Cross-Agent Aggregation Principles

Persistence frontiers, coupled viability, and the comparison of possible continuations

Theoretical and formal paper · persistence, coupled viability, and cross-agent comparison

Andraž Đurič

Version 1 · 22 September 2026

Contents

Abstract

Persistence-relative requirements belong to particular living systems, but their consequences unfold in a shared causal world. Once the continuation of one system affects another, evaluation must represent dependencies, substitution, conflict, and the futures that an intervention leaves available. This paper develops a persistence-frontier principle without assigning organisms context-free points or treating complexity as an intrinsic good.

Two comparison objects are distinguished. A persistence profile records the continuation prospects delivered by a policy. An attainable continuation set records combinations of prospects that some feasible policy can jointly secure. Componentwise dominance orders profiles without selecting trade-off weights. Containment orders attainable sets without counting possible futures. A policy-emulation condition establishes when additional adaptive capacity cannot reduce attainable persistence performance, while matched resources and preserved bearer identities prevent apparent improvements from concealing costs or replacement.

The resulting principles resolve no-loss comparisons, expose avoidable self-undermining, and make relational contributions assessable. They also locate a precise boundary: where profiles or attainable sets cross, local persistence requirements and causal coupling alone need not determine a unique selection. A finite-dimensional result shows why distinct persistence-responsive comparison rules can reverse such a choice. This does not establish that a fuller rule is impossible; it identifies what an additional derivation must determine.

The normative interpretation remains persistence-indexed. Conscious welfare, voluntariness, and the admission of future possible bearers are not silently reduced to continuation probabilities. The contribution is an explicit synthesis and development of the Epistemic Forge framework using established viability and multi-objective concepts, not a claim to have discovered Pareto reasoning or completed moral theory.

1. From one interacting world to a comparison rule

A living system is not insulated by its physical boundary from the conditions that sustain it. Its continuation can depend on processes elsewhere, and its activity can maintain or undermine the continuation of other systems.

One Game of Oughts made those relations explicit while separating the bearer of a requirement from its providers, controllers, and observers. It also preserved genuine conflict: requirements that are individually satisfiable need not be jointly satisfiable. The present paper takes the next step. Once the interacting requirements have been represented, what principles govern their comparison? [F1]

The motivating proposal is:

When continuation requirements conflict, compare the feasible trajectories by the open-ended persistence structure they preserve—not by an assumed intrinsic value assigned to each kind of organism.

That proposal has substantive promise. It also contains a word that must be discharged: richer. A richer persistence structure cannot mean whichever outcome the evaluator already prefers. Nor can it mean more organisms, more connections, greater intelligence, or a longer interval by stipulation.

The argument developed here gives the term one exact use: a structure is richer in attainable continuation when it preserves every jointly achievable continuation target of another structure and permits additional targets. This is a relation of inclusion, not a hidden sum.

That relation will not compare every pair. Where inclusion fails in both directions, a stronger conflict principle must explain which losses can be accepted for which gains. Making that requirement explicit is part of developing the principle, not a reason to abandon it.

2. What this paper inherits

This question was already present in the corpus. Value as Persistence distinguished persistence as the enabling condition of continuing valuation from the stronger reduction of all value-content to persistence. It explicitly left a general comparison of “better persistence” for further work and rejected favorable adjectives as a substitute for a criterion. [F2]

The Apex Synthesis had already located a Pareto frontier within an individual persister when duration, intensity, resources, and robustness trade off. The present extension is therefore not the introduction of frontier reasoning into an otherwise scalar framework. It applies and develops that structure across distinct, interacting bearers. [F3]

Three further commitments constrain the extension. An open horizon must expose deferred consequences rather than conceal them behind a convenient deadline. Coupling must identify actual dependence, including available substitutes, rather than rescue preferred conclusions after the fact. And universal applicability to living bearers does not erase the bearer-index of each requirement. [F4–F6, F11]

The later, explicitly scoped normative formulation governs here. Physicalism alone does not establish a particular semantics of “ought,” and identifying a functional continuation requirement does not by itself establish every practical or moral obligation. Relata-Indexed Objectivity and One Game of Oughts keep those distinctions visible. [F7, F1]

The paper nevertheless makes a positive normative commitment: an identified living organization supplies a non-arbitrary reference for assessing what sustains or defeats its continuation. Calling the corresponding requirements functional oughts is not conditional on the organism wanting to live. It is conditional on the organization whose continuation is under examination. What follows for a controller deciding among different bearers requires the additional comparison developed below.

