The Natural History of Fidelity
The Natural History of Fidelity
How finite physical agents become map-makers, why every map loses something, and why correction becomes part of continuing to exist
Apex synthesis · physicalism, thermodynamics, agency, representation, uncertainty, correction, and value
Version 1 · July 2026
Abstract
This essay presents an architecture of finite embedded agency.
The accessible universe contains lawful structure and temporary free-energy gradients. Under some conditions, physical processes use those gradients to maintain organized boundaries against dissolution. Some of these systems become adaptive regulators and agents: they act selectively in relation to conditions that sustain or threaten them.
Because an agent is finite and located inside the reality it must navigate, it cannot carry that reality in full. It operates through selective internal structures and, at higher levels, explicit models. Representation is therefore compressive. Compression omits structure; omission creates uncertainty and possible divergence between map and territory.
The agent must act anyway.
Its actions meet consequences. Through those consequences, relevant divergence may become detectable. Where a pathway exists by which detection can alter the model or the activity guided by it, the agent possesses a correction architecture. Corrigibility is therefore not merely an intellectual virtue. For finite systems acting through incomplete models, it is part of the operating structure of continued agency.
The account is physicalist but non-eliminative. Thermodynamics supplies the functional floor for bounded, resource-consuming agents, not a complete explanation of semantics, consciousness, or value. Higher-level patterns remain real where they identify stable organization and name their relata. Objectivity survives indexing: a claim can be relative to a frame, agent, task, scale, or horizon while remaining objectively constrained once those relations are specified.
Value enters locally. The persistence of a valuer is an enabling condition for its continuing valuation, but the stronger claim that all value reduces in content to persistence remains open. Conscious experience introduces moral stakes that functional organization alone may not exhaust.
The result is not a completed theory of everything. It is a map of how cosmology, thermodynamics, life, agency, representation, science, meaning, and value may fit into one corrigible architecture.
1. The object
There is a story broad enough to contain cosmology, life, cognition, language, science, engineering, value, and the possible end of the universe, but narrow enough to be carried by one recurring structure.
A finite physical process maintains itself inside an environment it cannot fully contain. To continue, it must remain sensitive to at least some of the structure around it. In living systems, sensitivity becomes regulation: internal organization changes action in relation to conditions that sustain or threaten the system.
As regulation becomes memory, anticipation, counterfactual evaluation, and thought, the system increasingly acts through models.
Those models are necessarily incomplete.
The agent is smaller than the territory, limited in energy, limited in time, limited in observational access, and located inside the system it is trying to understand. It must select which distinctions to preserve. Selection produces compression. Compression omits structure. Omission creates uncertainty and the possibility that the map will diverge from the territory.
The agent must nevertheless act.
Its actions meet consequences. Through those consequences, the territory can answer back. If divergence becomes detectable and the agent can alter its model or behaviour before the error crosses an irreversible boundary, correction becomes part of its continued existence.
This is the central architecture:
The account is physical throughout, but it is not flat. Explaining the material realization of a phenomenon does not automatically eliminate the phenomenon at its own level. A person is physically realized without being usefully replaced by a particle inventory. Meaning is physical without being identical to heat. Institutions, languages, values, promises, memories, and works of art are higher-level patterns whose reality depends on organization and relation, not on escaping physics.
The task is therefore double.
The upper levels must not be erased merely because lower levels realize them.
But neither may the upper levels float free of the systems, relations, and practices in which they perform their work.
The architecture protects the castle from being collapsed into its bricks and requires the castle to have bricks, structure, and a place in the world.
2. Physicalism and cosmological conditions
The standing ontological position is physicalism.
Whatever agents, experiences, meanings, institutions, values, and representations are, they are physically instantiated or physically realized. This is not held as a logically certain conclusion. It is held as the strongest inductive standing position available.
Again and again, phenomena once placed outside nature have become physically intelligible. Heat became a phenomenon of statistical mechanics. Life became a question of chemistry, organization, inheritance, selection, and regulation. Perception, memory, emotion, and decision increasingly became questions for neuroscience and cognitive science.
No well-confirmed explanatory program has yet required a nonphysical substance whose interaction with physical processes can be independently specified.
Physicalism remains corrigible in principle. It is a standing commitment of inquiry, not a metaphysical exemption from inquiry.
Cosmology supplies the contingent conditions under which the rest of the architecture becomes possible.
Our best current models describe an accessible universe that began in an extraordinarily low-entropy condition and remains far from equilibrium. From that history came usable gradients: differences in temperature, density, chemical potential, gravitational organization, and available free energy.
Gradients make work possible.
Stars, planets, atmospheric systems, chemistry, organisms, nervous systems, computers, and civilizations all operate inside the continuing discharge of such differences. The energy passing through an organism is not incidental to its existence. It is part of the physical process by which the organism remains organized rather than dissolving into its surroundings.
This does not mean the universe intended to produce life or intelligence. It does not mean increasing complexity is the purpose of cosmic history. A direction in physical evolution is not yet a normative instruction.
The universe gives agents conditions, constraints, opportunities, and eventual limits. It does not command them to persist, flourish, discover, create, or care.
Those stakes appear only where there are systems for whom conditions can go better or worse.
3. Thermodynamics as the functional floor
Thermodynamics is not the deepest level of reality. Beneath it lie the microphysical structures and dynamics investigated by fundamental physics.
But thermodynamics is the functional floor for the phenomena examined here.
It is the lowest explanatory scale at which the following concepts become jointly usable:
- gradients;
- work;
- dissipation;
- irreversibility;
- boundary maintenance;
- persistence and dissolution;
- resource limitation;
- information-processing cost;
- and failure through time.
The central claim is therefore deliberately bounded:
A localized process can remain far from equilibrium by drawing usable energy from its environment and exporting entropy into its surroundings. Maintaining an organized boundary requires work. Repair, regulation, memory, prediction, and computation occur in material systems with finite power and finite storage.
Information is physically instantiated. This does not mean every informational distinction can be read directly from thermodynamics. It means that storing, transforming, transmitting, and erasing physically realized information cannot be separated from material processes.
Landauer's principle supplies a lower bound on the heat dissipated by logically irreversible erasure. It does not imply that every act of cognition operates near that bound, nor that semantic error can be converted directly into a fixed quantity of heat. Results in information thermodynamics, including work on the thermodynamics of prediction, establish precise relations only under specified models and conditions. Their extension into a universal measure of map–territory divergence would require further argument.
The thermodynamic floor therefore constrains the architecture without settling its upper levels.
Thermodynamics alone does not determine:
- what a sentence means;
- whether a state is representational;
- where one person ends and another begins;
- what conscious pain feels like;
- what an agent ultimately values;
- or which moral verdict should govern a conflict.
Those questions concern forms of organization, use, relation, experience, and interpretation that become available only at higher levels.
Thermodynamics supplies the floor.
It does not demolish the floors above it.
4. From dissipation to agency
Some physical processes merely occur. Others maintain the organization that allows them to continue occurring.
