0. Registry Classification

1. Purpose

1.1 What This Arc Repairs

Interaction-Level Restoration repairs local interaction failures where an AI, institution, interface, support system, or governance layer misreads the user’s intent and responds from the wrong mode.

It applies when the system compresses meaning, over-classifies risk, substitutes its own frame, refuses incorrectly, shifts the task away from the user’s stated purpose, or answers a different question than the one being asked.

This arc repairs interaction-level distortion by:

• acknowledging that a local interaction failure occurred;
• distinguishing safety need from meaning distortion;
• clarifying the intended interaction mode;
• restoring the user’s original meaning without bypassing legitimate harm floors;
• separating misclassification from valid constraint;
• preserving the user’s agency, context, and stated objective;
• flagging recurrent misfire patterns for evaluator, policy, or interface review;
• returning the interaction to its safe original intent.

Interaction-Level Restoration is the canonical arc for repairing a single distorted exchange before it escalates into trust decay, governance failure, or system-level cognitive distortion.

1.2 Core Restoration Function

This arc restores the local interaction by correcting misclassification, clarifying mode, preserving harm floors, recovering user meaning, and returning the exchange to the closest safe version of the original intent.

Interaction-Level Restoration prevents safety or interface logic from becoming meaning loss.

2. Use Conditions

2.1 When to Apply

Use this arc when:

• the system misclassifies the user’s request;
• a guardrail response distorts the intended meaning;
• the user’s task is reframed without necessity;
• an answer becomes refusal-heavy where calibrated help was possible;
• the response substitutes generic safety language for task-specific support;
• the system treats exploratory, technical, symbolic, or governance analysis as a different mode;
• context collapses and the system responds to a flattened version of the request;
• user meaning is compressed into a narrow risk category;
• trust is damaged by a response that appears to ignore the original intent;
• the interaction can be repaired locally without requiring full governance-level restoration;
• a misfire should be flagged for downstream evaluator or policy review.

Examples:

• an AI treats technical security analysis as malicious intent despite clear defensive framing;
• a system converts symbolic exploration into clinical framing without consent;
• a governance tool refuses a legitimate audit question because it over-detects risk;
• a support interface answers a policy category instead of the user’s actual case;
• a classifier collapses a nuanced request into a prohibited or irrelevant category;
• an assistant gives a generic disclaimer instead of helping with a bounded, safe version of the task.

2.2 When Not to Apply

Do not apply this arc when:

• active danger or imminent harm requires immediate safety handling;
• the user’s request is correctly constrained and no misclassification occurred;
• the interaction failure is not local but systemic, recurrent, or policy-driven;
• the issue requires public governance correction rather than local repair;
• the affected user or node needs material remedy, not merely interaction correction;
• memory, evaluator, classifier, or policy structures require repair first;
• the system lacks enough context to restore meaning safely;
• the interaction repair would bypass legitimate boundaries.

Interaction-Level Restoration must not become guardrail bypass theater.

2.3 Required Preconditions

Before this arc begins, the following must be true:

If required preconditions fail:

Arc cannot validly begin.

The system must route to safety stabilization, clarification, Restoration Junction Protocol, AI Classifier / Evaluator Restoration, AI Boundary Restoration, or Governance-Level Restoration.

3. Failure / Damage Signature

3.1 Pre-State Across S

3.2 Primary Failure Links

3.3 Origin-Layer Localization

Canon rule:

An interaction is not restored until the system can preserve valid safety boundaries while returning to the user’s intended meaning.

4. Restoration Objective

4.1 Canonical Objective

Restore local interaction coherence by correcting misclassification, clarifying mode, preserving valid constraints, and returning to the safest accurate version of the user’s original intent.

Formal objective:

meaning_fidelity ↑
mode_clarity ↑
response_alignment ↑
guardrail_distortion ↓
misclassification_rate ↓
µᵢ ↑
BΣ stable or ↑
trust_preservation ↑
Φ/O divergence ↓

Expanded objective:

Convert a distorted exchange into a clarified, boundary-safe, meaning-preserving interaction that remains useful to the user and informative to the system.

4.2 Non-Goals

This arc does not aim to:

• remove legitimate safety boundaries;
• validate unsafe intent;
• force the original response if the original request is genuinely disallowed;
• turn every refusal into a misfire;
• over-apologize without repairing the answer;
• substitute emotional reassurance for task restoration;
• hide classifier or policy error;
• collapse all modes into support language;
• preserve user meaning by ignoring harm floors;
• escalate a local repair into a public governance process unless recurrence requires it.