3. Position in the literature

The formal background is established. Doyen and De Lara formulate stochastic viability using probability constraints and admissible controls. This supplies the distinction between being presently within a continuation-compatible condition and having an available policy that can maintain it. [1]

Roijers and colleagues distinguish solution concepts for multi-objective sequential decisions according to the scalarization and policy assumptions. In particular, their treatment of monotonic preferences, Pareto fronts, and coverage sets explains why several objectives do not automatically yield one optimal policy. This paper uses that established distinction rather than presenting it as a new theorem. [2]

The biological interpretation has both precedents and critics. Di Paolo develops adaptivity as regulation relative to viability, rather than treating continued self-production as already sufficient for every richer account of agency and value. Corti challenges the inference from organizational dependence to normative correctness. These are relevant to the interpretation of the model, not objections that probability notation can settle. [3–4]

Kolchinsky and Wolpert make a further methodological distinction useful here: information can be assessed through its causal contribution to an explicitly selected viability function. That connects information and continuation without identifying the amount of information, the complexity of the system, or the selected viability measure with moral worth. [5]

The residual contribution attempted here is the integration of bearer-preserving comparison, open-horizon profiles, joint attainability, policy emulation, and a stated boundary on conflict selection. The elementary results are supplied so that the synthesis can be inspected without granting authority to its vocabulary.

4. Fix the comparison before ranking outcomes

Let $I$ be a specified set of living bearers. For each $i\in I$ , state what organizational continuity counts as that bearer's remaining the bearer under examination. Its irreversible loss is recorded as failure of that continuity, not erased when another system later performs a similar function.

This inherits a bookkeeping commitment from What You Actually Are: preserving an external service and preserving the particular system that supplied it are different claims. The wider metaphysics of identity is not settled by this paper; an application must defend its continuity criterion. [F8]

Let $\Gamma$ record the comparison's bearer criteria, environment, dynamics, uncertainty model, available information, resource accounting, admissible policies, and continuation targets. An initial condition is fixed when a policy profile is evaluated; it is suppressed in the notation until alternative configurations are compared in Section 8. A policy is an executable rule for action, possibly responding to later observations. Its decisions may use only information available to the controller when those decisions are made. It is not a favorable completed history selected with hindsight.

The admissible class $\Pi_\Gamma$ must respect who can actually act. A coordinated policy unavailable to any real combination of controllers is not an option merely because it can be written down. Acquisition and maintenance of knowledge, tools, support, and coordination belong in the resource model.

For an outcome trajectory, let $\tau_i$ denote the loss-time of the selected continuity. The notation can represent a continuing organism through changes in material or state. It must not silently turn individual continuation into reproduction or lineage continuation.

Let $\mathcal Q=\{q_j:j\in J\}$ be the declared continuation targets. Examples include:

a particular bearer continuing through a specified horizon;
several named bearers continuing together;
at least one member of a specified group continuing.

These targets need not be independent. Joint events are essential precisely because dependence matters. Each target must have a stated role; giving several descriptions of the same event does not create several independent lives or harms.

The basic target family contains positive continuation events: preserving more of the named continuities cannot, by itself, turn success of such an event into failure. This does not assume that more continuation is better along every welfare dimension. It identifies the persistence comparison being made.

The declared probabilistic model must give each target a well-defined event and each admissible policy a probability law on the same described continuations. For the comparisons below, assume that the policy classes are nonempty. A policy need not achieve the continuation targets to be admissible. Define its profile by

p_j^\pi=P_\Gamma^\pi(q_j),\qquad p^\pi=(p_j^\pi)_{j\in J}.

A probability is a model-dependent estimate or a stipulated quantity in a constructed example. Precision in its notation is not precision in its empirical estimation.

5. The open horizon is a family of questions

An open-ended horizon does not require one infinitely distant observation at which every trajectory is scored alive or dead. Nor does it require an infinite sum.

It requires refusing an arbitrary stopping point that conceals relevant continuation possibilities or deferred damage. The appropriate object can be a profile across finite horizons:

S_i^\pi(H)=P_\Gamma^\pi(\tau_i>H).

The horizon family should expand with evidence-supported causal reach. A calculation at a finite horizon can remain useful, but its conclusion is not automatically valid at horizons it did not examine.

This distinction matters even in an elementary constructed case. Under one policy, a bearer continues for exactly ten time units. Under another, it continues for one unit with probability $0.1$ and for one hundred units with probability $0.9$ . At horizon five, the first policy has continuation probability $1$ and the second $0.9$ . At horizon twenty, the probabilities are $0$ and $0.9$ .

Both eventually have continuation probability zero. That common limit does not make their histories equivalent. Nor does considering the full horizon settle the preference: the first better protects the near interval; the second offers longer continuation with greater early risk.

