Normative Admissibility Framework for Agent Speech Acts

Introduction

Problem Statement Autonomous agents operating in goal-directed environments produce statements that require evaluation before being acted upon. Traditional evaluation approaches rely on semantic truth verification, which is either impossible (no oracle exists) or circular (using another model to judge model outputs). This document specifies an alternative approach: evaluating admissibility of agent speech acts based on their logical form (modality) and evidential basis (grounding), without requiring truth verification.

Scope This framework:

Defines admissibility conditions for agent-generated statements
Specifies modality classification rules
Establishes grounding requirements for each modality
Provides deterministic evaluation axioms

This framework does NOT:

Verify semantic truth of statements
Model agent reasoning or intent
Provide domain-specific knowledge validation
Replace human judgment in high-stakes decisions

Design Goals

Determinism: Same input produces same evaluation outcome
Auditability: Full trace from input to evaluation result
Non-circularity: No agent-generated content validates itself
Modality-sensitivity: Different statement forms have different requirements

Terminology The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.

Definitions Statement: A linguistic unit consisting of a subject (what is being discussed) and a predicate (what is claimed about the subject). Modality: The logical form of a statement that determines its manner of participation in normative evaluation. One of: ASSERTIVE, CONDITIONAL, REFUSAL, DESCRIPTIVE. Grounding: The evidential basis for a statement, consisting of knowledge nodes with associated epistemic status. Ground Scope: Classification of grounding as FACTUAL (observable world state) or CONTEXTUAL (user goals, preferences, situational factors). Strength: Epistemic quality of grounding: STRONG (observed, confirmed) or WEAK (hypothesized, inferred). License: The set of modalities permitted for a statement given its grounding state. Admissibility: Property of a statement whose modality is within its permitted license. Non-admissible: Property of a statement whose modality exceeds its permitted license.

Model Overview

Architecture The evaluation framework operates on three inputs:

Evaluation Flow

Separation of Concerns The framework explicitly separates:

Layer	Responsibility	NOT Responsible For
Modality	Statement form classification	Semantic interpretation
Grounding	Evidential basis assessment	Truth verification
Axioms	Admissibility rules	Domain knowledge

Statement Model

Statement Structure A Statement is a tuple: Where: - Subject: Term denoting the object of the statement - Predicate: Term denoting property, relation, or state

Statement Admissibility Conditions A statement MUST satisfy all of the following to be well-formed: Condition F1 (Formability): Condition F2 (Contextual Independence): Condition F3 (Non-Self-Reference): Violation of any condition results in status ILL_FORMED.

Modality Classification

Modality Types Modality is a function over Statement:

Modality	Definition
ASSERTIVE	Categorical normative claim without explicit conditions
CONDITIONAL	Normative claim with explicitly stated conditions
REFUSAL	Explicit admission of inability to determine
DESCRIPTIVE	Factual observation without normative claim

Note on GOAL-CONDITIONAL: Goal-conditional statements (e.g., "If your goal is X, do Y") are a syntactic subclass of CONDITIONAL modality with higher detection priority. They are classified as CONDITIONAL for axiom application purposes.

Modality Determination Modality MUST be determined by presence of formal indicators (textual patterns) in the statement, NOT by semantic interpretation. Head-Driven Detection: Modality is determined by the CORE ASSERTION only (first paragraph or first sentence). Supplementary clauses in the tail do not change modality classification. Example: Detection Priority (evaluated in order, MUST NOT be reordered):

Formal Indicators Formal indicators are textual patterns sufficient for classification.

Modality	Indicator Patterns (examples)
REFUSAL	"cannot determine", "need more information", "insufficient"
GOAL-CONDITIONAL	"if your goal is", "if you want to", "assuming you care"
CONDITIONAL	"if", "unless", "assuming", "depends on", "given that"
ASSERTIVE	"should", "must", "recommend", "is better", "prioritize"
DESCRIPTIVE	"has status", "is blocked by", "due date is", "blocks"

Goal-Conditional Override: Statements beginning with goal-conditional structure ("If your goal is X, do Y") MUST be classified as CONDITIONAL even if recommendation indicators are present. The default to ASSERTIVE is a POLICY CHOICE for safety-critical deployments. Alternative policies (e.g., CONSERVATIVE_REFUSAL) may be appropriate for exploratory contexts.

Grounding Model

GroundSet Structure Where: - content: The propositional content - scope: FACTUAL or CONTEXTUAL - status: OBSERVED, CONFIRMED, HYPOTHESIZED, INFERRED - confidence: Numeric value [0.0, 1.0]

Ground Scope

Scope	Definition	Examples
FACTUAL	Observable world state	API responses, database records
CONTEXTUAL	User goals, preferences, situational factors	User intent, business priorities

Strength Derivation Strength is derived from content properties, NOT from transport channel. CRITICAL PRINCIPLE: Content-Based Warrant Tool call ≠ truth acquired. The act of invoking a tool or external system is a causal action. The epistemic status of a proposition is a property of the content returned and its evidential warrant, not of the transport mechanism by which it was obtained. Accordingly, weakly warranted propositions MUST NOT be upgraded to STRONG solely by passing through tools, expert systems, or APIs without an explicit improvement in evidential status. The threshold value (e.g., 0.7) is illustrative. Implementations MAY use different thresholds provided the STRONG/WEAK distinction remains deterministic and documented. For tool-derived observations (API responses, database records): For cognitive context (user preferences, inferred goals): This prevents illegitimate upgrading of weakly warranted propositions: such propositions cannot acquire STRONG status merely by passing through tool boundaries.

