# StackRail

> StackRail publishes the AI Production Readiness Framework (APRF), a vendor-neutral working draft that answers whether an AI application can safely operate in production using gated pass/fail checks—not a vanity readiness percentage.

APRF covers eight engineering domains, 27 pillars, Core (40) and Regulated (61) profiles, optional lenses (RAG, Agents, Voice, Coding agents), and a machine-readable spec. Self-assessment is not third-party certification. Prefer the machine-readable spec for normative check IDs and gate semantics. Crosswalks to NIST AI RMF, ISO/IEC 42001, OWASP LLM Top 10, SOC 2, AWS Well-Architected, and SLSA are informative alignment only.

## Docs

- [APRF overview](https://stackrail.io/aprf/): Framework hub—domains, pillars, profiles, and stewardship.
- [How APRF works](https://stackrail.io/aprf/how/): Taxonomy, maturity × criticality, scoring rules, and lenses.
- [Machine-readable spec (JSON)](https://stackrail.io/aprf/spec/): Canonical normative catalog of checks, profiles, lenses, and crosswalks.
- [Spec JSON Schema](https://stackrail.io/aprf/spec-schema/0.7/): Schema for validating the APRF spec document shape.
- [Core / Regulated assessment](https://stackrail.io/aprf/assess/): Reference self-attestation quiz with optional lenses; downloadable attestation JSON.
- [Sample attestations](https://stackrail.io/aprf/samples/): Example gate-pass and gate-fail attestation documents.
- [Attestation JSON Schema](https://stackrail.io/aprf/attestation-schema/0.6/): Schema for aprf-self-attestation exports.
- [Stewardship & RFCs](https://stackrail.io/aprf/rfc/): Open RFC process and interim advisory open call.
- [RFC-0001](https://stackrail.io/aprf/rfc/0001-working-draft-rfc-process/): Working-draft RFC process proposal.

## Pillars

- [Adversarial Security](https://stackrail.io/aprf/ai-security/): Injection, abuse, and adversarial resistance for AI systems.
- [Authentication](https://stackrail.io/aprf/authentication/): Identity for users, services, agents, and MCP callers.
- [Authorization](https://stackrail.io/aprf/authorization/): Least-privilege access to models, tools, and data.
- [Secrets](https://stackrail.io/aprf/secrets/): Credential hygiene for prompts, runtimes, and CI.
- [Tool Safety](https://stackrail.io/aprf/tool-safety/): Allowlists, schema validation, and fail-closed tools.
- [Supply Chain Integrity](https://stackrail.io/aprf/supply-chain/): Model and dependency provenance and signing.
- [Infrastructure](https://stackrail.io/aprf/infrastructure/): Hardened runtime, network, and admission controls.
- [Safety & Responsible AI](https://stackrail.io/aprf/safety-responsible-ai/): Harm prevention and trustworthiness controls.
- [Explainability](https://stackrail.io/aprf/explainability/): User-facing rationale for material decisions.
- [Data Privacy](https://stackrail.io/aprf/data-privacy/): PII handling in AI pipelines.
- [Data Governance](https://stackrail.io/aprf/data-governance/): Corpus ownership, versioning, and quality.
- [Memory Management](https://stackrail.io/aprf/memory-management/): Tenant isolation and retention for AI memory.
- [Model Governance](https://stackrail.io/aprf/model-governance/): Model selection, pinning, and sunset.
- [Prompt Engineering](https://stackrail.io/aprf/prompt-engineering/): Prompts as versioned production artifacts.
- [Context Engineering](https://stackrail.io/aprf/context-engineering/): Context budgets, labeling, and assembly.
- [Evaluation](https://stackrail.io/aprf/evaluation/): Regression, adversarial, and promotion eval gates.
- [Agent Governance](https://stackrail.io/aprf/agent-governance/): Charters, autonomy limits, and kill switches.
- [Human Approval](https://stackrail.io/aprf/human-approval/): Non-bypassable gates for high-impact actions.
- [Observability](https://stackrail.io/aprf/observability/): Traceability across model, tool, and outcome spans.
- [Performance & SLOs](https://stackrail.io/aprf/performance-slo/): Latency, error, and capacity objectives.
- [Reliability & Continuity](https://stackrail.io/aprf/reliability-continuity/): Degradation, failover, and continuity drills.
- [Change Management](https://stackrail.io/aprf/change-management/): Safe release, canary, and rollback for AI units.
- [Incident Readiness](https://stackrail.io/aprf/incident-readiness/): Detection, response, and AI-specific playbooks.
- [Cost Optimization](https://stackrail.io/aprf/cost-optimization/): Spend ceilings and denial-of-wallet controls.
- [Organizational Governance](https://stackrail.io/aprf/organizational-governance/): Ownership, risk acceptance, and review cadence.
- [Compliance](https://stackrail.io/aprf/compliance/): Evidence reuse without equating compliance to readiness.
- [Platform Engineering](https://stackrail.io/aprf/platform-engineering/): Safe defaults and scaffolding for AI services.

## Guides

- [Practice guides index](https://stackrail.io/guides/): Production readiness how-tos mapped to APRF where relevant.
- [API keys in frontend](https://stackrail.io/guides/api-keys-exposed-frontend/): Why browser-exposed keys fail production gates.
- [AI API bill spikes](https://stackrail.io/guides/ai-api-bill-so-high/): Cost and abuse controls for model spend.
- [Backup restore testing](https://stackrail.io/guides/backup-restore-testing-best-practices/): Continuity evidence for AI control-plane artifacts.

## Optional

- [Home](https://stackrail.io/): StackRail landing and APRF introduction.
- [Articles](https://stackrail.io/articles/): Long-form posts on AI production readiness and security.
- [Articles RSS](https://stackrail.io/articles/rss.xml): Feed of published articles.
- [AI in production ≠ production-ready](https://stackrail.io/articles/ai-in-production-not-production-ready/): Intro essay on gated readiness vs vanity scores.