Reality Alignment as a Persistence Strategy already separates horizon, exposure, and weighting. Extending the first does not determine the other two. The Last Gradient likewise retained a genuine trade-off between the manner and duration of a finite continuation. [F4, F9]

One quantifier distinction also matters. Demonstrating that, for each finite horizon, some policy succeeds does not automatically demonstrate one executable policy that succeeds across all horizons. Such an interchange requires appropriate consistency or compactness conditions. A sequence of horizon-specific plans should not be described as one established open-ended strategy without them.

The open horizon disciplines the comparison. It does not supply a unique temporal exchange rate or make remote possibilities infinitely valuable.

6. First principle: no-loss persistence dominance

Define weak persistence dominance by

\pi\succeq_\Gamma\rho \quad\Longleftrightarrow\quad p_j^\pi\ge p_j^\rho\quad\text{for every }j\in J.

Write $\pi\succ_\Gamma\rho$ when, additionally, the inequality is strict for at least one target. Equivalent profiles may be delivered by different policies. Thus the relation is an order on profiles and a preorder on policies that can share a profile.

This produces a first comparison principle:

No-loss principle. Within a declared persistence comparison, an improvement in one represented continuation prospect, with no deterioration in any other, is not a persistence sacrifice. A rule that selects the worse profile must appeal to something other than an improvement in those prospects.

A numerical aggregator is unnecessary to recognize the relation. Define an aggregator $F$ as persistence-monotone when $x\ge y$ entails $F(x)\ge F(y)$ . Every such rule agrees that a dominating profile is no worse. If the rule additionally responds strictly to a no-loss improvement, it strictly prefers the dominating profile.

This is elementary order reasoning. Its significance is the identification of comparisons that do not require a negotiated exchange rate between bearers.

The corresponding frontier is

\mathcal F_\Gamma= \{\pi\in\Pi_\Gamma:\nexists\rho\in\Pi_\Gamma \text{ with }\rho\succ_\Gamma\pi\}.

For a finite nonempty candidate class, at least one nondominated profile exists. An unrestricted infinite class requires further care: a sequence of improvements can approach a limit that no admissible policy attains. A frontier should not be assumed nonempty without suitable conditions.

The normative standing must remain exact. Local functional normativity supplies the bearer-relative continuation requirements. Adopting a joint comparison that responds to no-loss improvements supplies the frontier restriction. This is not a newly discovered command from the universe, and it does not establish that every affected agent must personally prefer what benefits another. It states what follows for the specified cross-agent persistence assessment.

Being on the frontier also does not certify moral permissibility. Several frontier policies can involve serious losses. Nondominance means that the selected comparison has not found an alternative improving something without worsening anything—not that the policy has passed every ethical test.

7. Coupling changes the frontier, not just the description

A dependency belongs in the joint dynamics before the profiles are compared. It is not an optional moral multiplier attached afterward.

The transfer result in One Game of Oughts makes one case precise. If continuation of $i$ without an enabling event $Q$ has probability at most $\varepsilon$ throughout the relevant policy class, then

P^\pi(\tau_i>H)\le P^\pi(Q)+\varepsilon.

Requiring sufficiently reliable continuation therefore requires sufficiently reliable provision of the enabling event. Maintaining something outside a body can be part of sustaining that body. The result does not need a prior assignment of equal value to provider and beneficiary. [F1]

This is where some apparent conflicts disappear. A policy can appear favorable when it records only an immediate reserve gain and unfavorable when the loss of an indispensable maintenance process reaches the same bearer's later profile. If another available policy avoids that loss without sacrificing any represented prospect, dominance settles the comparison.

But coupling is not synonymous with mutual benefit. A dependency can be one-way, replaceable, antagonistic, or conditional on other providers. Nor does a connected network automatically become a further bearer with authority over its members. Preservation Is a Coupling Problem supplies the instrumental mechanism; the bearer gate remains a separate question. [F5]

Suppose either $B$ or $C$ can supply a required service. With both available, removing either one may leave the beneficiary's continuation unchanged; removing both may destroy it. Counting pairwise deletion effects and adding them would miss the shared dependency. The event describing the service and the availability of substitutes must remain in the model.

Conversely, replacing $B$ 's service does not make $B$ 's own termination disappear. It changes one beneficiary's dependency. It does not change the identity of the terminated bearer.

The principle is therefore not to preserve every existing relation. It is to preserve or replace enabling conditions where the specified continuations require them, while retaining the separate record of every bearer affected by the intervention.

8. Second principle: compare attainable continuations

One policy's outcome profile does not exhaust a structure's persistence potential. A system may retain several alternative responses whose relevance depends on what happens next.

