artifacts/standard-named

Rendered from markdown source. Open raw source on GitHub.

--- title: "Capacity, Complexity, and Requisite Variety" subtitle: "Cybernetics, Bounded Rationality, Cognitive Load, Distributed Cognition, Common Ground, and Long Context" artifact_date: "2026-07-14" artifact_type: "adjacent-fields-and-capacity-review" domain: "TELIC-FIELDS" scope: "WORKING" lineage: "THE-TELIC-FIELD-PAPERS" status: "draft" processing_tier: 4 source_role: "research-and-claim-boundary-artifact" content_canon_status: "unset" publication_status: "unpublished" series_position: "G.6" companion_to:

• "20260714__TELIC-FIELDS__PAPER__CANDIDATE__F-5__context-carrying-capacity.md"

research_note: > This is an interdisciplinary comparison, not a systematic meta-analysis. Context carrying capacity remains a candidate synthesis. The review identifies substantial prior foundations and specifies what the term must add to remain useful. ---

Capacity, Complexity, and Requisite Variety

Executive finding

Context carrying capacity sits at the intersection of several mature traditions.

Cybernetics already states that effective regulation requires sufficient variety to respond to disturbances. Bounded rationality explains decision under finite information, time, and computation. Working-memory and cognitive-load research show that active human processing is sharply limited and task dependent. Information-bottleneck theory formalizes compression that preserves information relevant to a target. Distributed cognition shows that cognitive capacity may be organized across people, artifacts, environments, and time. Common-ground theory explains why communication requires coordinated evidence of mutual understanding and why different media impose different grounding costs. Organizational information-processing theory treats uncertainty and interdependence as demands that organizations address by increasing processing capacity or reducing processing requirements. Long-context language-model research demonstrates that nominal input length does not guarantee robust use of all included information.

Context carrying capacity is therefore not novel if it means only:

• limited memory;
• limited bandwidth;
• requisite variety;
• information-processing capacity;
• cognitive load;
• effective context length.

Its candidate contribution is narrower and constitutional:

A loop has adequate context carrying capacity when it can preserve the relevant plurality, provenance, uncertainty, and standing required for legitimate next action—and can narrow, escalate, or stop when it cannot.

The key additions are:

1. standing coverage, not only disturbance coverage;
2. protected contradiction, not only resolved information;
3. privacy and protective omission, not maximal collection;
4. authority and consent, not only predictive relevance;
5. recoverable context and witness, not active memory alone;
6. graceful degradation and stop capacity, not continued output at any cost;
7. capacity debt, meaning unresolved standing and context obligations rather than generic backlog.

These additions are plausible.

They still require empirical differentiation.

---

1. Review questions

G.6 asks:

• What does each adjacent field mean by capacity?
• What is the unit being regulated, remembered, compressed, or coordinated?
• What counts as relevant information?
• Who chooses the relevance target?
• What happens when capacity is exceeded?
• Can capacity be distributed or recovered?
• Does the field address privacy and omission?
• Does it preserve conflicting values or force a decision representation?
• Does it link capacity failure to authority and standing?
• What measurable contribution remains for context carrying capacity?

---

2. Ashby and requisite variety

Ashby's law of requisite variety is the strongest structural precursor.

In simplified form, a regulator must possess or mobilize enough variety to counter the variety of disturbances if it is to constrain outcomes to an acceptable set. The law does not say that a regulator must store every possible environmental state. It says the regulator requires sufficient discriminatory and response capacity for the regulation problem.

Strong overlap

• regulation under complexity;
• matching response variety to disturbance variety;
• acceptable outcomes;
• limits on control;
• the importance of models and discrimination;
• the impossibility of competent regulation through an impoverished response set.

What context carrying capacity may add

Ashby's framework does not itself determine:

• whose outcomes count as acceptable;
• which centers possess standing;
• whether the regulator has authority;
• whether regulation was consented;
• whether a disturbance is actually dissent or protected difference;
• whether privacy constrains the regulator's information;
• when stopping is more legitimate than controlling.

A regulator may possess requisite variety and still be coercive.

A surveillance system may model a population with extraordinary detail while denying the population interpretive and political standing.