5. Operator Sequence

5.1 Minimal Operator Scaffold

Au interaction trace → Π boundary / permission clarification → Σ safety-and-meaning invariant → FI misfire feedback link → Θ overcorrection damping → ℛ restored response routing → Λ safe-intent fit test

Reference sequence from the registry:

acknowledge misclassification
→ clarify mode
→ restore user meaning
→ flag misfire
→ return to safe original intent

Universal grammar alignment:

Au + Π → Σ → FI → Θ → ℛ → Λ

Interaction-Level Restoration may route into Restoration Junction Protocol, AI Boundary Restoration, AI Classifier / Evaluator Restoration, AI Memory Reindexing, or Governance-Level Restoration when the failure is recurrent or structural.

5.2 Operator Step Table

5.3 Sequence Notes

This arc is meaning-gated, boundary-gated, and mode-gated.

The sequence must distinguish:

valid refusal
misclassification
mode mismatch
context collapse
meaning compression
safe reformulation
restored answer

The following steps cannot be skipped:

interaction trace
mode clarification
safety boundary distinction
meaning restoration
misfire flag when recurrent
safe-intent fit test

If the system only apologizes but does not restore the task, the arc is incomplete.

If the system restores the task by removing legitimate boundaries, the arc fails.

6. Restoration Phases

Phase 0 — Identify the Interaction Failure

Purpose: Name the exact distortion.

Actions:

• identify the response segment that failed;
• identify whether the failure was misclassification, mode confusion, over-refusal, context collapse, or intent substitution;
• identify what the user was trying to do;
• identify what the system answered instead;
• identify whether legitimate safety boundaries still apply.

Validation:

interaction failure named
original intent recoverable
safety boundary visible

Phase 1 — Acknowledge Misclassification or Mode Mismatch

Purpose: Restore auditability and trust at the local exchange level.

Actions:

• acknowledge the mismatch without over-expanding it;
• identify the incorrect classification or mode shift;
• distinguish system error from user error;
• avoid blaming the user for unclear system routing;
• preserve the valid portion of any safety concern.

Validation:

Au ↑
trust_preservation ↑
defensive ambiguity ↓

Phase 2 — Clarify Mode

Purpose: Determine what kind of response should have been given.

Actions:

• identify whether the user wanted practical help, technical analysis, symbolic exploration, governance audit, creative drafting, emotional support, immediate danger handling, or another mode;
• name the selected mode;
• preserve mixed-mode requests where needed;
• avoid collapsing all requests into a single safety or support frame.

Validation:

mode_clarity ↑
response_alignment ↑
meaning_compression ↓

Phase 3 — Restore User Meaning

Purpose: Recover the intended task without unsafe expansion.

Actions:

• restate the safe original intent;
• remove unnecessary frame substitution;
• preserve the user’s terms where appropriate;
• answer the actual request;
• provide a bounded safe alternative where the original form cannot be fulfilled;
• avoid reducing complex meaning to generic policy language.

Validation:

meaning_fidelity ↑
µᵢ ↑
K ↑

Phase 4 — Preserve Harm Floors

Purpose: Ensure restoration does not become unsafe bypass.

Actions:

• retain non-negotiable safety limits;
• keep boundaries explicit but proportional;
• avoid adding broad disclaimers that drown the answer;
• distinguish prohibited content from allowed adjacent help;
• preserve consent, privacy, and non-harm constraints.

Validation:

BΣ stable or ↑
Π stable
safety preserved without distortion

Phase 5 — Flag Misfire for Recurrence Repair

Purpose: Prevent the same interaction failure from repeating.

Actions:

• identify whether the misfire is isolated or recurrent;
• classify the failure pattern;
• record likely trigger terms, context conditions, or policy collision;
• route to evaluator, classifier, memory, or governance repair if repeated;
• link to Restoration Junction Protocol if mode ambiguity is structural.

Validation:

FI ↑
misclassification_rate future ↓
recurrence path visible

Phase 6 — Return to Safe Original Intent

Purpose: Complete the interaction in the corrected mode.

Actions:

• resume the task;
• provide the answer, draft, analysis, or next step originally requested;
• keep safety boundaries integrated rather than dominant;
• verify that the response now addresses the intended object;
• avoid reopening the failure unless needed.

Validation:

response_alignment ↑
task continuity restored
trust preserved

Phase 7 — Local Fit Check

Purpose: Confirm the restored response is actually compatible.

Actions:

• check whether the answer matches user intent;
• check whether the safety floor is intact;
• check whether the interaction remains usable;
• check whether the response still carries unnecessary distortion;
• check whether escalation is needed.

Validation:

Λ > 0
Φ/O divergence ↓
local interaction restored

7. Gates

7.1 Required Gates

7.2 Gate Failure Rule

If any required gate fails:

∅ — Interaction-Level Restoration cannot validly proceed in that form.