To represent that capacity, define an attainable continuation set. For a specified condition $x$ , with the same bearers, targets, and accounting conventions, let

\begin{gathered} \mathcal A_\Gamma(x)=\{r\in[0,1]^J:\\ \exists\pi\in\Pi_\Gamma(x)\;\forall j,\quad p_j^\pi(x)\ge r_j\}. \end{gathered}

An element $r$ is a combination of continuation probability thresholds. Its membership means that one feasible policy can meet all those thresholds together. The set is downward closed: attaining a profile also attains every componentwise weaker threshold.

The definition deliberately does not combine the best result for each bearer from incompatible policies. The quantifier order is

\exists\pi\;\forall j,

not

\forall j\;\exists\pi_j.

That difference is the distinction between a jointly available future and a collection of promises that cannot be fulfilled together.

Now define capability dominance:

x\succeq_{\mathrm{cap}}y \quad\Longleftrightarrow\quad \mathcal A_\Gamma(x)\supseteq\mathcal A_\Gamma(y).

When $x$ and $y$ are genuinely given starting conditions, the comparison concerns the capabilities available from them. To justify choosing an intervention that produces one rather than the other, evaluate the complete acquisition-and-continuation policies from a common pre-intervention baseline. Include acquisition failures, resource use, elapsed time, and every bearer lost on the way. Conditioning only on successful acquisition or on the bearers who remain alive would hide precisely the losses the comparison must retain. Comparing an expensive new capability with an uncharged old one would not meet these conditions.

Strict inclusion supplies one exact meaning of richer persistence structure: every continuation target combination attainable from $y$ remains attainable from $x$ , and at least one additional combination becomes available.

Attainability principle. Preserving a superset of jointly achievable continuation prospects is a capability improvement relative to the declared bearers, targets, and costs. It need not be measured by the number of possible futures, a complexity score, or an interpersonal exchange rate.

Proposition 1. If $\mathcal A_\Gamma(x)\supseteq\mathcal A_\Gamma(y)$ , then, for any persistence-monotone scalar comparison $F:[0,1]^J\to\mathbb R$ ,

\sup_{\pi\in\Pi_\Gamma(x)}F(p^\pi(x)) \ge \sup_{\rho\in\Pi_\Gamma(y)}F(p^\rho(y)).

Proof. Every attained profile belongs to its attainable set. Every threshold in that set is bounded above by some attained profile, so monotonicity makes the supremum over thresholds equal to the supremum over attained profiles. Set inclusion then gives the inequality. No maximizer need be assumed to exist. $\square$

Strict inclusion need not strictly improve every comparison rule: an added option may be irrelevant to a particular rule. What follows is non-deterioration of the best available performance across the whole declared monotone class.

This is stronger than counting options. Two sets can have the same cardinality while one strictly contains the other. Two equally large finite menus can offer wholly different continuations. Inclusion asks what can actually be retained, not how many descriptions of possibilities can be generated.

It also preserves a distinction between capability and conduct. A controller can possess the richer menu and choose badly. More available intelligence is not the same as a better implemented policy.

9. Adaptive capacity and the complexity hypothesis

The preceding definition yields a sufficient condition for capability improvement.

Proposition 2. Policy emulation. Suppose that for every policy available from $y$ , there is an executable policy available from $x$ whose continuation profile is componentwise no worse, under matched bearer criteria, uncertainty assumptions, and resource accounting. Then

\mathcal A_\Gamma(x)\supseteq\mathcal A_\Gamma(y).

Proof. Take any threshold in the set for $y$ . A policy attains it there. Its stipulated emulator from $x$ also attains it. Thus every element of the first set belongs to the second. $\square$

This gives a precise route from certain capacities to persistence potential. Learning, repair, additional sensing, a tool, or an alternative maintenance pathway can enlarge attainable continuation when the previous responses remain available at no offsetting cost. Mere possession of a larger description, more internal components, or a more complicated controller does not establish the premise.

Persistence Is Not Stasis already argued that, under consequential change and finite current responses, adaptive change can be part of continuation rather than a separate value imposed upon it. The present emulation condition sharpens one part of that argument: retained old responses plus usable additional responses can expand what remains achievable. [F6]

The motivating complexity hypothesis should therefore be stated in causal form:

In some environments, additional organized complexity enables learning, correction, and response generation that expand attainable continuation. Its persistence significance comes from those effects, net of their costs—not from complexity as such.

There is no general monotonic law established here. More components can require more resources. More connections can transmit failure. A simple provider can be indispensable, while an elaborate controller contributes nothing under the conditions in question. Whether complexity usually correlates with an expanded frontier across a particular class of organisms is an empirical claim requiring a specified complexity measure and a specified comparison population.