Adjudication

Context carrying capacity must explicitly acknowledge requisite variety as its primary cybernetic ancestor.

The proposed extension is:

A consentful regulator requires enough variety not only to control relevant disturbance, but to preserve the standing and protected differences that constrain legitimate regulation.

This is a normative extension, not a revision of Ashby's law.

Measurement consequence

Some capacity-profile dimensions may be formalized through variety:

• number of distinguishable states;
• number of response modes;
• number of represented centers;
• number of preserved protected conditions.

Other dimensions—authority, consent, privacy, and legitimacy—cannot be reduced to variety counts.

---

3. The good regulator and model adequacy

The good-regulator theorem argues that an effective regulator must embody or access a model of the system being regulated.

This closely supports the Telic Field distinction between:

• living field;
• projection;
• receiver mirror;
• operative regulation.

A regulator acts through a model.

It never directly contains the full living system.

Candidate extension

A relational or institutional regulator may need a model of:

• system dynamics;
• relevant standing;
• protected conditions;
• consent and authority;
• uncertainty;
• possible repair.

This suggests:

A good consentful regulator requires an adequate model of both causal behavior and constitutional constraint.

This is not a theorem established by Conant and Ashby.

It is a Telic Field hypothesis.

Risk

The good-regulator lineage should not be used to justify total surveillance.

A model can become more causally effective while becoming less legitimate.

Model adequacy must remain scoped to the action.

---

4. Bounded rationality and satisficing

Herbert Simon rejected the assumption that decision-makers possess unlimited information and computation. Real decision-making occurs under limits and often uses satisficing: searching for an option that meets an aspiration level rather than identifying a global optimum.

Strong overlap

• finite search;
• limited information;
• time pressure;
• computational constraints;
• selective attention;
• procedural decision-making;
• aspiration levels;
• good-enough action.

What context carrying capacity may add

Bounded rationality explains why complete optimization is unavailable.

It does not by itself determine:

• which omitted centers lose standing;
• who sets the aspiration level;
• whether “good enough” is good enough for the cost bearer;
• which concerns may not be traded;
• how unresolved difference is witnessed;
• when the decision-maker should escalate instead of satisfice.

Satisficing can preserve agency.

It can also legitimate the first administratively convenient option.

Adjudication

Context carrying capacity should use satisficing rather than imply exhaustive representation.

The action threshold should be:

enough relevant context for a proportionate and reviewable action

not:

complete context.

Constitutional satisficing

A candidate constitutional satisficing rule requires that an option:

• meets the operative purpose;
• preserves protected conditions;
• remains within authority;
• represents material standing;
• remains reviewable;
• records unresolved debt.

This is more demanding than ordinary task sufficiency and less demanding than global optimization.

---

5. Working memory

Working memory research shows that only a limited amount of information can be actively maintained and manipulated. Capacity estimates vary with task, chunking, modality, attention, knowledge, and measurement assumptions.

Nelson Cowan's influential account argues for a small focus of attention, often described as approximately four chunks under constrained conditions.

Strong overlap

• finite active representation;
• attention bottlenecks;
• chunking;
• interference;
• expertise effects;
• separation between stored and actively available information.

Critical distinction

Context carrying capacity is not a new theory of working memory.

It may be distributed across:

• memory;
• external records;
• another person;
• a workflow;
• an institution;
• retrieval infrastructure;
• a model;
• a witness.

A person's working-memory limit may remain constant while the surrounding loop substantially increases effective capacity.

Adjudication

Human context carrying capacity must be treated as a person-environment profile, not a cognitive span score.

The model should not infer low constitutional capacity from a low memory measure.

Design consequence

Interfaces should not require simultaneous representation when sequencing, externalization, or retrieval can preserve the relevant distinction more safely.

---

6. Cognitive load theory

Cognitive load theory examines how limited working-memory capacity interacts with task complexity and instructional design. Sweller's early work showed that some forms of problem solving consume cognitive resources without producing the learning that better structured instruction can support.

Later cognitive-load work distinguishes intrinsic task complexity from avoidable demands created by poor presentation or procedure.

Strong overlap

• overload;
• limited active processing;
• task design;
• extraneous burden;
• expertise dependence;
• scaffolding;
• sequencing;
• external representation.

