AI Release Readiness Checklist

A practical, risk-tiered checklist framework for evaluating AI release readiness, with a packaged CLI validator for local use and CI pipelines.

Choose this repo when

Use this repository when you need a working validator for YAML-based release-readiness configurations.

This repo is intentionally narrower than:

• release-governance, which explains the broader release lifecycle and gate model
• governance-playbook, which covers the wider AI operating model
• regulated-ai, which is a starter template repo

What this repository provides

• a packaged release-checklist CLI for validating YAML-based release gate configurations
• starter templates generated with release-checklist init
• example configurations for medium-risk and high-risk AI systems
• typed validation for known metadata fields, boolean gates, bounded numeric values, and expected mapping shapes
• text, JSON, and Markdown reporting for local use and CI pipelines
• GitHub Actions CI covering supported Python versions and packaged CLI behavior

How it works

Three risk tiers are supported, chosen based on safety impact, regulatory exposure, and reversibility:

Tier	Use when
Low risk	Internal tools, no safety impact, easily reversible
Medium risk	Customer-facing, some regulatory context, limited fallback
High risk	Safety-critical, regulated environment, hard to reverse

Higher tiers inherit the required gates from lower tiers and add stricter requirements.

The validator expects a nested YAML structure with these top-level sections:

• metadata
• model_validation
• governance
• infrastructure
• optional but supported: incident_readiness

Known nested sections such as model_validation.performance, governance.approvals, and infrastructure.testing are expected to be mappings rather than free-form lists or strings.

Quick start

git clone https://github.com/simaba/release-checklist.git
cd release-checklist
python -m pip install -e .

Validate a working example configuration:

release-checklist validate configs/medium-risk-example.yaml

Generate a report:

release-checklist report configs/medium-risk-example.yaml --format markdown

Create a starter template:

release-checklist init --industry healthcare

Legacy direct execution is still supported for local source checkouts:

python src/check_release.py validate configs/medium-risk-example.yaml

Install development dependencies:

python -m pip install -e ".[dev]"

Example configuration shape

metadata:
  project: "IVI assistant"
  version: "1.0.0"
  environment: "staging"
  regulated_industry: "general"
  risk_classification: "medium"

model_validation:
  performance:
    accuracy_threshold: 0.90
    bias_evaluation_complete: true

governance:
  documentation:
    risk_assessment_complete: true
  approvals:
    technical_review: true

infrastructure:
  testing:
    unit_tests_passing: true
  rollback:
    rollback_plan_documented: true

Validation behavior

The validator currently enforces:

• required top-level sections
• required metadata fields
• allow-listed values for environment, industry, and risk tier
• semver-like version formatting such as 1.0.0
• mapping/object shape checks for known structural sections
• boolean typing for known gates
• bounded numeric validation for known fields such as accuracy_threshold
• positive numeric validation for known monitoring fields such as latency_ms

This repository is meant to be useful in real workflows, but it is still a lightweight validator rather than a full policy engine.

Repository structure

configs/
  medium-risk-example.yaml
  high-risk-example.yaml
src/
  airc/
    cli.py
    validator.py
    report.py
    templates.py
  check_release.py
tests/
  test_validator.py
requirements.txt
pyproject.toml
.github/workflows/ci.yml

Related repositories

Repository	What it adds
release-governance	Broader framework this checklist operationalizes
governance-playbook	End-to-end operating model
regulated-ai	Starter template repo with governance artifacts

---

Shared in a personal capacity. Open to collaborations and feedback via LinkedIn or Medium.