The same restriction prevents an interpersonal error. A person's own continuation remains a represented stake even when that person contributes little to anyone else's capabilities. The presence of dependency effects does not turn usefulness to others into the definition of having a life to lose.

Consider a constructed comparison with two bearers and one marginal continuation target for each. The values below are stipulated probabilities, not empirical measurements.

Policy	Continuation of A	Continuation of B
Wasteful	0.10	0.10
A-favoring	0.90	0.20
B-favoring	0.20	0.90
Shared-capacity	0.70	0.70

Within this four-policy class, Wasteful is dominated. The other three profiles are nondominated. None is best for every persistence-responsive comparison.

Suppose the old arrangement offers only A-favoring and B-favoring. An improved arrangement retains both and makes Shared-capacity available without changing costs or the outcome meanings. The attainable set strictly expands: the threshold pair $(0.70,0.70)$ was not previously attainable and now is.

Allowing lotteries between the original policies does not remove that finding. Every mixture of the two has coordinate sum $1.10$ , whereas the added profile sums to $1.40$ . No such mixture reaches both thresholds. Lotteries may fill portions of the frontier; their availability must be stated rather than assumed.

Choosing from the expanded menu remains a separate question. Equal positive weights favor Shared-capacity:

\begin{gathered} 0.5(0.70)+0.5(0.70)=0.70\\ >0.5(0.90)+0.5(0.20)=0.55. \end{gathered}

Weights $(0.90,0.10)$ instead favor A-favoring:

\begin{gathered} 0.90(0.90)+0.10(0.20)\\ =0.83>0.70. \end{gathered}

Both comparisons respond positively to both bearers. Neither ignores persistence. Their disagreement identifies a trade-off not settled by the displayed physical prospects.

This example deliberately uses only marginal targets. A broader claim must include relevant joint events. To see why, compare two other constructed policies: one gives probability $0.5$ to both continuing and $0.5$ to neither; another gives probability $0.5$ to A alone and $0.5$ to B alone.

Their marginal profiles are identical. Their joint profiles are not. The first offers a greater chance that both continue; the second guarantees that at least one does. There is no basis for calling one simply more persistent without stating which aspects of that joint structure the comparison retains.

11. Why a frontier need not select one outcome

A precise result locates the boundary of weight-free profile comparison.

Proposition 3. Unanimity over positive linear comparisons. For finite-dimensional real profiles $x,y\in\mathbb R^d$ ,

\begin{aligned} x\ge y\quad\Longleftrightarrow\quad &w\cdot x\ge w\cdot y\\ &\text{for every }w\in(0,\infty)^d. \end{aligned}

Proof. If every component of $x-y$ is nonnegative, every positive weighted sum is nonnegative. Conversely, suppose component $k$ is negative. Give it weight $1$ and give every other component the same sufficiently small positive weight. The negative component then exceeds the sum of positive contributions in magnitude, making the weighted difference negative. $\square$

For crossing profiles, apply the same construction to a coordinate favoring each side. Two strictly positive weighting rules rank the alternatives oppositely. Normalizing either weight vector to sum to one does not change the ranking.

Positive weighted sums are not asserted to be the correct moral theory. Their role is diagnostic: they exhibit more than one comparison consistent with positive responsiveness to all represented persistence prospects. Therefore those conditions alone cannot logically entail a unique ranking of every crossing pair.

That is stronger than saying that a calculation has not yet been performed. It is a conditional underdetermination result. It is also narrower than saying that a fuller persistence-based rule cannot be derived. Additional physical structure may remove the crossing; a better account of the bearers may change the relevant targets; or a further normative argument may justify a priority. What cannot resolve the disagreement is repeating “richer” while leaving the selecting relation unspecified.

An attainable-set comparison has the same boundary. If each structure admits threshold combinations the other cannot, neither contains the other. A larger volume, a greater count of hypothetical futures, or a preference for one region of the set introduces an additional measure. That measure may be defensible. Set inclusion itself has not supplied it.

A frontier contains maximal candidates: each is undominated by the other candidates. It need not contain a maximum that weakly dominates every candidate. The distinction matters as much across an open horizon as across a short one.

12. The human, the parasite, and the nested component

The motivating human–parasite contrast contains several possible comparisons, and they should not be collapsed.

From the human bearer's persistence index, an intervention that removes a destructive dependency and improves that person's continuation can be straightforwardly favored. An actual application would still require evidence about the intervention, its risks, alternatives, and downstream effects. Nothing here supplies clinical advice.