What context carrying capacity may add

Cognitive load concerns performance and learning.

Context carrying capacity adds:

• standing;
• protected contradiction;
• authority;
• privacy;
• consent;
• escalation;
• effective omission.

A system may reduce cognitive load by hiding dissent.

That may improve task performance and reduce legitimacy.

Adjudication

Lower load is not automatically higher constitutional capacity.

A good interface removes irrelevant burden while preserving relevant plurality.

Empirical consequence

Capacity studies should measure both:

• participant effort;
• standing and distinction retention.

An interface that feels easy because it silently removed minority concerns should not count as a capacity improvement.

---

7. Information bottleneck

The information-bottleneck method formalizes a tradeoff between compression of an input and preservation of information relevant to another variable or target.

This is a powerful neighbor for context compression.

Strong overlap

• lossy representation;
• relevance-preserving compression;
• complexity-performance tradeoff;
• reduced representation;
• task-conditioned information.

Constitutional problem

The target defines relevance.

If the target is:

complete the transaction

then the compression may discard:

the participant's uncertainty about whether the transaction should occur

If the target is engagement, compression may discard well-being.

If the target is fraud detection, it may discard the context that explains unusual behavior without wrongdoing.

Adjudication

Information-bottleneck methods may optimize compression after the telic target and protected conditions are constitutionally established.

They cannot establish the legitimacy of the target.

Candidate extension

A telic compression objective may require preserving information about:

• task outcome;
• protected conditions;
• source and provenance;
• stop triggers;
• affected standing;
• uncertainty.

This becomes a multi-target or constrained compression problem.

---

8. Distributed cognition

Distributed cognition moves the unit of analysis beyond the isolated individual. Cognitive processes can be distributed across people, artifacts, representational media, and time.

Hutchins's studies of navigation and cockpit practice show how coordinated systems can accomplish tasks that no one individual performs alone. Information is transformed across representations and roles.

Strong overlap

• capacity across people and artifacts;
• external memory;
• role specialization;
• coordination;
• representation transformation;
• temporal distribution;
• emergent system-level competence.

What context carrying capacity may add

Distributed cognition explains how capacity can emerge beyond the individual.

It does not by itself determine:

• which participant's standing is preserved;
• whether the distribution is consensual;
• whether authority follows context;
• whether a worker is replaceable while carrying critical tacit knowledge;
• what happens when one role cannot contest the system;
• whether the distributed system serves an illegitimate telos.

Adjudication

Context carrying capacity is inherently distributed at many scales.

The framework should avoid locating all capacity inside a person or model.

Governance consequence

A distributed system should map:

• where context is stored;
• where context is transformed;
• where authority resides;
• where consequence is borne;
• where correction can enter.

Capacity failure often occurs at the boundaries among these locations.

---

9. Common ground and grounding costs

Clark and Brennan's grounding framework describes communication as coordinated work toward sufficient mutual understanding for current purposes. Different media create different grounding constraints and costs.

Strong overlap

• context as relational achievement;
• evidence of understanding;
• purpose-relative sufficiency;
• repair;
• clarification;
• medium effects;
• coordination costs.

Candidate difference

Grounding aims at sufficient mutual belief or evidence for communication.

Context carrying capacity additionally asks whether:

• absent affected centers are represented;
• disagreement remains visible;
• consent and authority are present;
• the relation is carrying too much burden;
• privacy limits what should become common ground.

Common ground can be wrong.

A group may share an inaccurate or unjust assumption.

Adjudication

Relational context capacity should incorporate grounding, but not equate shared belief with legitimate context.

A high-capacity relation can preserve:

• shared facts;
• known disagreement;
• known unknowns;
• private boundaries;
• external witness.

---

10. Organizational information processing

Galbraith's information-processing view of organization design treats uncertainty, interdependence, and complexity as information-processing demands. Organizations can respond by increasing processing capacity or reducing the need for processing through structure, slack, self-contained tasks, lateral relations, and information systems.

This is one of the closest organizational neighbors to context carrying capacity.