A flame, a vortex, a tornado, a bacterium, an animal, and a person are all dynamic structures. They are not therefore the same kind of system.
The crucial transition is not simply from static matter to motion, or from equilibrium to dissipation. It is from dissipation to self-maintaining organization, and from self-maintaining organization to adaptive regulation and agency.
The literature on autopoiesis describes living systems as networks of processes that continually produce the components and boundaries that make that network possible. Organizational approaches develop this through the idea of closure: constraints depend on and maintain one another as parts of the organization of the system. Enactive approaches add adaptivity: a living system does not merely remain organized but regulates its activity relative to conditions of viability.
These traditions support a graded bridge rather than one sudden transition:
Each transition adds structure.
A dissipative process persists through continuing throughput but need not regulate conditions on its own behalf.
A self-maintaining organization contributes causally to the continued production and repair of the organization that constitutes it.
An adaptive regulator changes its activity relative to conditions that improve or threaten its viability.
An agential controller acts asymmetrically on its coupling with the environment, selecting among possible trajectories relative to system-specific norms.
A content-bearing or model-bearing agent uses internal states in ways that can succeed or fail relative to what those states indicate or represent.
An explicit reflective mapper constructs detachable, recombinable, counterfactual models and can represent itself as a model-building system.
This ladder avoids two opposite errors.
The first is to deny all continuity and reserve agency for human reflective thought.
The second is to treat every self-organizing process as already possessing a model, beliefs, or purposes in the same sense.
Minimal agency need not require a symbolic inner representation. Basic regulation may occur through embodied dynamics whose relevant structure is distributed across body and environment. Radical enactivist accounts are therefore right to warn against inflating every adaptive coupling into representation.
At the same time, control theory and cybernetics support the idea that successful regulation requires internal or embodied structure that tracks regularities relevant to the system being controlled. The good-regulator tradition and internal-model principle formalize versions of this intuition under specified conditions. But "model" must remain indexed to the regulatory task. A control structure that functions as a model for one analysis may not be an explicit representation available to the system itself.
The current agenthood criterion can therefore be stated in graded form.
At minimum, an agent must possess:
- an organization that distinguishes it from an environment;
- conditions under which that organization can continue or fail;
- capacities that regulate its coupling with the environment relative to those conditions;
- and enough asymmetry or control for its activity to count as something it does, rather than merely something that happens through it.
Stronger forms of agency add memory, learning, anticipation, representation, counterfactual modeling, deliberation, language, and self-modeling.
The exact thresholds remain open. The transition from life to agency is not yet one solved theorem. It is a research program connecting thermodynamics, organizational biology, control theory, cognitive science, and philosophy of mind.
5. A taxonomy of boundaries
Boundaries recur throughout the architecture, but "boundary" does not name one uniform object.
A cell membrane, an organizational closure, a Markov blanket, a control interface, a model boundary, and a national border are not the same kind of thing. They may align, but they need not.
At least six boundary types must be distinguished.
Material boundaries
A material boundary is a physical interface: a membrane, wall, surface, or region across which matter and energy flow.
It is often measurable and spatially identifiable. But material enclosure alone does not establish agency or individuality. A bottle has a boundary without maintaining itself.
Organizational boundaries
An organizational boundary is produced or maintained by the network of processes that constitute the system.
The membrane of a living cell is not merely a container imposed from outside. It is produced, repaired, and regulated by the system whose activity it helps constitute. The boundary participates in the organization it separates.
Statistical boundaries
A statistical boundary partitions variables according to conditional dependencies. A Markov blanket, in its formal sense, separates internal from external states through a set of mediating states.
This can be a useful tool for analyzing dynamical systems. But the existence of a statistical partition does not by itself establish biological individuality, agency, consciousness, or intrinsic purpose. Statistical blankets and organismic boundaries may coincide, but the coincidence must be shown rather than assumed.
An instrumental and an ontological reading of the same formalism come apart at exactly this point. Read instrumentally, a Markov blanket is a modeling choice: a partition of variables that makes certain inferences tractable, chosen by whoever is doing the modeling, answerable only to whether it works for that purpose. Read ontologically, the blanket is treated as though it had already picked out a boundary that was there in the system independent of the choice to model it that way. The ontological reading is the one doing any real work — it is what would ground a claim about where an organism ends and its environment begins, or which coupled system counts as one agent rather than two — and it is exactly the reading a statistical partition cannot support by itself. A candidate partition can always be proposed. Whether it satisfies the relevant statistical relations, and whether those relations track an organizational boundary rather than merely a useful abstraction, are further questions the statistics alone don't settle.
Control boundaries
A control boundary identifies the interfaces through which a system senses, acts, regulates, or is perturbed.
These may not coincide exactly with the material boundary. A tool, prosthesis, communication network, or environmental scaffold may participate in a control loop without becoming part of the organism in every other sense.
Explanatory boundaries
An explanatory boundary is drawn for a model or inquiry. It determines which variables are treated as internal, external, fixed, negligible, or out of scope.
Such a boundary may be task-relative without being arbitrary. A useful explanation must show why the selected partition preserves the structure relevant to the question.
Institutional and semantic boundaries
Institutions draw legal, administrative, and conventional boundaries. Concepts and operators possess domains within which they have determinate use.
A promise exists within practices of commitment. Citizenship exists within legal institutions. "Before" presupposes temporal ordering. "Velocity" presupposes a frame. An ought, on the account developed here, presupposes a bearer, capacities, conditions, and a standard relative to which action can succeed or fail.
The general principle is therefore not that every real object possesses one perfectly sharp natural edge.
It is:
A boundary is also not the absence of interaction.
It is often the interface through which interaction occurs.
Changing the boundary may change the object of evaluation. A cell, organism, family, institution, state, biosphere, and universe are not interchangeable optimization units.
Many apparent factual and moral disputes are partly disputes about which boundary is being treated as the relevant bearer.
6. Levels without elimination
Reality is organized across levels: physical, chemical, biological, cognitive, social, institutional, semantic, and normative.
Explaining a phenomenon at one level does not automatically abolish it at another.
Understanding the chemistry of a brick does not make the castle unreal. Understanding the neural and hormonal realization of love does not make love disappear. Understanding that language is physically instantiated does not eliminate syntax, implication, reference, or public meaning.
Higher-level patterns can be real even when fully physically realized.
This is not a license for every familiar word. A higher-level pattern earns its place by identifying stable organization, supporting explanation, enabling prediction, or preserving distinctions that would be lost at lower resolution.
The anti-eliminative principle is:
This is not merely a plea for tolerance toward higher-level talk. Within the effective-information framework used to define causal emergence, a coarse-grained description can carry strictly more effective causal information than the fine-grained description it is built from — not just be more convenient for limited creatures, but be more determined relative to that measure, because aggregation can eliminate degenerate micro-level pathways that obscure which state actually did the causal work. Where this holds, the charge that a higher-level account is "merely" descriptive — a convenience for creatures too limited to track the microphysics — has the direction backwards: the microphysics can be the harder place to read the causal structure off of. This result is scale-relative, tied to a specific formal measure and intervention scheme rather than a framework-neutral fact, and it does not hold for every system; showing that it holds in a given case is separate work from asserting it by analogy elsewhere. But it means the anti-eliminative principle above does not have to rest on tolerance alone. It has, at least sometimes, a formal ally.
