Skip to main content

Judge Persona Loader for MoE Verification Layer.

Loads and manages judge personas from config/judge-personas.json for the Constitutional Court architecture (H.3.1.7).

Features:

  • Load personas by ID or get all enabled personas
  • Generate prompt templates with artifact/context injection
  • Validate persona schema completeness
  • Model routing configuration (configurable, not hard-coded)
  • Environment variable overrides for models
  • Diversity verification across model families

Model Configuration Priority (highest to lowest):

  1. Runtime override via get_model_for_persona(override_model=...)
  2. Environment variables: CODITECT_JUDGE_MODEL_<PERSONA_ID>
  3. Model routing config file: config/judge-model-routing.json
  4. Persona-specific config in judge-personas.json

File: persona_loader.py

Classes​

Verdict​

Unified verdict types for consensus.

StrictnessLevel​

Persona strictness levels.

ModelRouting​

Model routing configuration for a persona.

EvaluationDimension​

A single evaluation dimension with scoring.

RedFlag​

A red flag pattern to identify during evaluation.

Demographics​

Persona demographic information.

EvaluationStyle​

Evaluation style configuration.

JudgePersona​

Complete judge persona with all attributes.

PersonaConfig​

Global persona configuration.

PersonaRegistry​

Registry of all loaded personas with configuration.

Functions​

get_default_loader()​

Get or create the default persona loader.

load_persona(persona_id)​

Load a persona by ID using the default loader.

get_prompt_template(persona, artifact, context)​

Generate prompt template using the default loader.

validate_persona_schema(persona)​

Validate a persona schema using the default loader.

verify_panel_diversity(persona_ids)​

Verify panel diversity using the default loader.

get_persona(persona_id)​

Get a persona by ID.

get_all_personas(enabled_only)​

Get all personas, optionally filtered by enabled status.

get_personas_for_artifact(artifact_tags)​

Get personas relevant for a given artifact based on trigger conditions.

get_prompt_template(persona, artifact, context)​

Generate the evaluation prompt for a persona with artifact injection.

validate_persona_schema(persona)​

Validate that a persona has all required fields.

validate_all_personas()​

Validate all loaded personas.

verify_panel_diversity(persona_ids)​

Verify that a panel of personas meets diversity requirements.

get_model_for_persona(persona_id, use_backup, override_model)​

Get the model to use for a persona with configuration priority.

get_all_model_mappings()​

Get all persona-to-model mappings with resolved priorities.

map_verdict_to_consensus(persona_id, verdict)​

Map a persona-specific verdict to unified consensus verdict.