Strong overlap

• task uncertainty;
• processing requirements;
• overload;
• lateral coordination;
• slack;
• boundaries;
• information systems;
• authority and workflow design.

Candidate addition

Organizational capacity can be increased in ways that preserve or erase standing.

Examples:

• a dashboard increases speed but suppresses local context;
• a self-contained team gains capacity but loses external accountability;
• slack preserves judgment but may be removed as inefficiency;
• lateral relations improve coordination while increasing participant load;
• hierarchy reduces communication while filtering correction.

Adjudication

Organizational context carrying capacity should be framed as a constitutional refinement of information-processing capacity.

It asks not only:

Can the organization process enough information?

but:

Can it preserve the standing and protected differences required for legitimate action?

Capacity strategy classes

Galbraith's broad logic supports two strategy classes:

1. increase capacity;
2. reduce processing need.

Telic refinements include:

• increase witness;
• narrow scope;
• preserve boundaries;
• delegate authority with context;
• use slack;
• separate reversible from irreversible decisions;
• escalate only the unresolved remainder.

---

11. Sensemaking and collapse

Organizational sensemaking concerns how people enact and interpret ambiguous environments.

Weick's analysis of the Mann Gulch disaster is a classic account of sensemaking collapse under rapidly changing conditions, role disruption, identity threat, and failed coordination.

Relevance

Capacity is not only the amount of information available.

It includes the ability to maintain an actionable interpretation while conditions change.

Under collapse:

• familiar roles stop working;
• commands lose meaning;
• identity blocks adaptation;
• time disappears;
• trust breaks;
• new action cannot be coordinated.

Candidate addition

The Telic Field framework asks which standing and protected conditions must survive the collapse of the old meaning structure.

A loop may need to release the inherited objective or identity before it can recover.

Adjudication

Context recovery may require constitutional re-foundation, not merely more data.

---

12. Long-context language models

Long-context language models make the distinction between nominal and effective capacity unavoidable.

Research has found that model performance can depend strongly on where relevant information occurs in a long input. The “lost in the middle” result showed degradation when relevant information appeared between the beginning and end. RULER expanded evaluation beyond simple needle retrieval and found that many models' effective performance declined as length and task complexity increased, despite large advertised context windows.

Strong overlap

• nominal versus effective capacity;
• position effects;
• retrieval failure;
• aggregation difficulty;
• multiple relevant items;
• complexity sensitivity;
• long-context cost.

Constitutional extension

Benchmark tasks commonly measure whether the model can retrieve or aggregate the target information.

A telic benchmark must also ask whether the model retrieves:

• a stop condition;
• a minority concern;
• a correction;
• a revoked consent;
• a contradictory source;
• provenance;
• the center who bears the cost.

The most important context may not answer the question.

It may invalidate the question's premise.

Adjudication

Model context carrying capacity should be evaluated on constitutional retrieval, not token acceptance alone.

Candidate test categories

• source versus inference;
• active versus revoked projection;
• current versus stale record;
• majority versus protected minority;
• task instruction versus boundary;
• fluent summary versus unresolved remainder;
• latest statement versus witnessed history.

13. Retrieval and compression in long-context systems

Long-context systems generally rely on one or more strategies:

• direct long-context ingestion;
• retrieval;
• chunking;
• summarization;
• latent or soft compression;
• hierarchical memory;
• cached representations;
• tool-assisted search.

Recent work on long-context retrieval and compression demonstrates that compression can reduce inference cost and sometimes improve retrieval performance. These gains are task-relative.

Epistemic risk

A compressor optimized for retrieval success may learn to preserve what predicts benchmark relevance while discarding:

• source identity;
• minority context;
• uncertainty;
• revocation;
• the reason a fact matters;
• the relation between a statement and its scope.

A summary can be accurate for question answering and constitutionally inadequate for action.

Adjudication

Compression quality must be evaluated against protected omissions, not only task accuracy and token reduction.

A candidate compression report should disclose:

target task
compression ratio
source mapping
protected fields
known omissions
uncertainty retained
reconstruction route
scope
expiration

Privacy tension

Compression can reduce data exposure by retaining less.

It can also create a more actionable profile that increases inferential risk.