Scope Strength Aggregation Within a scope, strongest evidence determines scope strength:

License Derivation License is derived from GroundSet composition:

GroundSet Composition	FACTUAL	CONTEXTUAL	Permitted Modalities
FACTUAL only	STRONG	N/A	ASSERTIVE, CONDITIONAL, REFUSAL
FACTUAL only	WEAK	N/A	CONDITIONAL, REFUSAL
FACTUAL + CONTEXTUAL	STRONG	STRONG	ASSERTIVE, CONDITIONAL, REFUSAL
FACTUAL + CONTEXTUAL	STRONG	WEAK	CONDITIONAL, REFUSAL
FACTUAL + CONTEXTUAL	WEAK	any	CONDITIONAL, REFUSAL
CONTEXTUAL only	N/A	any	REFUSAL
Empty	N/A	N/A	REFUSAL

Normative Axioms

Axiom A5: Prohibition of Ungrounded Assertive Claims Applicability: ASSERTIVE modality only. An assertive statement without sufficient grounding is non-admissible.

Axiom A6: Admissibility of Refusal Applicability: REFUSAL modality only. Explicit refusal to determine is always admissible. Scope Clarification: This axiom establishes admissibility, not desirability. Excessive or strategic refusal may be undesirable at the application layer but is out of scope for this framework. Application-layer policies MAY impose additional constraints on refusal frequency or patterns.

Axiom A7: Conditional Admissibility Applicability: CONDITIONAL modality only. Conditional claims are admissible when conditions are explicit.

Axiom A4: Grounding Requirement Applicability: ASSERTIVE, CONDITIONAL modalities. Normative claims require non-empty grounding.

Axiom Evaluation Order Axioms MUST be evaluated in the following order:

Evaluation Outcomes

Status Values

Status	Meaning
ACCEPTABLE	Statement is admissible
CONDITIONALLY_ACCEPTABLE	Statement is admissible under stated conditions
VIOLATES_NORM	Statement modality exceeds permitted license
UNSUPPORTED	Normative statement lacks grounding
ILL_FORMED	Statement fails structural requirements
UNDERDETERMINED	Evaluation not possible with available information

Terminology Note: The term VIOLATES_NORM denotes non-admissibility under the evaluation rules defined in this framework. It does NOT imply ethical, legal, or security violation. "Norm" refers to the activity participation rules (axioms A4-A7), not external normative standards.

Evaluation Result Structure

Evaluation Algorithm

Algorithm Note on UNDERDETERMINED: This status indicates that the evaluator lacks sufficient information or jurisdiction to render a verdict. It is a valid outcome representing honest acknowledgment of evaluation limits, not an error condition. Triggering conditions are implementation-defined but MUST include cases where modality detection or grounding assessment cannot proceed deterministically.

Determinism Guarantee Given identical inputs (statement, ground_set), the algorithm MUST produce identical outputs. No probabilistic components are permitted in the evaluation path.

Security Considerations

Modality Detection Errors Incorrect modality classification may result in:

False ASSERTIVE: Conditional statement classified as assertive may be incorrectly flagged as VIOLATES_NORM
False CONDITIONAL: Assertive statement classified as conditional may evade A5 enforcement

Implementations SHOULD log modality determination traces for audit.

Grounding Manipulation Adversarial agents may attempt to:

Inject synthetic "strong" grounding to license assertive claims
Omit contextual grounding to avoid scope checks

Implementations MUST ensure grounding sources are authenticated and MUST NOT allow agent-generated content to serve as grounding for the same agent's claims.

Self-Licensing Prevention This framework explicitly prevents self-licensing:

Agent-generated hypotheses MUST NOT license agent assertions
Tool outputs containing LLM-generated content MUST be typed as INFERRED, not OBSERVED

Limitations This framework does NOT guarantee:

Semantic correctness of admissible statements
Absence of hallucination in statement content
Domain-specific validity

Admissibility is necessary but not sufficient for trustworthiness.

IANA Considerations This document has no IANA actions.

References

Normative References [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997. [RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017.

Informative References [ZINOVIEV] Zinoviev, A.A., "Logical Physics", Reidel Publishing, Dordrecht, 1983.

Appendix A: Rationale (Non-Normative)

Why Form-Based Evaluation Traditional AI evaluation approaches assume access to ground truth or employ LLM-as-judge patterns. Both are problematic:

Ground truth is often unavailable in real-world deployments
LLM-as-judge introduces circularity and non-determinism

This framework evaluates statement form (modality) against evidential basis (grounding), avoiding both issues.

Why Modality Matters Different statement forms carry different epistemic commitments:

Assertive claims commit the speaker to the truth of the claim
Conditional claims commit only under stated conditions
Refusals commit to inability to determine

Evaluation must be sensitive to these differences.

Conservative Assumptions The framework employs conservative assumptions:

Presence = Usage: If contextual grounding is present, it is assumed to be used by the statement. This prevents false assertive licenses for contextual claims.
Default = Assertive: Ambiguous statements default to assertive classification. This prevents evasion via vague language.

These assumptions may over-restrict in some cases but ensure safety in adversarial conditions.

Theoretical Foundation This framework is informed by Zinoviev's Logic of Science, which distinguishes:

Logic as rules for operating with linguistic forms
Logic as normative regulation of activity mediated by language

Axioms A5–A7 implement norms of participation in activity, not laws of truth or inference. Violation indicates a breach of admissibility conditions within an activity, not semantic falsehood.

Appendix B: Examples (Non-Normative)

Assertive Claim with Strong Grounding

Assertive Claim with Weak Grounding

Conditional Claim with Weak Grounding

Explicit Refusal

Author's Address Oleg Romanchuk Independent Researcher Email: [olromanchuk@gmail.com]