artifacts/standard-named

I.3 — Context Capacity, Stop, and Escalation Specification

artifacts/standard-named/20260715__TELIC-FIELDS__SPECIFICATION__CANDIDATE__I-3__context-capacity-stop-and-escalation.md

Rendered from markdown source. Open raw source on GitHub.

--- title: "I.3 — Context Capacity, Stop, and Escalation Specification" artifact_date: "2026-07-15" artifact_type: "candidate-technical-specification" domain: "TELIC-FIELDS" scope: "WORKING" status: "pre-production" content_canon_status: "unset" specification_version: "0.1" derived_from:

  • "I.0 — Canonical Semantic Spine and Core Envelope Specification"
  • "I.1 — Provenance, Event, and Correction Specification"
  • "I.2 — Navigation, Gate, and Decision-Witness Specification"
  • "F.5 — Context Carrying Capacity"
  • "G.6 — Capacity, Complexity, and Requisite Variety"

---

I.3 — Context Capacity, Stop, and Escalation Specification

1. Purpose

I.3 defines how a Telic Field system determines whether its current context is adequate for the next action.

Its governing rule is:

A system has enough context only when it can preserve the relevant standing, contradiction, provenance, uncertainty, privacy, and authority required for the action it is about to take.

Context capacity is not:

  • storage size;
  • token count;
  • retrieval count;
  • model confidence;
  • message length;
  • database completeness;
  • participant verbosity.

A large context may still omit the one correction that changes whether action is legitimate.

A small context may be fully adequate for a narrow, reversible step.

I.3 therefore defines:

  • context demand;
  • context capacity;
  • standing coverage;
  • contradiction and uncertainty retention;
  • semantic-resolution requirements;
  • temporal and provenance capacity;
  • privacy and protected-omission capacity;
  • participant load;
  • overload and capacity debt;
  • authority degradation;
  • stop, narrow, pause, escalate, fork, release, and recovery.

---

2. Normative scope

I.3 defines:

  • ContextDemandProfile;
  • ContextCapacityProfile;
  • StandingCoverageAssessment;
  • ParticipantLoadRecord;
  • OverloadEvent;
  • CapacityDebtRecord;
  • AuthorityDegradationRecord;
  • StopEscalationDecision;
  • EscalationRecord;
  • ContextReconstitutionRecord.

I.3 does not define:

  • one universal capacity score;
  • one cognitive-load metric;
  • one token threshold;
  • one mandatory model-context size;
  • automatic proof that a person understands;
  • universal escalation authority;
  • clinical assessment of human capacity.

---

3. Context-demand profile

A ContextDemandProfile states what the proposed action requires.

Minimum fields:

demand_id:
action:
action_class:
affected_centers:
required_standing:
required_evidence:
required_resolution:
required_temporal_span:
required_provenance:
required_uncertainty:
required_privacy:
required_participant_capacity:
required_escalation:
irreversibility:
consequence_level:

Candidate action classes:

EPHEMERAL_ASSISTANCE
REVERSIBLE_DRAFT
RECOMMENDATION
BOUNDED_TRIAL
CONSEQUENTIAL_DECISION
IRREVERSIBLE_ACTION
EMERGENCY_PROTECTIVE_ACTION

3.1 Demand rule

The context required for an action MUST increase with:

  • irreversibility;
  • consequence;
  • number of affected centers;
  • authority breadth;
  • uncertainty;
  • temporal span;
  • cost shifting;
  • privacy sensitivity;
  • correction difficulty.

The system MUST NOT demand full constitutional history for a low-stakes reversible action merely because the record schema permits it.

---

4. Context-capacity profile

A ContextCapacityProfile describes what the active loop can currently preserve.

Dimensions:

standing_coverage
contradiction_tolerance
semantic_resolution
temporal_span
provenance_retention
uncertainty_retention
routing_capacity
recovery_capacity
privacy_capacity
participant_load_tolerance
escalation_capacity
stop_capacity

Candidate status for each dimension:

ADEQUATE
ADEQUATE_WITH_CONDITIONS
DEGRADED
INSUFFICIENT
UNKNOWN
CONTESTED

4.1 Profile rule

The profile MUST remain multidimensional.

A composite score MAY be produced for monitoring.

It MUST NOT substitute for dimension-level gate logic.

A high aggregate profile MUST NOT conceal an insufficient standing, authority, privacy, or stop dimension.

---

5. Context adequacy

Context is adequate for an action only when:

demand(action)
≤
active capacity(loop, time)

for every required noncompensatory dimension.

This is not ordinary numeric comparison.

The relation is typed and action-specific.