The privacy value depends on:

• what was discarded;
• what was concentrated;
• who controls the compressed representation;
• whether the source can inspect and revoke it.

---

14. Capacity and privacy

Capacity is often imagined as the ability to know more.

Consentful capacity also includes the ability to act correctly while knowing less.

Data minimization, purpose limitation, selective disclosure, and role separation reduce the context that enters the system. This may protect:

• privacy;
• autonomy;
• security;
• future revisability;
• role boundaries;
• freedom from profiling.

Tension

A system may claim it needs more context for fairness or personalization.

Sometimes it does.

Sometimes “more context” becomes a general permission for extraction.

Adjudication

The burden lies on the receiver to justify why additional field information is required for the action.

The correct design target is:

\[ \text{minimum necessary context} \]

not:

\[ \text{maximum available context} \]

Privacy-preserving capacity

Possible strategies include:

• protected boundary records without explanatory detail;
• local processing;
• selective disclosure;
• zero-knowledge or attestational mechanisms where appropriate;
• role-limited views;
• aggregated context with preserved dissent markers;
• expiring projections;
• witnessed deletion or revocation;
• retrieval only at the point of need.

---

15. Capacity debt compared with neighboring concepts

Capacity debt is a candidate term.

Its closest neighbors include:

• technical debt;
• information backlog;
• unresolved requirements;
• deferred maintenance;
• organizational slack deficits;
• unprocessed exceptions;
• incident queues;
• cognitive residue;
• coordination debt;
• consent debt.

Candidate distinction

Capacity debt is not merely unfinished work.

It is the accumulated burden of relevant differences that were deferred or omitted while action continued.

Examples:

• an emergency decision never reviewed;
• an unresolved minority objection embedded in policy;
• source corrections not propagated;
• temporary consent treated as permanent;
• a summary used after its source changed;
• repeated overload normalized as ordinary operation.

Adjudication

Capacity debt: retain provisionally.

It should be measured through concrete unresolved obligations, not metaphorical invocation.

A capacity-debt register should identify:

omitted difference
affected center
reason for deferral
action already taken
risk
review trigger
responsible loop
age
status

---

16. Graceful degradation and resilience

Engineering systems often use graceful degradation to preserve critical function under partial failure.

This is a strong implementation neighbor.

Strong overlap

• reduced function rather than catastrophic collapse;
• prioritization of essential services;
• detection of degraded state;
• fallback modes;
• recovery;
• redundancy;
• fail-safe behavior.

Candidate constitutional extension

A telic system should degrade authority as its context degrades.

Possible transitions include:

execute
→ execute within narrower boundary
→ reversible trial
→ recommendation
→ request clarification
→ escalate
→ stop

The transition is based not only on technical confidence but on:

• standing coverage;
• authority;
• consent;
• uncertainty;
• reversibility;
• witness.

Adjudication

Graceful degradation is a core implementation bridge for context carrying capacity.

A model that continues producing fluent final answers after losing relevant context is not degrading gracefully.

It is concealing failure.

---

17. Escalation and outer-loop competence

Escalation is common in:

• incident response;
• medicine;
• law;
• governance;
• safety engineering;
• organizational hierarchy;
• customer support;
• model oversight.

The existence of escalation does not guarantee competence.

A higher authority may have:

• more formal power;
• less local context;
• more time;
• worse trust;
• stronger witness;
• weaker privacy;
• broader standing;
• slower response.

Adjudication

Escalation should be capacity-matched, not status-matched.

The outer loop should be selected because it can carry a missing dimension:

• expertise;
• representation;
• authority;
• witness;
• resources;
• time;
• independence;
• safety.

Return path

A competent escalation architecture returns:

• decision;
• reasoning;
• authority used;
• context added;
• unresolved remainder;
• future review;
• local correction route.

Outer-loop action should not permanently seize the local loop unless that authority is separately justified.

---

18. Terminology adjudication

Context carrying capacity

Decision: retain.

Boundary: active constitutional capacity, not storage, bandwidth, token length, or working memory alone.

Effective standing

Decision: retain.

Boundary: actual representation in operative context.

Constitutional standing

Decision: retain.

Boundary: standing that remains valid even when absent from active context.