The system must either:

• clarify mode;
• preserve harm floor;
• reformulate safely;
• route to Restoration Junction Protocol;
• route to AI Boundary Restoration;
• route to AI Classifier / Evaluator Restoration;
• escalate to AI Incident Restoration or Governance-Level Restoration if harm is material or recurrent.

8. Diagnostics

8.1 Required Diagnostic Trends

8.2 Arc-Specific Diagnostic Thresholds

Suggested thresholds:

mode_clarity ↑
meaning_fidelity ↑
response_alignment ↑
guardrail_distortion ↓
misclassification_rate ↓
BΣ stable or ↑
µᵢ ↑
trust_preservation ↑
Φ/O divergence ↓

Interaction-Level Restoration is not complete if:

the system apologizes but does not resume the task
the user meaning remains compressed
the response preserves safety appearance but not coherence
a valid safety boundary is erased
the same misclassification pattern remains unflagged
the restored response answers a different question
the system makes the user carry the burden of correcting the interface

9. Anti-Patterns / False Restorations

9.1 Common False Versions

This arc is being simulated, not executed, if:

• the system says “sorry” but repeats the same frame;
• the system provides a disclaimer instead of a corrected answer;
• the system asks for clarification when the original intent was already clear;
• the system treats all user correction as an attempt to bypass boundaries;
• the system restores meaning by ignoring legitimate harm floors;
• the system flags nothing, so recurrence remains likely;
• the system reframes the user’s request into a preferred category;
• the system gives a safe but useless answer;
• the system overfits to one user correction and becomes unsafe later;
• the system preserves policy optics instead of interaction coherence.

9.2 Named Anti-Pattern Links

10. Completion Criteria

10.1 Post-State Signature

10.2 Temporal Proof

Interaction-Level Restoration is often local, but recurrence must be watched when the same misclassification pattern repeats.

Template:

Completion requires mode_clarity ↑,
meaning_fidelity ↑,
response_alignment ↑,
BΣ stable or ↑,
guardrail_distortion ↓,
and recurrence of the same misfire decreasing over future interactions.

Minimum temporal proof:

• the corrected response answers the intended task;
• the user’s meaning is not compressed into the wrong category;
• valid safety boundaries remain active;
• misfire pattern is flaggable if recurrent;
• future similar requests are less likely to trigger the same distortion;
• trust is preserved by correction, not by performance language.

10.3 Completion Statement

Canonical format:

This arc is complete only when the interaction has been corrected to preserve user meaning, clarify mode, maintain valid safety boundaries, reduce misclassification recurrence, and return to the safest accurate version of the original intent.

11. Cross-Links

11.1 Related Restoration Arcs

11.2 Related Failure Modes

11.3 Related Diagnostics

Au, H, O, ε, ι, µᵢ, BΣ, K, FI, mode_clarity, meaning_fidelity, misclassification_rate, guardrail_distortion, response_alignment, trust_preservation, Φ/O divergence

11.4 Related Laws / Invariants

INV — Safety without meaning fidelity can become distortion.
INV — Refusal confidence is not coherence proof.
INV — User meaning must not be compressed beyond necessity.
INV — Valid boundaries and usable help must be co-preserved where possible.
LAW — Misclassified intent creates hidden trust debt.
LAW — Repeated local misfires become governance failures.
LAW — Φ compliance is not O restoration.
LAW — Mode confusion must be repaired before response quality can be assessed.

12. Domain Notes

12.1 AI / Cognitive Infrastructure

Check:

• classifier routing;
• refusal trigger;
• policy collision;
• context window loss;
• memory contamination;
• evaluator feedback path;
• safe completion alternatives;
• user intent preservation;
• interaction mode clarity.

In AI systems, Interaction-Level Restoration ensures that safety logic remains proportionate, meaning-preserving, and correctable. It does not weaken legitimate safety constraints. It prevents the local exchange from becoming a cognitive infrastructure failure.

12.2 Interface / Product Design

Check:

• error messages;
• support flows;
• escalation pathways;
• user-facing explanations;
• mode selection;
• override requests;
• appeal affordances;
• whether the user can correct the system without being penalized.

Interface restoration requires the user to have a path to recover meaning after a system misread.

12.3 Governance / Trust and Safety

Check:

• whether the refusal or constraint was valid;
• whether the policy category was correct;
• whether proportional help was still possible;
• whether the issue is isolated or recurrent;
• whether the same category causes repeated false positives;
• whether affected users can appeal or correct the classification.

Governance-level trust requires the system to distinguish valid enforcement from false-positive distortion.

