ADRI Architecture Guide

Simple, clear explanations of how ADRI works and why each piece matters. This page is intended for maintainers and contributors; user-facing documentation lives under the Users section.

ADRI in 30 Seconds

• The Problem: AI agents break when you feed them bad data.
• The Solution: One decorator that checks data quality before your function runs.
• The Result: Prevent failures and ship reliable agents faster.

@adri_protected  # This line prevents most AI agent failures
def your_agent_function(data):
    return expensive_ai_call(data)  # Now protected from bad data

How ADRI Works (The Flow)

Five Dimensions

• Validity · Completeness · Consistency · Plausibility · Freshness

Components Overview

Guard Decorator (src/adri/decorator.py)

• Intercepts function calls and enforces on_failure: raise | warn | continue
• Visible, explicit configuration (standard, data_param, min_score, dimensions)

Validator Engine (src/adri/validator/)

• Loads data, applies rules, returns AssessmentResult with dimension scores
• Multi-format loading (CSV/JSON/Parquet), detailed failure reporting

Protection Modes (src/adri/guard/modes.py)

• FailFastMode, SelectiveMode, WarnOnlyMode orchestrated by DataProtectionEngine

Standards System (src/adri/standards/)

• YAML rule loading via StandardsParser; schema.yaml drives structure

Analysis (src/adri/analysis/)

• Profiling and generation of YAML standards from sample “good” datasets

Configuration (src/adri/config/)

• Creates ADRI/<env> paths, loads overrides, sane defaults

Logging (src/adri/logging/)

• Local CSV/JSON logs and Enterprise streaming (Verodat MCP)

CLI (src/adri/cli.py)

• setup, generate-standard, assess, list/show/validate-standard, show-config, list-assessments

Quality and Testing

We track multi-dimensional quality beyond line coverage:

• Line Coverage · Integration Tests · Error Handling · Performance

Quality gates for release ensure critical components are robust.

Developer Notes

• Keep user-facing docs in Users section.
• Keep contributor docs in Contributors; link from README to this page.
• Prefer Mermaid diagrams for reviewable architecture visuals.