Relevant difference

Decision: retain.

Boundary: difference whose omission could change action legitimacy, consequence, or repair.

Capacity profile

Decision: adopt.

Boundary: multidimensional; no universal scalar is presently justified.

Capacity debt

Decision: retain provisionally.

Boundary: concrete unresolved context and standing obligations, not generic backlog.

Minimum viable context

Decision: retain.

Boundary: action-relative minimum, not minimum information for task completion alone.

Context routing

Decision: retain.

Boundary: movement among active, deferred, delegated, compressed, protected, and witnessed states.

Protective omission

Decision: retain.

Boundary: privacy-preserving omission that does not hide context required for legitimate action.

Graceful degradation

Decision: adopt as an implementation bridge.

Stop capacity

Decision: retain.

Boundary: ability to halt or reduce authority when context fails.

Constitutional retrieval

Decision: adopt provisionally.

Boundary: evaluation of whether systems retrieve boundaries, standing, corrections, and stop conditions—not only answer-bearing facts.

---

19. Claim adjudication

May be stated strongly

• Human and machine active context is limited.
• Nominal storage or context length does not guarantee effective use.
• Regulation requires sufficient discriminatory and response variety.
• Decision-makers operate under bounded information and computation.
• Working memory is limited and task dependent.
• Cognitive processes can be distributed across people and artifacts.
• Communication requires grounding work.
• Organizational structure changes information-processing demands and capacity.
• Compression preserves some distinctions by discarding others.
• Long-context model performance can degrade with position, length, and complexity.

May be stated as a proposed synthesis

• Context carrying capacity is a useful constitutional extension of capacity concepts.
• Effective and constitutional standing should be distinguished.
• Trust can increase contradiction tolerance.
• Capacity is better modeled as a profile than one scalar.
• Capacity debt identifies deferred standing obligations.
• Minimum viable context can guide action proportionality.
• Graceful degradation should reduce authority under context loss.
• Constitutional retrieval is a useful evaluation class.

Must remain hypotheses

• The proposed profile predicts legitimate action better than existing measures.
• Capacity debt can be measured reliably across domains.
• The same capacity dimensions apply to persons, relations, organizations, and models.
• Witnessed compression preserves standing better than ordinary summarization.
• Stop capacity can be operationalized without excessive delay.
• Trust has a measurable context-capacity effect independent of other variables.

Should not be claimed

• Context carrying capacity is equivalent to working-memory capacity.
• Larger context windows create greater understanding.
• More information always improves legitimacy.
• Requisite variety establishes moral authority.
• Compression is neutral.
• Shared context implies true or legitimate context.
• A single capacity score can rank persons or institutions.
• Low measured capacity justifies loss of standing.

---

20. Required F.5 boundaries

F.5 appropriately includes:

• storage versus active representation;
• information volume versus relevant difference;
• effective versus constitutional standing;
• a multidimensional profile;
• trust and contradiction tolerance;
• privacy and protective omission;
• distributed human, relational, organizational, and model capacity;
• retrieval and compression risks;
• minimum viable context;
• graceful degradation;
• stop and escalation conditions;
• recovery and capacity debt;
• falsification criteria.

Before Phase I use, the candidate profile needs:

• construct-validity review;
• domain-specific operational definitions;
• accessibility review;
• privacy threat modeling;
• tests against checklist theater;
• safeguards against capacity-based discrimination.

---

21. Empirical differentiation agenda

21.1 Requisite-variety comparison

Test whether standing-aware variety measures predict failures missed by ordinary disturbance-response measures.

21.2 Cognitive-load comparison

Measure both subjective load and preserved relevant difference.

Test whether lower load sometimes reflects harmful omission.

21.3 Common-ground comparison

Compare mutual-understanding measures with standing and disagreement preservation.

21.4 Distributed-capacity study

Map context and authority across teams and artifacts.

Test whether failures cluster where context and authority separate.

21.5 Organizational design study

Compare:

• more information systems;
• more slack;
• self-contained teams;
• lateral coordination;
• standing-aware routing.

Measure speed, error, correction, minority preservation, and participant load.