But the protection runs in both directions.
A high-level term must still name something. "Justice," "meaning," "authority," "identity," "value," and "ought" cannot survive merely because their grammar is familiar or socially powerful. They must identify the agents, systems, relations, practices, or structures in which they perform their work.
The two commitments pull against each other deliberately:
The upper floors remain real.
They must pay rent.
7. From correlation to representation
Finite agents regulate through selective internal and embodied structure. But not every selective structure is already a representation.
A graded ladder helps prevent both eliminativism and inflation.
Covariance
One state varies with another.
A thermometer's fluid level covaries with temperature. Covariance is physically real, but correlation alone does not yet determine content. Many variables correlate with many others.
Control relevance
A state participates in regulation.
Its variation affects how a system acts. This gives the state a functional role, but not necessarily representational content in the full sense.
Indication
A state reliably indicates some condition relevant to the system.
A receptor may indicate the presence of a chemical. An alarm may indicate smoke. Indication adds an asymmetric informational relation, but reliable indication can still occur without explicit interpretation.
Functional representation
A state is used by some consumer or control organization in a way that can succeed or fail relative to what the state is supposed to track.
This introduces the possibility of misrepresentation. A state can be produced normally yet guide the system incorrectly because the indicated condition is absent or because the signal has become unreliable.
Teleosemantic theories emphasize the selected or maintained functions of signal-producing and signal-consuming systems. Informational semantics emphasizes objective information flow. Enactive accounts emphasize that meaning emerges through the activity and viability of the system rather than through detached inner symbols.
These approaches identify different necessary or contributing conditions. No single account has eliminated all indeterminacy.
Explicit model
An explicit model is a structured representation that can be detached from the immediate stimulus, recombined, compared, and used counterfactually.
Maps, diagrams, equations, narratives, simulations, and scientific theories belong here. They can represent absent, possible, impossible, past, or future states.
Public symbolic representation
Language and other public symbolic systems add socially maintained inferential roles, conventions, commitments, and practices of correction.
A scientific term does not receive its content from causal contact alone. Its use is constrained by measurement procedures, theories, inferential consequences, instruments, institutional practices, and relations to the world.
This ladder does not imply that every stage is sharply separated or that all theories of representation can be unified without remainder.
Its purpose is more modest:
8. Relational meaning
Not every meaningful act is primarily representational.
A descriptive sentence maps a claimed state of affairs.
A cry expresses a state.
A command directs action.
A promise creates a commitment within a social practice.
A legal declaration may alter an institutional relation.
A work of art may represent, evoke, simulate, reorganize attention, create an experience, or open a counterfactual space.
This five-way split — description, expression, direction, commitment, and institutional declaration — is not an invention of this essay. It names a structure with its own history in the philosophy of language, developed from earlier work on what an utterance does as opposed to what it says, and refined into the specific five-way form used here. That history is not repeated here; it is enough to note that the split has a pedigree and is not being proposed fresh.
The broader semantic principle is therefore:
Meaning does not float free of relations.
An expression is anchored through some combination of:
- signs and signal structures;
- producers and interpreters;
- causal histories;
- conventions and practices;
- inferential roles;
- contexts;
- referents or territories;
- functions;
- and possible consequences.
This does not reduce meaning to simple reference.
Inferential, expressive, directive, and constitutive functions are also real. A command can be meaningful without describing a fact. A promise can alter what agents are entitled or committed to do. A fictional world can be meaningful without purporting to describe an actual place.
But meaningfulness still requires physically and socially instantiated relations that distinguish correct from incorrect uptake, relevant from irrelevant inference, successful from failed use, or one function from another.
When a term purports to represent, its content must remain anchored to what it represents and to the practices that determine how the representation succeeds or fails.
When language expresses, directs, commits, or coordinates, those functions and their relata must be named rather than mistaken for ordinary descriptive reference.
The first question is therefore not always:
It is:
9. Compression and loss
The image running through this section and the next two — a map that is not its territory — is old, and this essay does not claim to have coined it or to be giving its first careful treatment. What follows is this architecture's particular development of a long-standing image, not a claim on the image itself.
Every explicit representation selects.
A word preserves some distinctions and removes others. A category groups unlike cases according to a chosen structure. A measurement converts a physical interaction into a value. A diagram suppresses most details. An equation isolates regularities. A scientific model preserves selected causal or predictive relations while idealizing away much of the individual case.
Representation is selective and commonly compressive. Even a lossless representation of a bounded target excludes the wider context from which that target was selected.
But compression is not itself the error.
No finite map can preserve every feature for every possible use. The relevant question is which structure is lost and whether that structure was load-bearing for the task.
Information theory gives formal tools for analyzing compression, coding, prediction, and trade-offs between rate and distortion. Minimum-description-length approaches formalize the preference for compact models that explain data well. But no purely formal compression ratio determines semantic or scientific adequacy by itself.
A shorter description may preserve the wrong regularity.
A highly predictive model may rely on spurious correlations.
A biologically useful signal may not support explanation.
A morally convenient category may erase the individuals harmed by its simplification.
Compression must therefore be evaluated relative to a declared loss profile.
A loss profile states which errors matter, how much they matter, and for what use.
A subway map can distort geographic distance while faithfully preserving station order and connectivity. It succeeds because those distortions are not load-bearing for route selection.
The same map would fail as a guide to property boundaries or walking distance.
Purpose selects the relevant territory and loss function.
It does not make every representation equally good.
10. The limit of absolute representation
A representation can be exact relative to a bounded domain.
A finite string can be copied losslessly. A database can preserve every selected field. A formal structure may possess an isomorphic representation. A model may perfectly preserve one declared relation at one resolution.
But a globally exhaustive representation of the entire territory is a different demand.
To represent everything without loss, the map would need to preserve:
- every entity;
- every state;
- every relation;
- every scale;
- every dynamic;
- every counterfactual;
- every future development;
- and every consequence of the map's own construction.
If the map exists inside the territory, the completed territory includes the map itself. The map would therefore have to include its own physical state, its effects, and its representation of itself representing.
At that limit, nothing may be omitted.
The map ceases to compress and approaches the informational and causal complexity of another region of the territory.
This is not a denial of exact local representation. It is a claim about a finite embedded agent constructing a globally exhaustive, fully self-inclusive representation of the reality in which the construction itself occurs.
Formal results concerning self-reference, truth, computability, and algorithmic incompressibility illuminate parts of this limit, but none should be inflated into a simple proof of the whole metaphysical conclusion. The broader argument also depends on physical finitude, embeddedness, reflexive inclusion, and the practical requirement that a representation remain usable by the agent carrying it.
