Skip to content
AIGP

RFC-035: Mediation Vector Profile — A Validity-Aware Measurement Model for Domains of Concern — 3. Ontological Foundation: What Exists to Be Observed

AIGP Specification › RFC-035: Mediation Vector Profile — A Validity-Aware Measurement Model for Domains of Concern › 3. Ontological Foundation: What Exists to Be Observed

← 2. Epistemological Foundation: What We Learn from Science · Section index · 4. Measurement Science Foundation: A Concern Vector Is a Measurement Claim →

3. Ontological Foundation: What Exists to Be Observed

3.1 From Epistemology to Ontology

Section 2 establishes how we know — the scientific method as observation protocol. This section establishes what exists to be observed — the irreducible entities that constitute a mediated intelligence event.

Before we can define variables, declare evidence sources, or compute gaps, we must answer: what are the fundamental things that exist in every governed AI mediation event, regardless of industry, model, or domain? What is the minimum set of entities whose existence and relationships make observation meaningful?

3.2 The Seven Primitives of Mediated Observation

This RFC defines seven minimum primitives required for mediated observation. These are irreducible within this specification — remove any one and the observation system collapses into something that cannot produce reliable knowledge. Implementations MAY extend the ontology with additional primitives, but MUST preserve these seven and their relationships.

Primitive What It Is Why It Is Irreducible
The Artifact The thing produced by mediation — an output, action, decision, recommendation, or generated object Without an artifact, there is nothing to observe. Observation requires an object.
The Mediation The transformative act that produced the artifact — model inference, agent reasoning, tool invocation, autonomous decision Without mediation, the artifact is human-authored and does not require this apparatus. Mediation is what makes AI governance necessary.
The Authority The governance decision that permitted the mediation to occur — the CHECK that preceded execution Without authority, mediation is ungoverned. It happened, but nobody authorized it. This is the pre-invocation gap.
The Evidence The signed, traceable record that the mediation occurred under declared conditions — D-DNA, TRACE, RECORD Without evidence, observation is ungrounded. You believe it happened but cannot prove it. This is the difference between a log and a scientific record.
The Concern The evaluative question being asked of the artifact within its domain — the specific thing we need to know Without a concern, observation has no direction. You are measuring without knowing what matters. This is why RFC-034 (Domains of Concern) is a prerequisite.
The Observer The qualified entity — human, machine, or hybrid — that renders a verdict on the concern Without an observer, evidence accumulates without interpretation. Data without judgment is not knowledge. This is why the Fourth Law requires external observation.
The Verdict The observer’s judgment: the artifact satisfies or fails to satisfy the concern, given the evidence Without a verdict, the loop is open. Evidence was gathered but no knowledge was produced. This is the post-hoc gap that RFC-032 closes.

3.3 Irreducibility Test

Each primitive can be tested by asking: what happens if it is absent?

No Artifact   → Nothing to evaluate. The system has no object of observation.
No Mediation  → No governance need. The work is human-authored.
No Authority  → Ungoverned mediation. The pre-invocation gap (AIGP's founding problem).
No Evidence   → Ungrounded observation. Belief without proof.
No Concern    → Undirected measurement. Data without purpose.
No Observer   → Uninterpreted evidence. Signals without judgment.
No Verdict    → Unclosed loop. The post-hoc gap (RFC-032's founding problem).

If removing a candidate entity collapses the system, it is a primitive. If removing it merely reduces fidelity but the system still functions, it is a property of a primitive, not a primitive itself.

3.4 Relationships Between Primitives

The seven primitives form a directed observation graph:

graph TD
    AU[Authority] -->|permits| ME[Mediation]
    ME -->|produces| AR[Artifact]
    ME -->|is recorded as| EV[Evidence]
    CO[Concern] -->|scopes| OB[Observer]
    EV -->|is consumed by| OB
    AR -->|is subject of| OB
    OB -->|renders| VE[Verdict]
    VE -->|feeds back to| AU

Reading the graph:

  1. Authority permits Mediation (governance authorizes the AI to act)
  2. Mediation produces an Artifact (the AI generates output)
  3. Mediation is recorded as Evidence (D-DNA captures what happened)
  4. Concern scopes the Observer (the Domain of Concern defines what to look for and who is qualified)
  5. Evidence is consumed by the Observer (the observer examines the record, not the artifact in isolation)
  6. The Artifact is the subject of the Observer’s attention (the observer evaluates the thing that was produced)
  7. The Observer renders a Verdict (judgment is produced)
  8. The Verdict feeds back to Authority (governance adapts based on empirical results — the closed loop)

3.5 The Cycle

The feedback edge (Verdict → Authority) is what makes this a cycle rather than a pipeline. Each verdict informs future authority decisions:

  • MATCH → authority can relax (scope expansion, higher autonomy)
  • MISMATCH → authority tightens (scope narrowing, added gates)
  • VIOLATION → authority halts (circuit break, revoke scope)

This is the empirical governance adaptation loop — the system learns from its own verdicts. In scientific terms: the body of evidence accumulates, theories are revised, and experimental conditions are adjusted for the next observation.

3.6 The Four States Apply to Every Primitive

The four-state model (§8) applies not only to individual variables but to each primitive itself:

Primitive Anticipated Actual Evidenced Perceived
Artifact What it should contain (ANTICIPATE criteria) What it contains (execution output) What D-DNA proves it contains (signed record) What the user believes it contains (FEEDBACK)
Mediation What model/tool should run (PLAN_SUBMIT) What ran (STEP_COMPLETE) What TRACE proves ran (stage spans) What the operator believes ran (assumption)
Authority What should be authorized (scope envelope) What was authorized (CHECK decision) What D-DNA proves was authorized (signed permit) What the user assumes was authorized (trust)
Evidence What should be provable (admissibility requirements) What is provable (available signed artifacts) What is independently verifiable (recomputation witness) What the organization claims is provable (assertion)
Concern What should matter (Domain of Concern definition) What matters (active concern classes) What the domain formally declares (registered criteria) What stakeholders believe matters (organizational assumption)
Observer Who should judge (accreditation requirement) Who judged (grader_id in VERIFY) Who is certified to judge (calibration record) Who is assumed qualified (organizational belief)
Verdict What the outcome should be (passing threshold) What the verdict is (MATCH/MISMATCH/VIOLATION) What the signed conformance record proves What the consumer believes the verdict means (interpretation)

The gaps between columns are where governance failures hide. An organization that perceives its observer as qualified when the observer is actually uncalibrated has a perception gap on the Observer primitive. An organization that anticipates full evidence but actually has Level 4 (declared-only) evidence has an anticipation gap on the Evidence primitive.

3.7 Ontological Sufficiency Claim

These seven primitives constitute the minimum ontology required by this specification for mediated observation. Any mediated intelligence event — regardless of model, framework, industry, or jurisdiction — can be described in terms of these entities and their four states. Implementations MAY extend the ontology to capture domain-specific entities, but MUST preserve these seven primitives and the relationships among them.

This claim is testable: present a mediated intelligence scenario that cannot be described using these seven primitives. If such a scenario exists, an implementation extends the ontology to cover it — without removing or redefining the seven. The seven are the floor, not the ceiling.

The Mediation Vector (§7 onward) is the technical mechanism by which this ontology is made measurable — typed variables that quantify the state of each primitive and the gaps between states.

← 2. Epistemological Foundation: What We Learn from Science · Section index · 4. Measurement Science Foundation: A Concern Vector Is a Measurement Claim →