Skip to main content

PMDD Assessment Guidance for AIPCH

This section provides detailed guidance for evaluating qualified AI Products against the AI Product Characteristics (AIPCH01–AIPCH21) using the Product Maturity Driven Development (PMDD) model.

PMDD treats maturity as a continuously observable, evidence-backed, and improvable property of a specific AI Product version. It does not determine whether something is a Product or whether it belongs to the AI Product Family.


PMDD Starts After AI Product Qualification​

AIPS separates three questions that must not be collapsed into one score:

  1. Product qualification — does the thing qualify as a Product under BPS?
  2. AI Product qualification — does PIN establish artificial intelligence as the value-bearing essence, with a distinct intelligence proposition and current AI Primitive evidence?
  3. AI Product maturity — how mature is that qualified AI Product against the applicable AIPCH characteristics?

PMDD answers only the third question.

A low PMDD score does not turn an AI Product into an AI Asset. A high PMDD score cannot turn a Model Product, AI-enabled Product, or unqualified artifact into an AI Product.

The normal PMDD precondition is therefore:

  • the assessed Product version has CURRENT AI Qualification Evidence (AQE);
  • the Product version is classified in the AI Product Family;
  • its applicable Governance Obligation Profile is known or can be resolved.

If these preconditions are not met, PMDD is not assessable for that version. The correct response is to resolve Product/AI qualification first, not to assign maturity level L0.


What This Guidance Is​

This guidance is not a checklist to be completed once and is not an AI Product qualification mechanism.

It is:

A structured interpretation layer for consistent, automated, and agent-assisted evaluation of AI Product maturity over time.

Each AIPCH characteristic defines:

  • what it asserts beyond surface interpretation;
  • when it is applicable;
  • when it is met, partially met, or not met;
  • what evidence signals support the assessment;
  • how it contributes to maturity and evolution.

The Role of PMDD​

PMDD:

  • evaluates the applicable AIPCH characteristics continuously;
  • produces maturity scores and profiles for a Product version;
  • detects maturity regressions;
  • identifies improvement priorities;
  • consumes evidence from AIPROD, AIPDS, AIPS-DPP, Marketplace, COVP, observability, and governance;
  • feeds Product evolution without redefining PIN, Product Family, APT, or R0–R4 risk.

Observation is not ontology. Maturity is not qualification.


Maturity Levels​

PMDD uses three maturity bands over the applicable characteristic weight.

L2 — High-Maturity AI Product (≥ 80%)​

  • Product characteristics are broadly mature and evidence-backed.
  • The Product is well positioned for scaled use, subject to its mandatory governance obligations and trust currentness.

L1 — Evolving AI Product (50–79%)​

  • Core Product characteristics exist but remain incomplete or inconsistent.
  • Material hardening or operating-model improvements remain.

L0 — Emerging AI Product (< 50%)​

  • The thing is still a qualified AI Product, but its productization maturity is low.
  • Ownership, interfaces, evidence, trust, economics, reuse, or operational maturity may still be weak.

Not Assessable — Qualification Precondition Failed​

This is not a maturity level. It applies when AI qualification is absent, stale, invalid, or the thing belongs to another Product Family.

Examples include:

  • a raw model checkpoint;
  • a mature Model Product offered by AIModelCo;
  • an AI-enabled Service Product;
  • a previously qualified AI Product version whose qualification is now INVALID.

Maturity Does Not Override Governance​

A maturity percentage is not a deployment authorization.

Mandatory Governance Obligation Profile requirements, trust currentness, policy constraints, and R0–R4 controls remain independently binding.

A Product can score highly in PMDD and still be unsuitable for production promotion if a mandatory governance obligation is unmet. Conversely, a low-maturity AI Product does not cease to be an AI Product merely because it needs improvement.


Applicability-Aware Assessment​

Not every AIPCH characteristic is equally applicable to every AI Product version.

Examples:

  • model retraining is not applicable to a symbolic Product with learningMode = NONE;
  • subgroup fairness measurement may be irrelevant where no people, protected groups, or comparable allocative effects are involved;
  • per-decision human approval may not be required for low-risk autonomous behavior where the Governance Obligation Profile permits it;
  • Physical AI requires controls that a non-embodied Product does not.