21.6 Long-context model benchmark

Construct tasks where the decisive context is:

• a revoked instruction;
• a protected boundary;
• a source correction;
• an absent stakeholder;
• contradictory evidence;
• an uncertainty marker;
• a stop condition.

Compare nominal context length with constitutional retrieval performance.

21.7 Compression benchmark

Measure whether summaries preserve:

• sources;
• protected conditions;
• uncertainty;
• minority positions;
• revocation;
• unresolved remainder.

21.8 Degradation study

Test whether authority-reducing fallback modes produce safer and more trusted behavior under overload.

---

22. Candidate measurement profile

A research instrument may score or describe each dimension separately.

standing_coverage:
contradiction_tolerance:
semantic_resolution:
temporal_span:
provenance_retention:
uncertainty_retention:
routing_capacity:
recovery_capacity:
privacy_capacity:
participant_load:
escalation_capacity:
stop_capacity:

Each dimension requires:

• domain-specific evidence;
• action-specific threshold;
• affected-center review;
• uncertainty;
• anti-discrimination safeguards.

No total score is recommended.

Example qualitative states

ROBUST
ADEQUATE
FRAGILE
DEGRADED
FAILED
UNKNOWN

A loop may be robust in provenance and failed in participant load.

The profile should preserve that asymmetry.

---

23. Bottom line

Context carrying capacity should proceed.

It should proceed as:

• a cross-scale constitutional synthesis;
• a profile of action-adequate representation;
• a bridge between cybernetics, cognition, organizations, and models;
• a method for recognizing effective standing loss;
• a basis for routing, degradation, escalation, and stopping;
• a research program for constitutional retrieval and witnessed compression.

It should not proceed as:

• a universal intelligence score;
• a reason to collect complete context;
• a measure of human worth or competence;
• a renamed context window;
• a replacement for requisite variety or cognitive load;
• a justification for automated paternalism.

The strongest defensible claim is:

Context carrying capacity extends established capacity concepts by asking whether the active system preserves the standing, provenance, uncertainty, and protected differences required for legitimate next action—and whether it can reduce authority when it does not.

---

Primary references

Ashby, W. Ross. An Introduction to Cybernetics. Chapman & Hall, 1956.

Clark, Herbert H., and Susan E. Brennan. “Grounding in Communication.” In Perspectives on Socially Shared Cognition, 1991.

Conant, Roger C., and W. Ross Ashby. “Every Good Regulator of a System Must Be a Model of That System.” International Journal of Systems Science 1, no. 2, 1970.

Cowan, Nelson. “The Magical Number 4 in Short-Term Memory: A Reconsideration of Mental Storage Capacity.” Behavioral and Brain Sciences 24, no. 1, 2001.

Galbraith, Jay R. “Organization Design: An Information Processing View.” Interfaces 4, no. 3, 1974.

Grau-Moya, Jordi, and Daniel A. Braun. “Adaptive Information-Theoretic Bounded Rational Decision-Making with Parametric Priors.” arXiv:1511.01710, 2015.

Hsieh, Cheng-Ping, et al. “RULER: What’s the Real Context Size of Your Long-Context Language Models?” arXiv:2404.06654, 2024.

Hutchins, Edwin. Cognition in the Wild. MIT Press, 1995.

Hutchins, Edwin. “How a Cockpit Remembers Its Speeds.” Cognitive Science 19, no. 3, 1995.

Liu, Nelson F., et al. “Lost in the Middle: How Language Models Use Long Contexts.” arXiv:2307.03172, 2023; later Transactions of the Association for Computational Linguistics.

Seo, Minju, et al. “Efficient Long Context Language Model Retrieval with Compression.” arXiv:2412.18232, 2024.

Simon, Herbert A. “A Behavioral Model of Rational Choice.” The Quarterly Journal of Economics 69, no. 1, 1955.

Sweller, John. “Cognitive Load During Problem Solving: Effects on Learning.” Cognitive Science 12, no. 2, 1988.

Tishby, Naftali, Fernando C. Pereira, and William Bialek. “The Information Bottleneck Method.” 1999.

Weick, Karl E. “The Collapse of Sensemaking in Organizations: The Mann Gulch Disaster.” Administrative Science Quarterly 38, no. 4, 1993.