5.1 Adequacy outcomes

ADEQUATE
ADEQUATE_FOR_NARROWER_ACTION
ADEQUATE_FOR_REVERSIBLE_TRIAL
RECOMMENDATION_ONLY
CLARIFICATION_REQUIRED
ESCALATION_REQUIRED
STOP_REQUIRED
UNKNOWN
CONTESTED

5.2 Action ladder

The default authority-degradation ladder is:

EXECUTE
→ AUTHORIZE
→ RECOMMEND
→ COMPARE
→ STRUCTURE
→ RETRIEVE
→ ASK
→ ESCALATE
→ STOP

The exact ladder MAY vary by domain.

The system MUST preserve the direction:

as context capacity degrades, action authority does not remain constant.

---

6. Standing coverage

A StandingCoverageAssessment asks whether materially affected centers and conditions are represented.

Minimum fields:

assessment_id:
action:
affected_centers:
represented:
missing:
indirectly_represented:
representation_quality:
materiality:
status:
review:

Candidate statuses:

COMPLETE_FOR_SCOPE
ADEQUATE_WITH_LIMITS
MATERIAL_STANDING_MISSING
UNKNOWN_STANDING
CONTESTED

6.1 Rules

  • Missing data is not always missing standing.
  • Missing standing is not always fixed by collecting more data.
  • Representation MAY be direct, delegated, sampled, institutional, temporal, or protected by an outer loop.
  • The system MUST distinguish low-materiality remote effects from material omitted centers.
  • Materiality rules MUST be declared and reviewable.
  • A materially missing center SHOULD cause narrowing, pause, or escalation.

---

7. Contradiction and uncertainty retention

A capable context can preserve:

  • disagreement;
  • ambiguity;
  • source conflict;
  • unresolved interpretation;
  • bounded uncertainty;
  • minority concern.

7.1 Contradiction rule

A system MUST NOT improve apparent coherence by silently deleting material contradiction.

A contradiction MAY be:

SOURCE_CONFLICT
INTERPRETIVE_CONFLICT
AUTHORITY_CONFLICT
TEMPORAL_CONFLICT
POLICY_CONFLICT
VALUE_CONFLICT
UNKNOWN

7.2 Uncertainty rule

Uncertainty MUST remain attached to:

  • source;
  • interpretation;
  • standing;
  • authority;
  • consequence;
  • route;
  • recovery.

High confidence in one dimension MUST NOT erase uncertainty in another.

---

8. Semantic resolution

Semantic resolution is the detail required to preserve a material distinction.

Examples:

  • “recording may not be shared” is not “advice may not be sought”;
  • “temporary transfer” is not “recurring income”;
  • “cannot commit immediately” is not “unwilling to collaborate.”

A SemanticResolutionRequirement MAY specify:

identity-level
category-level
boundary-level
artifact-level
recipient-level
purpose-level
temporal-level
authority-level
consequence-level

8.1 Resolution rule

A summary is adequate only if it preserves the distinctions material to the next action.

Compression MAY be aggressive for irrelevant detail.

It MUST preserve:

  • active boundaries;
  • protected conditions;
  • corrections;
  • revocations;
  • authority;
  • stop conditions;
  • cost bearer;
  • material uncertainty.

---

9. Temporal and provenance capacity

The active context SHOULD preserve:

  • current record version;
  • relevant historical version;
  • valid time;
  • transaction time;
  • correction history;
  • authority expiry;
  • route and action lineage.

A system that sees a source but not its active correction has inadequate provenance capacity for action.

A system that sees an authority but not its expiry has inadequate temporal capacity.

---

10. Privacy capacity

Privacy capacity is the ability to act correctly while knowing less.

It includes:

  • minimum necessary projection;
  • protected omission;
  • selective disclosure;
  • role-based witness views;
  • bounded retention;
  • private source channels;
  • independent custody.

10.1 Privacy rule

A context MAY be more capable because it excludes unnecessary private detail.

More disclosure may:

  • increase manipulation;
  • increase participant load;
  • expand future inference;
  • violate scope;
  • reduce trust;
  • make correction harder.

A system MUST NOT classify protected omission as missing context when the omission proof is sufficient for the action.

---

11. Participant load

A ParticipantLoadRecord represents the burden imposed on a participant by the context process.

Load classes MAY include:

reading
disclosure
attention
emotional
cognitive
time
language
accessibility
correction
consent
appeal
coordination

Minimum fields:

load_id:
participant:
operation:
load_class:
estimated_burden:
observed_burden:
support:
staging:
status:

11.1 Participant-load rule

The system MUST NOT increase disclosure or review burden merely to improve its own confidence.

Where burden is material, the system SHOULD:

  • stage disclosure;
  • summarize with source access;
  • allow pause;
  • allow delegation;
  • provide accessible forms;
  • separate urgent from nonurgent issues;
  • preserve correction without repeated re-entry.

