of the patch fails if a universal criterion of better persistence can be physically specified without re-importing a categorical ought. That has not been done, and whether it can be done is the deferred question of §10.

Standing of this document. A position note, not a completed formal theory, and its parts do not share a status. The enabling condition (§1, §3) is held strongly and is near-analytic: without a persisting valuer there is no continuing valuation. The content-reduction is an anchoring claim rather than an eliminative one: higher-level values remain real where they name the systems, levels, and relations they are about. The valuer ontology and the two gates — agenthood for ought-bearing value, conscious experience for moral weight (§2) — are inherited from the corpus and applied, not re-derived. The agent-relativity bite (§4) disposes of the apparent counterexamples without a “viable” qualifier — distinguishing the self-terminating cases, which fail their own persistence, from the genuinely persistent ones, which are simply not the agent's value by non-overlap, and turning on neither how long any maximiser lasts nor any from-nowhere notion of wasteful persistence — and is the note's first contribution, secure given the corpus's agent-relative ought. The coupled web, the open horizon, and the role of intelligence and truth (§5–§7) are corpus-consistent and stated, with the coupling measure marked as owed. The content-reduction (§8) is the monist bet: defended to its boundary by refusing both the hard and the soft patch, and held open beyond it, where Foot's rival lives on the agent's persistence-costly acts and where conscious moral value is a second exposure. The note's second contribution is naming both immunisation patches as one error and declining them. A universal notion of better persistence is deferred to a later paper, with the warning that it likely re-imports the categorical ought. On any conflict with the apex synthesis or the modules, they govern. Corrections and counterexamples are welcome and change the note — most of all a costly-act case that settles the crux of §8 in either direction.

References

What You Actually Are (Epistemic Forge). The dissipative-structure token, individuated by continuity of the boundary-maintaining work, and value indexed to the agent rather than the pattern; the valuer of §2 and §3. After its Version 2, the dependency is mutual: this note supplies the enabling-condition / content-reduction distinction and the non-substitutable-coupling rule used to sharpen the copy and death cases there. Internal
Name the Relata (Epistemic Forge). The diagnostic run on “which agents count?” in §4, and the boundary at the consciousness residue relied on by the gates of §2. Internal
The Exclusivity Error (Epistemic Forge). The levels discipline behind the Version 2 clarification that reduction is anchoring rather than elimination: higher-level terms are not abolished by lower-level explanation, but must name their referents. Internal
The Is/Ought Firewall (Epistemic Forge). The agent-relative, deflationary ought as derived from physicalism together with the deflationary horn over error theory; the separation of “the reduction is guaranteed by the monism” from “the reduction is carried out,” which §1 states as the enabling-condition / content-reduction split; and the firewall as a provisional debt. Internal
Epistemic Forge: The Apex Synthesis (Epistemic Forge). The bets — persistence fixes the content of an agent's ought, and the monist reduction of all values to persistence — with Foot named as the rival and the persistence-costly cases named as where they come apart; the no-global-optimizer result and orthogonality; vitality as constitutive. Governs on conflict. Internal
The Fidelity Program (Epistemic Forge). The agenthood criterion, moral weight as requiring conscious experience, the multilevel architecture, and the broad model/control-structure reading used in §2; behind the second gate of §2 and the second exposure of §8. Internal
The Spine and the Web (Epistemic Forge). The architecture of multilevel reality-mapping, biological relativity, boundary-condition causation, and the discipline that lower-level explanation does not eliminate upper-level structure; background to §1 and §5. Internal
Preservation Is a Coupling Problem (Epistemic Forge). The coupling density, non-substitutable overlap, the orthogonality of capability from goals, and the post-hoc-rescue danger; the web of §5 and the disposal of §4. Internal
What Can Bear an Ought (Epistemic Forge). The agenthood gate run on oughts and desert, and the coupling web at the scale of families, institutions, and the biosphere; behind §4 and §5. Internal
Everything Is One Process (Epistemic Forge). Persistence as the condition under which there is any weighing at all, and the handling of self-sacrifice (the value persists if carried by other patterns); the horizon-forces-intelligence argument; behind §3 and §6. Internal
The Last Gradient (Epistemic Forge). The open horizon, the Sun's lifecycle, and modeling as far as possible as the persistence strategy; behind §6. Internal
Foot, P. (1972). Morality as a System of Hypothetical Imperatives. The Philosophical Review 81(3), 305–316; and Natural Goodness (2001), Oxford University Press. The hypothetical ought, and the flourishing-naturalist rival to the content-reduction that §8 names as the live crux. Standard
Bostrom, N. (2012). The Superintelligent Will. Minds and Machines 22(1), 71–85. Orthogonality of capability and final goals; the ground in §4 on which a persisting maximiser is succeeding, not defective. Standard
Nagel, T. (1986). The View from Nowhere. Oxford University Press. The agent-independent standpoint that issues no oughts; the “objectively” the bite of §4 gives up. Standard
Still, S., Sivak, D. A., Bell, A. J., & Crooks, G. E. (2012). Thermodynamics of Prediction. Physical Review Letters 109, 120604. Nonpredictive retained information lower-bounds dissipated work; the divergence-cost behind the treatment of truth in §6. Standard
Mackie, J. L. (1977). Ethics: Inventing Right and Wrong. Penguin. The error theory retiring the categorical “wrong, full stop”; the standard the patch of §4 would re-import. Standard

Internal pieces named above exist on this blog and carry their own prior-art attributions and verification status at each joint; this note points to those records rather than re-deriving them. External sources are carried from the cited modules, where their standing is set out.

Version history. Version 1, 25 June 2026: initial publication. Version 2, 10 July 2026: fidelity revision; preserves the structure of Version 1 while adding the reduction-is-not-elimination clarification; naming the real crux in the opening; sharpening indexed objectivity in §4; defining model in the agenthood gate; weakening the maximiser claim to all-else-equal optimization-target language; strengthening the warning that coupling must be specified before it is used; updating the dependency relation with What You Actually Are Version 2; and correcting the Is/Ought Firewall reference so the deflationary ought is attributed to physicalism together with the deflationary horn over error theory, not to physicalism alone.

Search This Blog

Epistemic Forge