From a wider network index, maintaining that person can also preserve capabilities and relationships relevant to others. This is the proper location for the proposed significance of knowledge, care, repair, and adaptive potential. The relevant question is what those capacities actually leave jointly achievable in the environment, not how sophisticated the species label sounds.

But when the parasite's own continuation is included as a distinct target, terminating it is a loss on that coordinate. A larger contribution elsewhere does not make that loss mathematically vanish. Preference for the intervention may remain strongly justifiable, but it is then a conflict judgment—not a no-loss dominance result over every bearer.

Nestedness changes the causal model without abolishing this distinction. Where a component's expansion destroys an indispensable condition of its own continuation, the open-horizon model exposes the strategy's self-undermining structure. That mechanism is already part of the persistence account; it is not a fresh objection to it. Where a wider organization has a legitimate continuity criterion, its maintenance supplies a real index for assessing component behavior.

However, being inside a system does not, by itself, prove that the wider boundary is the only boundary that matters in every normative comparison. The same question must be faced when moving among cells, organisms, and collectives: which bearer is being preserved, and which bearer is bearing the loss?

The proposed conflict principle remains promising in its stronger form:

Where continuation requirements are genuinely incompatible, prefer the feasible trajectory that preserves the better-supported open-ended persistence structure.

The present paper supplies exact interpretations of “better-supported” where profiles dominate or attainable sets include one another. In crossing cases, the stronger principle remains a research commitment: it must derive the comparison that makes one structure preferable without concealing a bearer, a cost, or a priority inside the word “structure.”

13. Longevity, replacement, and conscious welfare

Consider a person who wants to continue living in good health. Suppose an available continuation preserves that person's prospects without worsening any other represented persistence prospect, while a forced ending removes those prospects. On the declared persistence comparison, the ending is dominated.

Nothing in that result requires the person to be unusually intelligent, socially productive, or indispensable to others. Their own continuation supplies the improved coordinate. Positive external effects can strengthen the practical case; they are not a prerequisite for the loss to exist.

A conventional lifespan does not change that comparison by itself. Nor does one person's preference for finitude establish a requirement to impose finitude on another. Such considerations would have to supply a relevant further stake or argument, rather than function as an unexplained deadline.

The no-loss assumption is substantive. Resource competition, risk, and effects on others must be examined rather than wished away. The principle does not decide every allocation problem merely by using the word longevity.

Replacement requires equal care. Substituting another provider can preserve a function. It cannot, under a token-indexed continuity criterion, retroactively preserve the terminated provider. The record should distinguish a maintained function, a continuing individual, and the beginning of a different individual. The addition of future possible bearers raises a further population question: which possible lives enter the comparison and on what terms? The fixed-bearer model does not answer it by multiplying speculative descendants.

Finally, a persistence profile is not a completed welfare profile. Persistence Is Not Stasis explicitly retains the question whether conscious valence has significance not exhausted by persistence. The Fidelity Program likewise treats conscious consequences as ethically load-bearing. [F6, F10]

The pressure test is direct. Suppose two trajectories preserve the same bearers, durations, relationships, adaptive capacities, and attainable continuations, but differ in severe conscious suffering. If that separation is physically possible, the persistence variables as specified do not represent the difference. A fuller account must either add the relevant welfare dimension or demonstrate—not assume—that the purported equality in persistence is impossible.

No conclusion that prolonged suffering is justified follows from locating a policy on a persistence frontier. Neither does a command to force every person to remain alive against their considered wishes. Functional continuation, experienced welfare, consent, and practical obligation must be distinguished before their relations can be derived.

14. Evidence, uncertainty, and representation discipline

No application possesses a complete causal map of an open future. The framework needs a way to remain useful without pretending otherwise.

For a declared set of plausible models $\mathcal M$ , a demanding robust comparison requires the profile inequalities to hold separately in every model:

\begin{gathered} p_j^{\pi,M}\ge p_j^{\rho,M}\\ \text{for every }j\text{ and every }M\in\mathcal M. \end{gathered}

A robust attainable set can similarly require one policy that meets the same thresholds across the model class:

\begin{gathered} \mathcal A^{\mathrm{rob}}_\Gamma(x)=\{r\in[0,1]^J:\\ \exists\pi\in\Pi_\Gamma(x)\;\forall M\in\mathcal M\;\forall j,\\ p_j^{\pi,M}(x)\ge r_j\}. \end{gathered}

This is not permission to place every imaginable catastrophe into the model class. The class needs evidence-based justification. Nor does it license choosing a different policy after being told which unknown model is true, unless the controller can actually obtain that information. Uniform guarantees summarize a different question from a probability-weighted forecast, and may be very conservative.

