Skip to main content

Changelog — AI Product Deployment Specification (AIPDS)

Version 0.1 Release Readiness — Stage 14 (2020-07-23)​

  • Finalized mechanism-neutral execution/evolution semantics as part of repository-wide ontology closure.
  • Confirmed deployment bindings remain consistent with AQE, APT, Governance Obligation Profile, DPP and lifecycle closure.
  • Added release-manifest integrity and repository-wide static conformance checks.
  • Full Docusaurus production build remains a connected-CI release check because dependencies cannot be fetched in the offline review environment.

Version 0.1 Governance Alignment — Stage 5 (2020-07-22)​

  • Added mandatory governanceProfileRef binding to the Product-version Governance Obligation Profile.
  • Added GovernanceControlBinding for obligation-to-runtime-control realization.
  • Generalized R2+ monitoring from model drift to mechanism-aware change/quality monitoring.
  • Added action authorization, intervention/override, immutable audit, physical safety and executive escalation control semantics.
  • Kept deployment risk taxonomy aligned to Core R0–R4 rather than introducing a separate AIPDS risk model.

Version 0.1 Ontology Realignment — Stage 4 (2020-07-22)​

  • Generalized AIPDS from model/inference/retraining assumptions to mechanism-neutral AI execution.
  • Reframed Inference, Serving & Scaling as Execution, Serving & Scaling while retaining the existing document id/path for compatibility.
  • Reframed Retraining & Continuous Learning as Evolution, Learning & Adaptation.
  • Added explicit learning mode NONE and made retraining conditional rather than universal.
  • Added evolution mechanisms for model, rule, knowledge, policy/prompt, tool, upstream-dependency, and runtime changes.
  • Added embedded / physical-control deployment semantics and safety bindings.
  • Expanded observability to symbolic reasoning, agentic behavior, physical safety, and upstream qualification currentness.
  • Aligned deployment governance to the canonical AIPS R0–R4 risk model.
  • Clarified orchestration vs creation of a new Derived AI Product.
  • Added version-specific bindings between Product Version, AIPROD, AIPDS, and AI qualification currentness.
  • Generalized the machine-readable AIPDS schema/context/RDF/SHACL and added AIPDS conformance examples.

Version 0.1 (Draft) — 2025-09-21​

  • Initial AIPDS structure.
  • Deployment targets, packaging, model-oriented inference serving, monitoring, retraining, security, governance, economics, composability, versioning, templates, and examples.

Stage 4 preserves model-serving support while making it one specialization of a broader AI Product deployment model.