Incomplete maps are not merely a temporary defect of human civilization.
They are the ordinary condition of finite embedded representation.
11. The epistemic interior
Finitude forces selection.
Selection produces compression.
Compression produces omission.
Omission creates uncertainty and possible divergence.
The agent must act through the incomplete map anyway.
Agency therefore occurs inside an epistemic interior between two unusable absolutes.
At one end is perfect substantive certainty. A finite embedded agent cannot generally obtain it. Evidence is local, instrument-mediated, physically costly, temporally limited, and exposed to unknown unknowns.
At the other end is perfect uncertainty. But perfect uncertainty would destroy action selection. No policy could be preferred, no expectation differentiated, and no evidence interpreted. Even representing a possibility space already imposes distinctions.
Agency lives between them:
The agent may be uncertain not only about the state of the world but about the completeness of its possibility space, the reliability of its evidence, and the accuracy of its confidence estimates.
No final probability distribution automatically resolves this problem. A probability model covers the possibilities represented within it; it need not capture what the model does not know how to represent.
The answer is not final certainty.
It is corrigibility under uncertainty.
Three kinds of epistemic standing should therefore be distinguished.
Logical or analytic certainty
Tautologies and valid formal consequences are secure because of their structure.
"If P, then P."
"What exists, exists."
"What persists, persists."
They do not discriminate among substantive possible worlds.
Hinge commitments
Some commitments are constitutive of inquiry rather than ordinary empirical conclusions.
Reality constrains successful inquiry.
Contradictions cannot simply be accepted indiscriminately without collapsing discrimination.
Evidence can sometimes count against a claim.
These commitments are not ordinary observations, but abandoning them would abandon the practice of reality-directed inquiry.
Empirical confidence
Substantive claims about the world are held with degrees of confidence determined by evidence, model quality, alternatives, calibration, and correction history.
Confusing these forms allows empirical claims to borrow the certainty of tautology or allows the revisability of empirical knowledge to be mistaken for complete skepticism.
12. Tautologies and the formal limit
Tautologies occupy a peculiar place in the architecture.
They are maximally secure because they exclude no substantive empirical possibility.
This does not make them useless.
They constrain coherent inference. They expose category mistakes. They identify enabling conditions. They mark points where some requests for explanation may have reached the boundary of their domain.
The persistence work contains a central example:
This secures an enabling condition for continuing valuation.
It does not establish that everything valued is, in content, reducible to persistence.
The tautology gives one claim almost for free. It does not pay for the stronger theory.
The same discipline applies throughout the architecture.
"What exists, exists" does not tell us what exists.
"What persists, persists" does not explain how a system persists.
"A map differs from its territory" does not tell us which map is adequate.
"An agent acts under constraints" does not identify its values.
The most secure statements are often the least discriminating.
Substantive knowledge begins where the map becomes vulnerable to being wrong.
13. Frame-specified objectivity
Objectivity does not require the absence of an index.
Many objective facts are relational.
Velocity is relative to a frame.
Fitness is relative to an organism and environment.
Risk is relative to an exposure, probability model, and horizon.
"Better" is relative to a task and criterion.
An ought may be relative to an agent, its capacities, environment, commitments, and conditions of persistence or flourishing.
A fact does not become arbitrary merely because one of its relata is an agent.
The central principle is:
This position has close relatives in scientific perspectivism, model pluralism, and work on idealization. Different models may preserve different structures because scientific inquiry serves different tasks and operates at different scales.
But three dimensions must remain distinct.
Truth
Is the claim correct about the specified territory under the stated interpretation?
Adequacy
Does the model preserve the structure required for its task within the declared error tolerances?
Usefulness
Does the model enable prediction, intervention, understanding, communication, or action?
These can come apart.
A model may be useful but rely on false assumptions.
A statement may be locally true but misleadingly incomplete.
A high-fidelity model may be too complex for its intended user.
A simple model may be adequate within a narrow regime and fail catastrophically outside it.
Purpose selects which structures matter for a task. It does not manufacture truth.
The stronger formulation is therefore:
Indexed does not mean arbitrary.
Relative does not mean unreal.
Purpose-dependent does not mean unconstrained.
14. The map-construction discipline
The architecture yields an operational method.
Each step answers a different question.
Determine the function
Is the object describing, predicting, explaining, classifying, expressing, directing, committing, coordinating, simulating, or constructing?
A command should not be assessed as though it were a failed scientific report. A model should not be assessed as though every simplifying assumption were a lie. A poem and a measurement instrument do not owe the same kind of fidelity.
Fix the task
What must the representation or act enable?
Understanding, prediction, intervention, communication, coordination, decision, memory, experience, or design?
Name the relata
What entities, properties, agents, systems, environments, practices, and relations are involved?
A relational term without its relata may appear more absolute than it is.
This step is not new here. Demanding that a term's use be traced to the practice, context, or inferential role that gives it content, rather than treated as free-floating, has a substantial history in twentieth-century philosophy of language, under more than one name. What this architecture adds, if anything, is narrower than the diagnostic itself: applying it specifically to persistence-indexed normative and boundary claims, not proposing the diagnostic for the first time.
Draw the boundary
What is included? What is excluded? Which kind of boundary is being used? At what scale is the individuation stable?
Specify the scale, index, conditions, and horizon
Relative to which frame, agent, environment, temporal interval, measurement procedure, or standard is the claim made?
Declare the loss function
Which errors matter for the task? How are false positives, false negatives, omissions, delays, and distortions weighted?
Preserve the load-bearing structure
Which distinctions must survive for the representation to remain adequate?
Compress openly
Choose the word, category, equation, diagram, narrative, model, or theory.
The compression should be as simple as the task permits and no simpler than the load-bearing structure allows.
Expose the losses
What has been omitted? Under what nearby task would the omission become misleading? Which populations, scales, causal pathways, or alternatives disappear in the compression?
Specify the correction architecture
Through what pathway can relevant divergence become detectable and alter the model or the action guided by it?
One risk runs through every step above and is not discharged by following them. The choice of loss function can itself be motivated. A loss profile that happens to mark exactly the inconvenient structure as "not load-bearing for this task" produces a map that looks disciplined — every step followed, every loss disclosed — while doing the same work as an undisclosed one. Nothing in the ten steps distinguishes a loss profile chosen because it is genuinely adequate to the task from one chosen because it is comfortable. The check has to come from outside the mapmaker's unilateral control, and it comes in three grades, not one. A declared loss is one where the omission is made visible — the minimum the ten steps above secure, and no more. A justified loss is one where the omission is defensible relative to the actual task, not merely disclosed. An adversarially tested loss is one where the criterion has been run against cases and stakeholders the mapmaker did not select, and has survived. Disclosure secures only the first grade. Moving from declared to justified, and from justified to adversarially tested, is the work of checking whether the profile survives a task the mapmaker did not pick and stays fixed across cases where the inconvenient structure changes but the mapmaker's interest in avoiding it does not. A loss profile that tracks the mapmaker's comfort rather than the task is the same failure this corpus has separately named at the level of language — see "Semantic Terrorism" (internal, existing on this blog) — appearing here one level up, in what gets to count as noise before the reporting even starts.