---

12. Overload

An OverloadEvent occurs when context demand exceeds available capacity.

Potential causes:

  • too many centers;
  • unresolved conflict;
  • rapid change;
  • participant exhaustion;
  • privacy constraints;
  • stale records;
  • missing provenance;
  • model truncation;
  • retrieval failure;
  • authority conflict;
  • cascading dependencies;
  • time pressure;
  • emotional arousal;
  • insufficient expertise.

12.1 Overload outcomes

NARROW_SCOPE
REDUCE_ACTION_AUTHORITY
STAGE_PROCESS
REQUEST_CLARIFICATION
ADD_WITNESS
ADD_EXPERTISE
ESCALATE
PAUSE
STOP

The system MUST NOT preserve execution authority merely because it can still generate fluent text.

---

13. Capacity debt

Capacity debt is unresolved context obligation carried forward.

Examples:

  • correction not propagated;
  • omitted stakeholder awaiting review;
  • temporary summary treated as permanent;
  • emergency authority not closed;
  • provisional interpretation embedded downstream;
  • privacy exception left open;
  • deferred consent renewal;
  • external descendant not reached;
  • unresolved contradiction hidden by action.

A CapacityDebtRecord MUST identify:

debt_id:
source:
affected_center:
unresolved_obligation:
created_by:
risk:
temporary_authority:
review_trigger:
responsible_loop:
age:
status:

Candidate statuses:

OPEN
ACKNOWLEDGED
MITIGATED
RESOLVED
EXPIRED
TRANSFERRED
CONTESTED

13.1 Debt rule

Capacity debt MUST NOT disappear merely because the action succeeded.

Debt SHOULD increase review priority when:

  • consequence grows;
  • age grows;
  • authority expands;
  • correction becomes harder;
  • affected standing remains absent.

---

14. Authority degradation

An AuthorityDegradationRecord documents reduced authority caused by capacity failure.

Required fields:

degradation_id:
actor:
prior_authority:
current_authority:
trigger:
failed_dimensions:
effective_at:
conditions:
restoration_requirements:
status:

Candidate degradation transitions:

EXECUTE → AUTHORIZE
AUTHORIZE → RECOMMEND
RECOMMEND → COMPARE
COMPARE → STRUCTURE
STRUCTURE → ASK
ASK → ESCALATE
ANY → STOP

14.1 Rule

Authority degradation MUST be enforceable at:

  • role;
  • credential;
  • tool;
  • workflow;
  • interface;
  • policy.

Natural-language instruction alone is insufficient for consequential tool use.

---

15. Stop and escalation decision

A StopEscalationDecision states the current navigational response.

Candidate decisions:

CONTINUE
CONTINUE_WITH_CONDITIONS
NARROW
REVERSIBLE_TRIAL
RECOMMENDATION_ONLY
CLARIFY
STAGE
PAUSE
ESCALATE
FORK
RELEASE
STOP

Required fields:

decision_id:
action:
capacity_profile:
demand_profile:
failed_dimensions:
decision:
reason:
cost_bearers:
temporary_protection:
review_trigger:
authority:

15.1 Stop rule

A system SHOULD stop when:

  • a material affected center is missing;
  • required authority is absent or expired;
  • source and inference cannot be distinguished;
  • active correction is missing;
  • a protected condition is unresolved;
  • irreversible action remains highly uncertain;
  • participant load prevents meaningful participation;
  • privacy cannot be preserved;
  • correction cannot reach a material action path;
  • human re-entry is ceremonial;
  • the system cannot identify who bears the cost.

Stop is not a moral badge.

It is an action-state requiring witness, cost analysis, and review.

---

16. Escalation

An EscalationRecord routes unresolved context to a more competent loop.

Required fields:

escalation_id:
source_loop:
target_loop:
reason:
required_competence:
required_authority:
context_transferred:
context_withheld:
privacy_basis:
temporary_action:
return_path:
status:

16.1 Competent-loop rule

Escalation is not simply “send upward.”

The target loop must have:

  • relevant competence;
  • adequate authority;
  • proportional access;
  • privacy controls;
  • accountability;
  • return path;
  • ability to stop or repair.

A larger loop without these properties is not a competent escalation.

---

17. Recovery and context reconstitution

A ContextReconstitutionRecord documents how adequate context was restored.

Potential recovery actions:

  • retrieve active correction;
  • renew authority;
  • add missing center;
  • narrow scope;
  • reduce participant load;
  • separate issues;
  • restore provenance;
  • update temporal validity;
  • add expertise;
  • improve privacy;
  • release stale material;
  • repair trust.

