Normative Coherence Metric

The metric directly measures the ratio of conflicting rule pairs to the total number of possible rule interactions within a normative corpus. A lower density score indicates a more coherent system. This involves parsing the rule base to identify direct logical collisions, such as an obligation-obligation conflict where a single action is simultaneously mandated and prohibited. The calculation often relies on a Normative Collision Matrix to systematically enumerate and classify each pairwise interaction, providing a raw count of inconsistencies that the AI model must resolve.

During model training, the coherence metric is integrated as a custom loss component. The model is penalized not just for factual inaccuracy but for generating logically inconsistent rule interpretations. This is formalized as:

• Logical Consistency Loss: A penalty applied when the model's output violates deontic logic constraints.
• Entailment Fidelity: A measure of whether the model's conclusions are valid entailments from a conflict-free subset of rules. This steers the model toward solutions that respect the Maximal Consistent Subset (MCS) of the governing legal framework.

A core component of the metric evaluates the alignment of deontic modalities (obligation, permission, prohibition) across the reasoning chain. The metric checks for modality clashes by verifying:

• Obligation Consistency: No two rules simultaneously obligate and prohibit the same action for the same agent.
• Permissive Harmony: A permitted action is not simultaneously prohibited by a higher-priority rule. This relies on a formal Deontic Logic Tensor to represent the truth values of each modality, allowing for vector-space calculations of logical coherence.

The metric weights conflicts based on the precedence hierarchy of the colliding rules. A conflict between a constitutional norm and a regulation is scored as more severe than a conflict between two regulations of equal weight. The scoring algorithm traverses a Normative Hierarchy Graph, applying the principles of:

• Lex Superior Derogat Inferiori: Higher authority rules override lower ones.
• Lex Specialis Derogat Legi Generali: Specific rules override general ones. The final coherence score is adjusted by the severity of unresolved conflicts after precedence is applied.

The metric incorporates a temporal reasoning module to ensure that the sequence of rule enactments and repeals does not create logical gaps. It verifies that:

• Lex Posterior is correctly applied, where a later rule overrides an earlier conflicting one.
• Rule Suspension periods do not create undefined normative states.
• Contrary-to-Duty Obligations are properly modeled, ensuring that a secondary obligation triggered by a violation does not itself conflict with the primary rule system. This prevents the metric from rewarding models that ignore temporal dynamics.

The metric can also evaluate the minimality of a repair needed to restore coherence. It measures the semantic distance between an inconsistent rule base and its coherent successor after a Normative Repair Operator is applied. Key indicators include:

• Rule Modification Count: The number of rules that required weakening or exception clauses.
• Norm Abrogation Impact: The scope of rules that had to be permanently removed. A high-efficiency repair indicates the original system was close to coherent, while a low-efficiency score signals deep, systemic contradictions requiring extensive restructuring.

A fundamental legal principle dictating that a specific rule overrides a general rule when both apply. This is the primary mechanism for normative exception handling. In a computational context, this principle is implemented by assigning a higher specificity weight to rules with more detailed applicability conditions. For example, a statute governing 'commercial drones under 250g' will preempt a general aviation law when both are triggered. The Normative Coherence Metric penalizes systems that fail to correctly apply this principle, measuring the consistency of exception carving.

The hierarchical conflict rule stating that a law from a higher authority (e.g., a constitution) overrides a conflicting law from a lower authority (e.g., a statute). This is modeled in a Normative Hierarchy Graph where nodes represent rules and directed edges represent 'has-authority-over' relationships. The Normative Coherence Metric evaluates whether a reasoning system correctly resolves conflicts by traversing this graph, ensuring a federal regulation always defeats a conflicting municipal ordinance.

A temporal precedence maxim where a later-enacted law overrides an earlier one in case of irreconcilable conflict. This requires a system to maintain a temporal validity index for every rule. The Normative Coherence Metric checks for correct temporal reasoning, penalizing a model that incorrectly applies an obsolete regulation when a newer, conflicting version is active. This is critical for regulatory change detection and maintaining a coherent rule base over time.

A computational method for resolving global inconsistency by identifying the largest subset of non-contradictory rules. When a rule base is incoherent, an MCS algorithm finds a conflict-free core. The Normative Coherence Metric can be defined as the ratio of the size of the chosen MCS to the total rule base size. For instance, if a system must drop 3 out of 50 rules to achieve consistency, its coherence score is 94%. This provides a direct, quantitative measure of a system's internal logical integrity.

A formal logic system where adding a new premise can invalidate a previous conclusion. This is essential for legal reasoning because a new exception or overriding rule must retract a prior inference. The Normative Coherence Metric is inherently non-monotonic; the addition of a new, contradictory rule should decrease the metric until the conflict is resolved. This contrasts with classical logic, where adding axioms never shrinks the set of provable theorems.

An algorithmic function that minimally modifies an inconsistent rule set to restore coherence. Operations include:

• Rule Weakening: Narrowing a rule's scope to avoid collision.
• Exception Carving: Explicitly adding a lex specialis exception to a general rule.
• Rule Abrogation: Removing the lowest-priority conflicting rule. The Normative Coherence Metric serves as the loss function for training AI models to act as optimal normative repair operators, guiding them to make the smallest possible change to maximize consistency.