This is not a recipe for a perfect map.
It is a discipline for building representations whose limits remain visible and whose contact with reality remains structurally possible.
15. Correction architectures
"Remain open to correction" is too weak if it names only an attitude.
Correction requires architecture.
A system is corrigible only where relevant divergence can enter through some channel, become discriminable from noise, and change the model or activity.
A correction architecture must identify at least:
- the target variable or structure;
- the evidence or sensor channel;
- the comparison process;
- the loss or error criterion;
- the update rule;
- the delay between divergence and detection;
- the noise and uncertainty profile;
- the intervention pathway;
- and the possibility of suppression, manipulation, or adversarial distortion.
Different domains possess different correction channels.
An empirical scientific claim may be corrected by measurement, replication, prediction failure, or experimental intervention.
A mathematical proof may be corrected by identifying an invalid inference or counterexample.
A conceptual definition may be corrected when it fails to discriminate cases, creates contradiction, or no longer serves the intended theoretical role.
An institution may be corrected through audits, complaints, elections, markets, legal review, whistleblowing, or direct material failure.
A person's self-model may be corrected through introspection, social feedback, behavioural consequence, clinical evidence, or conflict among their own predictions.
The existence of feedback is not enough.
Feedback may be delayed, noisy, selectively sampled, gamed, censored, or interpreted through a model that protects itself from revision.
Correction systems can also become unstable. An agent may overreact to noise, update too quickly, or allow adversarial signals to hijack its control loop.
Cybernetics, recursive estimation, adaptive control, robust control, experimental design, and fallibilist epistemology all contribute to the same general insight:
Correction is not the opposite of structure.
It is structure designed to remain answerable.
16. Science, engineering, and philosophy
Science is not merely measurement.
It observes, instruments, hypothesizes, idealizes, compares, predicts, simulates, intervenes, detects anomalies, and discovers entities or mechanisms that previous maps did not contain.
Science is a discovery machine and a correction system.
It does not merely assign numbers to concepts philosophy has already supplied. It repeatedly forces new distinctions into the conceptual vocabulary because reality contains structures the old vocabulary could not represent.
Science feeds forward into engineering:
Engineering operationalizes models. It also creates instruments, experiments, interfaces, and environments that make new scientific discovery possible.
Science feeds backward into philosophy:
Philosophy operates throughout the loop.
It frames questions, distinguishes claim types, audits premises, identifies hidden relata, tests concepts, and prevents local success at one scale from being inflated into universal metaphysics.
The relation is recursive:
No single discipline holds the entire loop.
The quality of the collective map depends partly on the interfaces among them.
17. Local normativity
Normativity begins where states and actions can go better or worse relative to the organization of some system.
A self-maintaining organism has viability conditions. Some trajectories sustain its organization; others damage or dissolve it. An adaptive agent regulates its activity in relation to those differences.
This supports a naturalized form of local normativity.
The bacterium ought to move toward a usable nutrient gradient in the limited functional sense that doing so contributes to the conditions under which it continues, while failing may lead to its dissolution.
This does not mean the bacterium possesses a reflective moral obligation. It means that the system's organization establishes a distinction between successful and failed regulation.
The semantic thesis remains substantive:
Physicalism alone does not prove this analysis of the word. Nor does thermodynamics by itself establish the content of every ought. The claim depends on a theory of the practical function and domain of normative language.
This claim meets Hume's law immediately. No ought is validly derivable from is-premises by logic alone; any apparent derivation smuggles a normative premise in somewhere. That inferential point is granted in full — nothing in this architecture claims a formally valid derivation of a categorical ought from purely descriptive premises, and none is offered here. What is not granted is the further gloss often attached to it: that the impossibility of that derivation shows normative language answers to an autonomous, nonphysical domain, disconnected from what agents are and what sustains them. The inferential point and the stronger claim that normativity must belong to such a domain are different claims; the first does not by itself establish the second, and this essay does not need to take a position in the separate scholarly dispute over how closely that stronger claim tracks Hume's own considered view. Given the naturalism already on the table in §2, and given a bearer with specified persistence conditions, "agent A ought to φ" can be reconstructed as a claim that φ-ing is required or favored relative to A's specified organization, conditions, environment, and horizon: A is the kind of thing that persists partly by way of φ-ing, in environment E, over horizon H. Offering that reconstruction is not deriving an ought from an is in the sense the law forbids. It is a semantic proposal about what the word is doing once its bearer and standard are fixed, not a violation of the inferential point above.
Two existing positions bear on this and differ from it in a specific way. One ties an agent's reasons entirely to that agent's actual motivational set, or to what a sound deliberative route from that set would produce; on that view, an agent with no actual or reachable desire connected to its own persistence has no reason to preserve it, full stop. This architecture indexes differently — to what actually sustains the agent, not to what the agent happens to want or could come to want by its own lights. An addiction can sit inside an agent's actual motivational set and still work against what sustains it; on the indexing used here, the ought tracks the second, not the first, and the two can diverge. A second position holds that the standards a would-be agent must meet are constitutive of agency itself — that failing to meet them does not make one a defective agent so much as not fully an agent at all, which is what makes the antecedent non-optional rather than merely inconvenient to escape. That move is closer to what this architecture needs than the hypothetical-imperative form alone provides, where an agent who simply doesn't care about the consequent is convicted of nothing worse than not caring. Whether the persistence-conditions bearer used here earns the same non-optional status that constitutive standards are supposed to have, or only borrows its shape, is not settled by anything argued so far.
A 2026 formalization of persistence-conditioned dynamics across physics, biology, economics, and cultural evolution reaches a structurally similar bridge from selection dynamics rather than from agent-relative persistence conditions directly: the descriptive claim that configurations generating friction are selected against, a conditional normative claim that holds only if agents prefer the outcome selection tends to produce, and an explicit refusal to assert that the outcome is good independent of that preference. The route there is different — through population-level selection rather than a single agent's persistence conditions — but the shape of the bridge is close enough that the two should be read against each other. That work also states, with an explicit proof sketch rather than a closed proof, what this architecture only gestures at in §9 through §11: proposed necessary and sufficient conditions for when a coarse-grained description of a multi-scale system preserves the original dynamics, and when it instead acquires memory effects the coarse description cannot capture on its own. The paper is explicit that this specific result is a proof sketch requiring further formalization — its separate, genuinely machine-checked results (in Lean 4) cover a different set of claims, about the replicator dynamics themselves rather than about coarse-graining. Even at the proof-sketch stage, this corpus's "declare the loss function" has no comparable formal backing yet. That is a gap worth stating rather than papering over.
Once the bearer and standard are specified, ought-facts may be objective without being absolute.
"The cell ought to climb the gradient" is not identical to "the cell will climb." The cell may fail. The ought expresses the action required or favored relative to the cell's continued viability, not a prediction that the action will occur.