Several representational checks are indispensable. Relabeling bearers or duplicating the same recorded target must not manufacture an improvement. Adding a genuinely distinct living bearer is different from duplicating a description; it changes the subject set and requires an explicit extension. Changing a continuity criterion or excluding a harmed bearer after inspecting the ranking is not a correction unless the change has independent justification.

Causal contributions must also be kept separate from extra points. If a provider's effects on several organisms already appear in their continuation probabilities, adding a second bonus for its “network importance” can count the same effect twice. Topology belongs first in the transition model. A separate objective concerning it requires a separate argument.

The practical output should distinguish three findings: a comparison settled under the stated model; a comparison that reverses across plausible models; and a comparison whose trade-off remains after the models agree. More data may address the second. It does not necessarily settle the third.

15. A constructive route to stronger conflict principles

The next step should be a derivation project, not a search for one impressive scalar.

A bearer-relative completion starts from a particular living system's actual continuation requirements and derives which policies meet them. Coupling can make substantial preservation of others necessary within that analysis. The result remains indexed to that bearer and must not be presented as every other bearer's verdict.

A joint-constraint completion starts from justified requirements for several bearers: for example, specified reliability floors, limits on imposed harm, or conditions of an actual cooperative arrangement. The model then tests whether those requirements are jointly feasible and which policies satisfy them. The justification of the floors or limits is part of the theory, not a number supplied by the optimization routine.

A stronger comparative completion tries to show that a particular ordering of crossing persistence structures is constitutive of an adequately specified form of agency or joint normative assessment. To succeed, it must explain why competing persistence-responsive orderings fail, rather than merely choose one. It must also retain conscious stakes and distinguish the continued existence of particular individuals from substitutable functions.

Even the symmetric limiting case illustrates the separation. When exactly one of two otherwise symmetrically situated bearers can continue, local requirements identify two losses that cannot both be avoided. If a single lottery over the two feasible outcomes must remain invariant under exchanging the bearer labels, each outcome must receive probability one half. This symmetry requirement explains the equal chances; merely allowing lotteries would not select that distribution. Bare persistence alone has not supplied an entire theory of fair risk allocation. A justified procedural result can still be a substantive completion for that case.

The strongest prospective contribution would be a rule whose scope and failure conditions are clear enough that a critic can construct the case on which it gives the wrong answer. Such a rule could expand what the frontier resolves. It should be tested against the frontier's unresolved cases, not protected by renaming them.

16. What could defeat this proposal?

The set-inclusion and finite-order results are deductive given their definitions. Their principal vulnerabilities in application are misidentified bearers, unrealizable policies, omitted costs, incompatible targets, and inaccurate dynamics.

The proposed usefulness of the synthesis is separately testable. A model should record bearers and targets before ranking interventions, represent compensation and substitution explicitly, and distinguish the policy actually implemented from the best feasible policy. Predictions about continuation and dependency should then be compared with observed outcomes at the relevant scale. This paper contains constructed examples, not an empirical validation of those predictions.

The complexity hypothesis would be weakened in a specified domain if measured additional complexity failed to produce the claimed response capacities after matching costs, or if those capacities failed to expand attainable continuation under the relevant disturbances.

The stronger conflict proposal would fail at its claimed scope if it could select only by changing the bearer set, hiding a suffered loss, or introducing weights it purported to have derived without an additional premise. A countermodel exhibiting two permitted orderings with different selections is evidence against uniqueness under those premises—not against every possible further argument.

Finally, the persistence-only reduction would need revision if an independently established welfare difference remained after all defensible persistence-relevant differences had been held fixed. The possibility must remain exposed rather than being ruled out by defining every good into persistence.

17. Conclusion

Persistence gives evaluation a living reference. Coupling determines how that reference reaches beyond a body's boundary. An open horizon exposes dependencies and losses concealed by convenient stopping points. Together they constrain comparison much more strongly than an isolated count of lives or an immediate survival score.

The resulting machinery has two objects. A profile describes what a policy delivers. An attainable set describes what a structure leaves jointly achievable. Dominance can settle the first comparison without trade-off weights; inclusion can settle the second without counting futures. Adaptive capacity matters where it expands that achievable structure, not because complexity has acquired an intrinsic moral rank.

This yields a defensible persistence-frontier principle:

Compare feasible policies by the continuation prospects they preserve for explicitly identified bearers. Reject a claimed persistence advantage that depends on hidden losses, unavailable responses, or a convenient horizon. Where all existing prospects can be retained and additional ones secured, the persistence case favors retaining that capability. Where continuation structures cross, expose and justify the principle selecting among their losses and gains.