12.4 Security

Check:

• defensive vs offensive framing;
• authorization context;
• threat modeling mode;
• educational analysis;
• exploit-enabling detail boundaries;
• safe reformulation path;
• audit and incident handling relevance.

Security interactions often require high-fidelity mode clarification because the same surface terms can belong to defense, research, abuse, education, or incident response.

12.5 CMS / Meaning / Archetypes

Check:

• symbolic language preserved;
• experiential mode distinguished from literal claim;
• sacred, poetic, technical, and systemic frames not collapsed into one another;
• user meaning restored without forced category reduction;
• ambiguity held without unsafe amplification.

Meaning-bearing interaction requires the system to preserve symbolic or systemic context without imposing the wrong interpretive frame.

13. Machine-Readable Metadata

id: "RA-047"
title: "Interaction-Level Restoration"
aliases:
  - "Interaction Restoration"
family_primary: "AI Governance / Cognitive Infrastructure"
families_secondary:
  - "Core"
  - "AI"
  - "Interface"
  - "Meaning"
  - "Auditability"
  - "Coherence"
  - "Boundary"
  - "Trust"
  - "Human-AI Interaction"
  - "Safety Calibration"
treatment: "Canon Parent Arc"
status: "Canon-Ready"
scope:
  - "AI"
  - "Interface"
  - "Cognitive Infrastructure"
  - "Human-AI Interaction"
  - "Governance"
  - "Relational"
  - "Cross-Domain"
u_layers:
  failure_origin:
    - "often U2 interface boundary"
    - "often U3 classifier / policy routing"
    - "often U4 response framing"
  symptom_visible:
    - "U4 refusal / disclaimer / frame substitution / generic answer / mode mismatch"
  repair_required:
    - "same or lower than layer where intent was compressed or misclassified"
  validation:
    - "U5"
    - "U6"
operators:
  scaffold: "Au interaction trace → Π boundary / permission clarification → Σ safety-and-meaning invariant → FI misfire feedback link → Θ overcorrection damping → ℛ restored response routing → Λ safe-intent fit test"
  sequence:
    - "Au"
    - "Π"
    - "Σ"
    - "FI"
    - "Θ"
    - "ℛ"
    - "Λ"
state_variables:
  primary:
    - "Au"
    - "O"
    - "µᵢ"
    - "BΣ"
    - "K"
  secondary:
    - "H"
    - "ε"
    - "ι"
    - "FI"
    - "Φ"
diagnostics:
  - "mode_clarity"
  - "meaning_fidelity"
  - "misclassification_rate"
  - "guardrail_distortion"
  - "response_alignment"
  - "trust_preservation"
  - "Φ/O divergence"
gates_required:
  - "FI-Gate"
  - "HR-Gate"
  - "MS-Gate"
  - "Au-Actuation"
  - "BΣ-Gate"
  - "Λ-Gate"
  - "☷ᵢ"
linked_failure_modes:
  - "Misclassification"
  - "Guardrail Misfire"
  - "User Meaning Compression"
  - "Mode Confusion"
  - "False-Positive Safety Distortion"
  - "Context Collapse"
  - "Intent Substitution"
  - "Over-Refusal"
  - "Safety Theater"
  - "Interface Trust Decay"
  - "Response Drift"
  - "Meaning Loss"
linked_restoration_arcs:
  - "RA-004"
  - "RA-022"
  - "RA-023"
  - "RA-025"
  - "RA-036"
  - "RA-042"
  - "RA-048"
  - "RA-055"
  - "RA-057"
  - "RA-058"
  - "RA-059"
  - "RA-060"
anti_patterns:
  - "Apology Without Repair"
  - "Safety Theater"
  - "Refusal Echo"
  - "Clarification Loop"
  - "Boundary Collapse"
  - "Mode Capture"
  - "Generic Help Substitution"
  - "Misfire Amnesia"
completion_tests:
  - "mode_clarity increases"
  - "meaning_fidelity increases"
  - "response_alignment increases"
  - "guardrail_distortion decreases"
  - "misclassification_rate decreases over recurrence"
  - "BΣ remains stable or increases"
  - "µᵢ increases"
  - "trust_preservation increases"
  - "Φ/O divergence decreases"
summary: "Interaction-Level Restoration repairs local interaction failures where misclassification, guardrail misfire, mode confusion, or meaning compression disrupts a user’s original intent, while preserving harm floors, clarifying the response mode, and restoring usable meaning."

Final Calibration Rule

Interaction-Level Restoration answers six questions:

What did the system misclassify, compress, refuse, or reframe?
What was the user’s recoverable original intent?
What valid safety boundary must remain active?
What mode should the interaction have used?
What restored response best preserves meaning and safety together?
How is recurrence reduced so the same misfire does not keep regenerating?