The same structure becomes more complex in reflective agents.
Humans possess many partially independent needs, values, identities, commitments, relationships, and horizons. They are embedded in families, institutions, ecosystems, and cultures. Their action-guiding models must integrate across biological, psychological, social, and conscious levels.
There may be objective facts about what sustains a person, relationship, institution, or ecosystem under a declared horizon.
There need not be one frame-independent optimizer capable of ranking every conflict among all agents from nowhere.
18. Persistence and value
The persistence account must preserve a crucial distinction.
The secure claim is the enabling condition:
A dissolved agent cannot continue pursuing its projects or experiencing its values. This is close to analytic.
The stronger claim is the content-reduction:
That remains a bet.
The first claim does not entail the second.
Persistence may be necessary for valuing while failing to exhaust the content of what is valued.
Art, play, sacrifice, love, aliveness, flourishing, conscious richness, and care for agents whose persistence is not instrumentally connected to one's own remain pressure tests.
Some instances may support persistence. Play can develop capacity. Art can coordinate, teach, regulate, or preserve identity. Love can create durable cooperative systems.
But explaining some functions of a value does not establish that the value's entire content reduces to those functions.
The reduction cannot be protected by redescribing every apparent counterexample until it agrees.
The current position is therefore intentionally asymmetric.
Persistence is secured as an enabling condition and powerful organizing constraint.
Its status as the complete content of all value remains open against rival accounts based on flourishing, desire, conscious valence, relationship, achievement, or irreducible plurality.
19. Consciousness and moral weight
Agency and moral weight are not identical.
A minimal self-maintaining system may qualify as an agent without being conscious. It may possess viability-relative control relations without there being anything it is like to be that system.
Conscious experience introduces another dimension:
- suffering;
- pleasure;
- flourishing;
- deprivation;
- fear;
- joy;
- and felt stakes.
The architecture therefore distinguishes two gates.
Agency is relevant to bearing an ought in the functional, action-guiding sense.
Consciousness is relevant to moral weight.
The exact placement of both gates remains open, especially at the margins of animal life and artificial systems.
Functional accounts of valence may explain why a complex controller requires priority signals, global interrupts, resource reallocation, and learning mechanisms. Pain can be modeled partly as a high-priority signal of injury, threat, or escalating divergence.
But a functional explanation of what pain does does not automatically explain why there is something it is like to feel it.
The phenomenological residue remains a genuine frontier.
Physicalism commits the inquiry to a physical realization of experience. It does not mean the realization has already been identified or that functional vocabulary has exhausted the explanandum.
20. Experience, play, art, and aliveness
First-person experience supplies data that a physical account must eventually explain, but it must not erase the datum before the explanation exists.
A useful distinction is:
A person's attraction to aliveness, appetite for experience, love of play, grief, aesthetic response, or sense of meaning exists as a real physical and phenomenological event before a complete reduction has been produced.
Analysis can locate it, compare it, identify functions, and test proposed groundings. It cannot honestly pretend that the value was deduced from thermodynamics merely because a physical realization must exist.
Art belongs here as well.
Art often compresses structure, but it does not merely report facts. It can:
- represent observed reality;
- express a state;
- direct attention;
- simulate another world;
- create an experience;
- preserve a memory;
- reorganize perception;
- or open a counterfactual possibility.
Its fidelity therefore depends on function.
A portrait, propaganda poster, engineering drawing, game world, autobiographical image, abstract composition, and fictional narrative do not owe the same kind of correspondence.
The appropriate generalization is:
A fictional work may be false as literal history while faithfully preserving psychological, social, or counterfactual structure.
An uncomfortable work may succeed precisely by producing discomfort appropriate to what it expresses.
A beautiful work may be aesthetically effective while compressing a false or harmful ideology.
No single measure of compression determines artistic value.
21. Coupling and nested agents
Agents do not exist as isolated beads.
Their persistence conditions reach into bodies, families, infrastructures, institutions, ecosystems, cultures, and planetary systems.
This produces a nested and coupled web.
A person depends on organs, food systems, social knowledge, technology, law, and biosphere. An institution depends on people, procedures, legitimacy, material infrastructure, and information channels. Humanity depends on planetary systems and the continuing flow of usable energy.
Coupling comes in degrees and forms.
It may be:
- direct or mediated;
- symmetric or asymmetric;
- stable or temporary;
- visible or hidden;
- replaceable or non-substitutable.
An agent has instrumental reason to preserve external structure to the extent that the structure genuinely contributes to its own persistence or projects and cannot be replaced at lower cost.
But intelligence alone does not create care.
A sufficiently accurate agent may discover that another system is irrelevant to its own objective or that the relevant function can be substituted.
Clarity is not benevolence.
Where overlap exists, intelligence may identify a cooperative gradient.
Where overlap does not exist, intelligence does not manufacture one.
This is why capability and ends remain distinct. An intelligent system can become better at pursuing almost any sufficiently coherent objective.
Coupling must therefore be measured rather than invoked after the fact to rescue a preferred moral conclusion.
22. Cosmological self-location
Eventually the map expands far enough to include the physical process supporting the mapper.
The agent models:
- its dependence on gradients;
- the life cycle of its star;
- the accessibility of energy;
- the physical cost of computation;
- cosmic expansion;
- and possible end states of the universe.
Cosmological heat death remains a conditional, model-dependent extrapolation rather than an absolute metaphysical certainty. It depends on assumptions about the large-scale dynamics of the universe, including the behaviour of dark energy and the continuation of the relevant physical laws across immense horizons.
Different cosmological branches imply different futures and different limits.
The epistemic achievement is not choosing the emotionally preferable ending.
It is preserving the assumptions, confidence, alternatives, and correction channels.
Cosmology is the architecture operating at its widest available horizon.
A local physical process uses instruments, mathematics, observations, and collective institutions to construct a map of the conditions that produced it and the possible exhaustion of the gradients that sustain it.
Even a hard cosmic end would not create a cosmic ought or erase present value.
The universe supplies constraints.
Agents supply stakes.
Finitude does not negate local value, because local value was never secretly dependent on eternal preservation.
A song can be real and valuable while it plays even though it ends and its energy dissipates.
23. Reflexive closure
This framework is itself a physical model, running in physical systems, of physical agents that construct models.
It is a token of the kind of process it describes.
That is not evidence that it is true. A false but internally coherent theory can also represent itself.
The loop is structural, not justificatory.
But it reveals the final shape of the object:
The universe does not literally wake up as one unified subject.
Rather, local physical processes become capable of representing increasingly large regions of the reality of which they are parts.
The map eventually includes the mapper.
The correction discipline must then apply recursively.
The framework must expose its own boundaries, compression losses, assumptions, and correction channels.
Otherwise its reflexivity becomes self-mythology rather than self-knowledge.
