Skip to content
AIGP

RFC-035: Mediation Vector Profile — A Validity-Aware Measurement Model for Domains of Concern — 5. The Mediation Vector Profile: A Four-Layer Model

AIGP Specification › RFC-035: Mediation Vector Profile — A Validity-Aware Measurement Model for Domains of Concern › 5. The Mediation Vector Profile: A Four-Layer Model

← 4. Measurement Science Foundation: A Concern Vector Is a Measurement Claim · Section index · 6. Structural Parallel →

5. The Mediation Vector Profile: A Four-Layer Model

The combined force of the measurement science above is that a Mediation Vector must not be a flat vector. It must be a profile with four layers, each answering a distinct question:

graph TD
    L1[Layer 1: Construct<br/>What concern is being measured?] --> L2[Layer 2: Evidence<br/>What evidence is admissible?]
    L2 --> L3[Layer 3: Measurement<br/>What variables, typed, normalized, warranted?]
    L3 --> L4[Layer 4: Calculation<br/>How may RFC-036 compute concern posture?]
Layer Question Defines Source Discipline
1. Construct What concern is being measured? concern_id, construct_definition, observable_indicators, measurement_assumptions, known_invalid_uses Construct validity (Cronbach & Meehl; Messick)
2. Evidence What evidence is admissible? admissible_evidence_types, admissibility_levels, provenance_requirements W3C PROV; AIGP D-DNA
3. Measurement What variables, how typed, normalized, warranted? variable declarations with type, scale, allowed_operations, extraction, uncertainty, warrant Measurement theory (Stevens); GQM (Basili); Toulmin
4. Calculation How may concern posture be computed? aggregation rules, hard-stops, weights, handoff to RFC-036 Decision analysis (deferred to RFC-036)

A profile is incomplete if any layer is absent. A vector without a construct layer measures nothing defined. A vector without an evidence layer measures without grounding. A vector without a measurement layer permits bad math. A vector without a calculation layer cannot produce posture.

5.1 The Admissibility Rule

The governing constraint, stated normatively:

A variable MUST NOT participate in empirical concern calculation unless its value, type, admissible evidence source, extraction method, provenance, uncertainty, and allowed mathematical operations are declared.

This single rule is what separates a measurement profile from a scoring heuristic.

← 4. Measurement Science Foundation: A Concern Vector Is a Measurement Claim · Section index · 6. Structural Parallel →