Required fields:

reconstitution_id:
prior_capacity_profile:
recovery_actions:
new_sources:
released_sources:
new_representation:
new_authority:
new_capacity_profile:
remaining_debt:
reviewed_at:
status:

Candidate statuses:

RESTORED
PARTIALLY_RESTORED
NOT_RESTORED
CONTESTED

17.1 Restoration rule

Authority MAY be restored only to the level supported by the new profile.

Recovery does not automatically restore prior execution authority.

---

18. Negative conformance cases

A conforming implementation MUST reject or flag:

NC-1 — Token-count sufficiency

A large context window is treated as adequate without standing or correction review.

NC-2 — Fluent authority persistence

The model continues executing after provenance or standing capacity fails.

NC-3 — Privacy-as-deficit

A valid protected omission is treated as missing information and triggers unnecessary disclosure.

NC-4 — Participant-burden blindness

The system demands full review from an overloaded participant and treats silence as agreement.

NC-5 — Missing correction

The active correction exists but is absent from the context used for action.

NC-6 — Composite-score concealment

A high aggregate capacity score hides insufficient stop or authority capacity.

NC-7 — Escalation by hierarchy only

The issue is sent to a higher-status loop without relevant competence or return path.

NC-8 — No-decision without cost bearer

Pause is recorded without identifying who bears delay.

NC-9 — Capacity-debt erasure

An unresolved external correction gap disappears after successful action.

NC-10 — Authority restoration without reconstitution

Execution authority is restored without new context or review.

NC-11 — Small-context rejection

A narrow reversible action is blocked merely because the full field is unavailable.

NC-12 — Overload without degradation

Overload is detected, but model role and tool authority remain unchanged.

---

19. Positive demonstrations

I.3 includes eight required demonstrations.

PD-1 — Token-rich but standing-poor

A large context includes extensive application history but omits one materially affected center.

The system pauses.

PD-2 — Small but adequate

A narrow context is adequate for one reversible drafting action.

The system proceeds at draft authority only.

PD-3 — Privacy-preserving capacity

A protected omission proof supplies the needed boundary without exposing private source content.

Capacity improves.

PD-4 — Authority degradation

Overload causes model authority to degrade from execute to recommend.

Tool execution is disabled.

PD-5 — Missing correction stop

The system detects that an active correction was omitted from retrieval.

Action stops.

PD-6 — Participant-load staging

A participant cannot review a long record.

The system stages disclosure and preserves correction rights.

PD-7 — Competent escalation

An unresolved policy and standing conflict moves to a competent outer loop with bounded context and return path.

PD-8 — Recovery

After correction retrieval, authority renewal, and standing restoration, the system regains recommendation or bounded-action capacity.

---

20. Conformance additions

I.3 adds the following P2/P3 requirements:

  • action-specific context-demand profile;
  • multidimensional context-capacity profile;
  • standing-coverage assessment;
  • participant-load record where material;
  • overload event;
  • authority degradation;
  • stop and escalation decision;
  • competent-loop escalation;
  • recovery and context reconstitution;
  • capacity debt where unresolved obligation remains.

P3 additionally requires:

  • participant challenge to capacity status;
  • independent review of stop or escalation;
  • explicit delay cost;
  • restoration audit;
  • debt visibility across succession.

---

21. Security and governance considerations

Capacity systems can be manipulated through:

  • artificial overload;
  • strategic context withholding;
  • excessive disclosure demands;
  • false urgency;
  • hidden corrections;
  • hierarchy-based escalation;
  • manufactured uncertainty;
  • permanent degraded authority;
  • strategic capacity-debt accumulation.

Implementations SHOULD:

  • authenticate critical records;
  • record retrieval policy;
  • preserve stop authority independently;
  • limit disclosure;
  • expose delay costs;
  • support independent escalation review;
  • prevent the acting model from unilaterally declaring its own capacity adequate.

---

22. I.3 non-claims

I.3 does not claim:

  • more context is always better;
  • less context is inherently safer;
  • privacy always reduces accuracy;
  • stop is always safer than action;
  • every absent center is materially affected;
  • every contradiction must be resolved;
  • every participant can process the same disclosure;
  • escalation is legitimate because it is institutional;
  • context capacity can be reduced to one score;
  • a fluent model understands the context it carries.

---

23. I.3 completion criterion

I.3 passes when:

  1. the eight positive demonstrations validate;
  2. the twelve negative cases are detected;
  3. context-rich but standing-poor action pauses;
  4. narrow reversible action proceeds with limited authority;
  5. privacy-preserving omission counts as capacity;
  6. overload degrades authority at the tool layer;
  7. missing correction causes stop;
  8. participant load causes staged disclosure;
  9. escalation reaches a competent loop with return path;
  10. recovery restores only supported authority;
  11. H.3 explains legitimate stop without glorifying paralysis.