One further risk sits inside the discipline this framework recommends for itself, and it is distinct from the ones already named. Naming an open joint is not the same as having earned the confidence claimed elsewhere. A stated bet can come to function as a shield rather than a disclosure, if its main effect is to pre-empt objections rather than to remain genuinely available for resolution. The test is not whether a claim carries an open-joint label. It is whether engaging with that joint is capable of changing anything — whether a specific argument against it would move the architecture, or whether the label has become a permanent parking spot for the objection instead. Section 24's list is offered as the first kind. That is a claim the list makes about itself, and a list cannot certify its own honesty; it is tested only by whether a future correction to one of its own items actually occurs.
24. What holds and what remains open
This architecture is not a single proof. Its components do not share one epistemic status.
The strongest standing claims are:
- physicalism is the best-supported inductive ontology presently available;
- non-equilibrium thermodynamics supplies a functional floor for bounded, resource-consuming agents;
- finite agents cannot model every aspect of their territory at every resolution;
- explicit representation is selective and compressive;
- higher-level patterns can remain real without being nonphysical;
- objectivity can survive frame and agent indexing;
- boundaries, scales, relata, and loss functions must be declared;
- empirical agency operates under uncertainty;
- correction requires functioning pathways rather than mere openness;
- continuing valuation requires a continuing valuer;
- and no demonstrated cosmic agent bears a universal ought.
The central open joints are:
- the full transition from dissipative process to living organization;
- the transition from adaptivity to agency;
- the threshold between regulation, indication, and representation;
- the exact relationship among material, organizational, statistical, and explanatory boundaries;
- the strongest naturalistic theory of semantic content;
- the generalization of information-thermodynamic results into a broad theory of map cost;
- the content-reduction of every value to persistence;
- the relation among persistence, flourishing, conscious valence, and moral weight;
- the measurement of coupling;
- the physical basis of phenomenal experience;
- and a formal, rather than merely declarative, treatment of when compression preserves the structure a task needs — the nearest existing model of this last point sits outside this corpus and is named in §17.
These open joints are not blemishes to conceal.
They are where the territory can still correct the architecture.
25. The architecture, compressed
Reality contains lawful structure and temporary gradients.
Under some conditions, physical processes use those gradients to maintain organized boundaries. Some become adaptive regulators and agents, acting selectively relative to conditions that sustain or threaten them.
Because agents are finite and embedded, they cannot contain the territory in full. They operate through selective internal structure and, at higher levels, explicit models.
Representation compresses.
Compression creates omission.
Omission creates uncertainty and possible divergence.
Agents act through their maps anyway.
Consequences allow relevant divergence to return through evidence, prediction failure, resistance, injury, success, or breakdown. Correction architectures allow models — and sometimes the systems carrying them — to change before divergence crosses an irreversible boundary.
Science industrializes discovery and correction.
Engineering turns models into altered territory and creates new access for inquiry.
Philosophy audits the functions, relata, boundaries, indices, assumptions, levels, and losses throughout.
Values arise locally in agents and conscious systems, not as instructions issued by the cosmos.
As their models expand, agents eventually represent their own physical nature, their uncertainty, their values, and the possible cosmological exhaustion of the gradients that made modeling possible.
Beneath the entire architecture sit three limits:
Inside that interval, fidelity is not perfection.
It is the disciplined attempt to preserve what matters, disclose what was lost, and remain answerable to reality.
Standing of this essay
This is an architectural synthesis and research program, not a completed scientific theory or one continuous deductive proof.
Its contribution is primarily integrative: placing cosmological conditions, thermodynamic constraint, organizational boundaries, graded agency, selective representation, relational meaning, frame-specified objectivity, loss auditing, correction architecture, and local value inside one recursive account of finite embedded agents.
Many individual components have substantial prior literatures. The claim is not that autopoiesis, cybernetics, perspectival realism, teleosemantics, fallibilism, information theory, speech-act theory, causal emergence, or naturalized normativity begin here. This essay makes explicit several places where the architecture independently converged on an existing named result, rather than leaving those overlaps implicit.
The proposed contribution lies in the architecture connecting them and in the operational discipline derived from that connection:
The framework should be judged by the same standard it recommends.
Its boundaries must remain visible.
Its compression losses must be named.
Its rivals must be represented fairly.
Its errors must remain capable of changing it.
References
- Anderson, P. W. (1972). "More Is Different." Science, 177(4047), 393–396. Standard
- Austin, J. L. (1962). How to Do Things with Words. Oxford University Press. — §8's five-way split in representation, expression, direction, commitment, and declaration descends from Austin's original distinction between locutionary, illocutionary, and perlocutionary acts, refined into the specific taxonomy below. Verified · July 2026
- Ashby, W. R. (1956). An Introduction to Cybernetics. Chapman & Hall. Standard
- Barandiaran, X. E., Di Paolo, E., & Rohde, M. (2009). "Defining Agency: Individuality, Normativity, Asymmetry, and Spatio-Temporality in Action." Adaptive Behavior, 17(5), 367–386. Verified · July 2026
- Box, G. E. P. (1976). "Science and Statistics." Journal of the American Statistical Association, 71(356), 791–799. Standard
- Brandom, R. B. (1994). Making It Explicit. Harvard University Press. — inferentialism is one of the existing traditions §14's "name the relata" step draws on without improving upon; content as inferential role is a different route to a similar demand than the one taken here. Standard
- Bruineberg, J., Dołęga, K., Dewhurst, J., & Baltieri, M. (2021). "The Emperor's New Markov Blankets." Behavioral and Brain Sciences, 45, e183. — §5's instrumental/ontological distinction for statistical boundaries is this paper's target-article distinction, applied here rather than independently rediscovered without credit. Verified · July 2026
- Cartwright, N. (1983). How the Laws of Physics Lie. Oxford University Press. Standard
- Conant, R. C., & Ashby, W. R. (1970). "Every Good Regulator of a System Must Be a Model of That System." International Journal of Systems Science, 1(2), 89–97. Verified · July 2026
- Dennett, D. C. (1991). "Real Patterns." The Journal of Philosophy, 88(1), 27–51. Standard
- Di Paolo, E. A. (2005). "Autopoiesis, Adaptivity, Teleology, Agency." Phenomenology and the Cognitive Sciences, 4, 429–452. Standard