Each characteristic is therefore assessed as:

  • Met — full applicable evidence exists;
  • Partial — evidence or implementation is incomplete;
  • Not Met — an applicable requirement is unsatisfied;
  • Not Applicable — the characteristic or part of it does not apply to this Product version and the non-applicability is justified by evidence.

Not Applicable:

  • must include a rationale;
  • should preferably reference the Governance Obligation Profile, ontology dimensions, or Product constitution;
  • contributes neither points nor weight to the normalized score.

N/A is an applicability decision, not a convenient way to avoid a maturity gap.


What Makes AI PMDD Different​

AI Products introduce maturity concerns beyond conventional Product management, but those concerns vary by intelligence mechanism and context.

PMDD therefore considers evidence such as:

  • behavior quality and reliability;
  • AI qualification and dependency currentness;
  • explainability and transparency where applicable;
  • agency, embodiment, and intelligence effects;
  • policy and safety controls;
  • governed evolution and adaptation;
  • composition/dependency behavior;
  • usage, fitness, feedback, and value realization through COVP.

PMDD must not assume that every AI Product:

  • owns a machine-learning model;
  • learns continuously;
  • requires retraining;
  • exhibits model drift;
  • makes a decision;
  • requires human approval for every action.

Continuous vs Static Assessment​

PMDD is intended to be evidence-driven and periodically or event-triggered re-evaluated.

Relevant triggers include:

  • Product version change;
  • AI qualification or dependency currentness change;
  • Governance Obligation Profile change;
  • AIPDS realization change;
  • incident or assurance signal;
  • COVP fitness/value evidence;
  • material consumer or use-context change.

Manual review may contribute evidence, but the resulting maturity state should be structured, version-bound, auditable, and machine-consumable.


Relationship with AIPROD, AIPDS, DPP, Marketplace and COVP​

PMDD draws on the wider AIPS evidence fabric:

  • AQE → confirms that the assessed version currently qualifies as an AI Product;
  • AIPROD → semantic Product definition and ontology classification;
  • AIPDS → execution, control, evolution, and deployment realization;
  • AIPS-DPP → trust, assurance, qualification and governance evidence;
  • Marketplace → current discovery/publication status and consumption contracts;
  • COVP → usage, feedback, contextual fitness and value-realization evidence;
  • Governance Obligation Profile → determines mandatory and context-dependent controls.

Together these form the PMDD evidence surface.


How to Use This Guidance​

For each AIPCH characteristic:

  1. Confirm the assessed AI Product version and AQE are current.
  2. Determine whether the characteristic is applicable to that version and context.
  3. Read “What it is really asserting”.
  4. Inspect positive, negative, and edge-case evidence.
  5. Assess Met / Partial / Not Met / Not Applicable.
  6. For Not Applicable, record a rationale and evidence basis.
  7. Normalize the score over applicable weight only.
  8. Treat mandatory governance gaps independently from the maturity score.
  9. Feed gaps into Product evolution and Continuous Assurance.

The List of AIPCH Characteristics​

1–1112–21
AIPCH01 - Domain-OwnedAIPCH12 - Semantically Aligned
AIPCH02 - DeployableAIPCH13 - Consumption-Driven Intent
AIPCH03 - Declarative SpecificationAIPCH14 - Testable & Versioned
AIPCH04 - Declarative Intelligence Definition & CompilationAIPCH15 - Economically Accountable
AIPCH05 - DiscoverableAIPCH16 - Explainable & Transparent
AIPCH06 - Self-DescribingAIPCH17 - Bias-Controlled & Fairness-Measured
AIPCH07 - TrustworthyAIPCH18 - Governed Evolution & Adaptation
AIPCH08 - ReusableAIPCH19 - Safe & Policy-Bound Usage
AIPCH09 - SLA–SLO Backed & ObservableAIPCH20 - Interoperable & Composable
AIPCH10 - Compliant by DesignAIPCH21 - Human-Centered Oversight
AIPCH11 - Addressable

Machine-Readable PMDD Profile​

AIPS 0.1 provides a machine-readable PMDD profile at:

/spec/aips/pmdd/0.1/

It captures version-bound qualification, applicability-aware AIPCH assessments, maturity band, evidence, and governance blockers.


Final Principle​

PMDD continuously proves the maturity of a qualified AI Product; it does not decide whether the thing is a Product or whether it is intrinsically an AI Product.