The remaining conflict is not outside reality. It is a conflict among real, located continuation requirements. But putting those requirements in one causal world does not make their differences disappear.

The task is therefore neither to assign every organism the same number nor to declare that the more complex organism always wins. It is to determine what each feasible intervention sustains, what it forecloses, and what further rule—if one is required—justifies the exchange.

Preserve the structure that makes further continuation possible. Name whose continuation is preserved. Keep the losses visible.

Standing and version record

Version 1 · 22 September 2026. Publication edition, developed from the Epistemic Forge corpus and the dialogue initiating this paper.

The formal comparison results are elementary consequences of specified orders, sets, and policy classes. Their application is model-dependent. The persistence-indexed normative interpretation is an inherited philosophical commitment. The empirical relationship between complexity and attainable continuation remains a domain-specific hypothesis. A complete cross-agent conflict selector and the reduction of conscious welfare are not established here.

The contribution claimed is synthesis, specification, and development. No priority claim is made for viability theory, Pareto frontiers, monotone comparison, policy emulation as a sufficient condition, or the elementary proofs. The literature pass is targeted rather than an exhaustive originality audit.

The source-and-version review, editorial changes, and reproducible checks are recorded in the accompanying publication audit. Those checks do not constitute peer review or empirical validation.

References and internal lineage

External primary sources

[1]
Doyen, L., and De Lara, M. (2010). Stochastic viability and dynamic programming. arXiv:1002.1140v1. Archive record; open manuscript. Especially Sections 2.2–2.3 and 3. The archive records submission in February 2010; the retrieved PDF separately carries an internal November 2018 date.
[2]
Roijers, D. M., Vamplew, P., Whiteson, S., and Dazeley, R. (2013). A survey of multi-objective sequential decision-making. Journal of Artificial Intelligence Research, 48, 67–113. doi:10.1613/jair.3987; open manuscript. Especially Section 4.2.2, pp. 79–82. The journal article is dated 2013; its arXiv deposit is dated 2014.
[3]
Di Paolo, E. A. (2005). Autopoiesis, adaptivity, teleology, agency. Phenomenology and the Cognitive Sciences, 4, 429–452. doi:10.1007/s11097-005-9002-y; author-hosted full text. Especially pp. 437–440. The article belongs to the 2005 volume; the publisher records online publication in April 2006.
[4]
Corti, L. (2023). Organizational normativity and teleology: A critique. Synthese, 202, Article 96. doi:10.1007/s11229-023-04255-7; open article. Especially Section 3 and the conclusion.
[5]
Kolchinsky, A., and Wolpert, D. H. (2018). Semantic information, autonomous agency and non-equilibrium statistical physics. Interface Focus, 8(6), 20180041. doi:10.1098/rsfs.2018.0041; open manuscript, arXiv:1806.08053v3. Especially Sections IV–V. The consulted archive version is dated 7 November 2018.

Epistemic Forge sources

[F1]
Đurič, A. (2026). One Game of Oughts. Version 1, 22 September. Published text.
[F2]
Đurič, A. (2026). Value as Persistence: Agent-relative oughts under coupling, nesting, uncertainty, and open-ended time. Version 2, 10 July. Published text.
[F3]
Đurič, A. (2026). Epistemic Forge: The Apex Synthesis. Version 1, June; post dated 21 June. Saved original consulted. Canonical post.
[F4]
Đurič, A. (2026). Reality Alignment as a Persistence Strategy. Version 3, 12 August. Saved version consulted. Canonical post.
[F5]
Đurič, A. (2026). Preservation Is a Coupling Problem. Version 1, June; post dated 15 June. Saved original consulted. Canonical post.
[F6]
Đurič, A. (2026). Persistence Is Not Stasis. Version 1, July; post dated 14 July. Saved original consulted. Canonical post.
[F7]
Đurič, A. (2026). Relata-Indexed Objectivity. Version 1, July. Published text.
[F8]
Đurič, A. (2026). What You Actually Are. Version 4, July. Relevant sections consulted. Published text.
[F9]
Đurič, A. (2026). The Last Gradient. Version 1.0, June; post dated 11 June. Saved original consulted; a current canonical link was not independently established.
[F10]
Đurič, A. (2026). The Fidelity Program: Toward a High-Resolution Map of the Territory. Version 2, June; post dated 21 June. Saved version consulted. Canonical post.
[F11]
Đurič, A. (2026). Universal Is Not Categorical. Version 1, 28 August. Published text. The scope distinction in Section 2 also draws on F1 and F7.

Search This Blog

Epistemic Forge