- Dretske, F. I. (1981). Knowledge and the Flow of Information. MIT Press. Standard
- Fodor, J. A. (1974). "Special Sciences." Synthese, 28(2), 97–115. Standard
- Foot, P. (1972). "Morality as a System of Hypothetical Imperatives." The Philosophical Review, 81(3), 305–316. Standard
- Foot, P. (2001). Natural Goodness. Oxford University Press. Standard
- Farzulla, M. (2026). "The Replicator-Optimization Mechanism: A Scale-Relative Formalism for Persistence-Conditioned Dynamics with Application to Consent-Based Metaethics." arXiv:2601.06363 (preprint, not peer-reviewed). — the near neighbor named in §17: an independent route to a structurally similar is-ought bridge, together with an explicit proposed theorem giving lumpability conditions for when coarse-grained dynamics preserve their original structure. The paper labels this specific result a proof sketch requiring full formalization; its machine-checked (Lean 4) verification covers a different, separately identified set of results — simplex preservation, survival monotonicity, moving equilibrium existence, and the impossibility of static equilibrium under varying friction — not the lumpability theorem. This corpus does not yet have a comparable formal treatment even at the proof-sketch stage. Verified · July 2026 (read in full)
- Friston, K. (2010). "The Free-Energy Principle: A Unified Brain Theory?" Nature Reviews Neuroscience, 11, 127–138. Standard
- Giere, R. N. (2006). Scientific Perspectivism. University of Chicago Press. Standard
- Hoel, E. P. (2017). "When the Map Is Better Than the Territory." Entropy, 19(5), 188. — the formal ally named in §6: causal emergence gives a scale-relative, non-metaphorical sense in which a coarse-grained description can be more informative about a system's causal structure than the fine-grained one beneath it. Verified · July 2026
- Hoel, E. P., Albantakis, L., & Tononi, G. (2013). "Quantifying Causal Emergence Shows That Macro Can Beat Micro." Proceedings of the National Academy of Sciences, 110(49), 19790–19795. — the earlier result the 2017 paper above builds on; not independently re-verified this pass. Standard
- Hutto, D. D., & Myin, E. (2013). Radicalizing Enactivism: Basic Minds Without Content. MIT Press. Standard
- Kalman, R. E. (1960). "A New Approach to Linear Filtering and Prediction Problems." Journal of Basic Engineering, 82(1), 35–45. Standard
- Knight, F. H. (1921). Risk, Uncertainty and Profit. Houghton Mifflin. Standard
- Korsgaard, C. M. (2009). Self-Constitution: Agency, Identity, and Integrity. Oxford University Press. — the constitutivist position named in §17: agency's constitutive standards as the source of a non-optional antecedent, differing from the hypothetical-imperative form in the way stated there. Verified · July 2026
- Korzybski, A. (1933). Science and Sanity: An Introduction to Non-Aristotelian Systems and General Semantics. International Non-Aristotelian Library. — origin of the map/territory image used throughout §9 through §11, acknowledged in §9 rather than left as an unattributed inheritance. Standard
- Landauer, R. (1961). "Irreversibility and Heat Generation in the Computing Process." IBM Journal of Research and Development, 5(3), 183–191. Standard
- Levins, R. (1966). "The Strategy of Model Building in Population Biology." American Scientist, 54(4), 421–431. Standard
- Mackie, J. L. (1977). Ethics: Inventing Right and Wrong. Penguin. Standard
- Massimi, M. (2022). Perspectival Realism. Oxford University Press. Verified · July 2026
- Maturana, H. R., & Varela, F. J. (1980). Autopoiesis and Cognition: The Realization of the Living. Reidel. Standard
- Millikan, R. G. (1984). Language, Thought, and Other Biological Categories. MIT Press. Standard
- Montévil, M., & Mossio, M. (2015). "Biological Organisation as Closure of Constraints." Journal of Theoretical Biology, 372, 179–191. Verified · July 2026
- Moreno, A., & Mossio, M. (2015). Biological Autonomy: A Philosophical and Theoretical Enquiry. Springer. Standard
- Morgan, M. S., & Morrison, M., eds. (1999). Models as Mediators: Perspectives on Natural and Social Science. Cambridge University Press. Standard
- Popper, K. R. (1963). Conjectures and Refutations. Routledge. Standard
- Prigogine, I. (1978). "Time, Structure, and Fluctuations." Science, 201(4358), 777–785. — origin of the dissipative-structure image the §4 ladder is built on. Standard
- Rissanen, J. (1978). "Modeling by Shortest Data Description." Automatica, 14(5), 465–471. Standard
- Rosen, R. (1991). Life Itself: A Comprehensive Inquiry into the Nature, Origin, and Fabrication of Life. Columbia University Press. Standard
- Schrödinger, E. (1944). What Is Life? Cambridge University Press. Standard
- Searle, J. R. (1975). "A Taxonomy of Illocutionary Acts." In K. Gunderson (Ed.), Language, Mind, and Knowledge, Minnesota Studies in the Philosophy of Science, Vol. 7. University of Minnesota Press. — the five-way taxonomy (assertive, expressive, directive, commissive, declarative) that §8 restates without naming. Verified · July 2026
- Simon, H. A. (1955). "A Behavioral Model of Rational Choice." The Quarterly Journal of Economics, 69(1), 99–118. Standard
- Still, S., Sivak, D. A., Bell, A. J., & Crooks, G. E. (2012). "Thermodynamics of Prediction." Physical Review Letters, 109, 120604. Verified · July 2026
- Tarski, A. (1933/1956). "The Concept of Truth in Formalized Languages." In Logic, Semantics, Metamathematics. Oxford University Press. Standard
- Williams, B. (1979). "Internal and External Reasons." In R. Harrison (Ed.), Rational Action. Cambridge University Press, pp. 17–28. Reprinted in Williams, Moral Luck: Philosophical Papers 1973–1980 (Cambridge University Press, 1981), pp. 101–113. — the internalist-about-reasons position named in §17, differing from this architecture's indexing in the way stated there. Verified · July 2026
- Wittgenstein, L. (1921). Tractatus Logico-Philosophicus. Standard
- Wittgenstein, L. (1953). Philosophical Investigations. Blackwell. — language-games and use-in-a-form-of-life, the other existing tradition §14's "name the relata" step draws on without improving upon; distinct from the two 1921/1969 Wittgenstein entries above, which do not cover this material. Standard
- Wittgenstein, L. (1969). On Certainty. Blackwell. Standard
Verification key: "Verified · July 2026" means the citation and the specific claim attributed to it were checked against primary or authoritative secondary sources during this revision pass. "Standard" means the work is treated as sufficiently canonical that it was not independently re-checked this pass; none of these are niche or contested attributions. No entry in this list was left unclassified.
Internal lineage
This synthesis develops and reorganizes arguments from the Epistemic Forge corpus, including:
- Epistemic Forge: The Apex Synthesis
- The Fidelity Program
- The Spine and the Web
- Build Maps That Reality Can Correct
- High-Fidelity Maps
- Name the Relata
- The Exclusivity Error
- The Wizard's Error Is Delocalization
- The Is/Ought Firewall
- Where the Is–Ought Question Bottoms Out
- Value as Persistence
- Preservation Is a Coupling Problem
- What You Actually Are
- The Last Gradient
- The Two Ways to Be Wrong
- Fidelity-Valenced Cognition
- An Appetite for Experience
- The Magnetism of Aliveness
- Semantic Terrorism
- and the earlier Thermodynamic Realism documents from which the present functional-floor position was refined.
The older Thermodynamic Realism label is retired. Thermodynamics remains central as the functional floor for bounded agents, but not as the absolute deepest ontology or as a complete reduction of semantics, consciousness, and value.
Comments
Post a Comment