AI Systems (Operational Governance)

Layer-2 domain pack: descriptive invariants about metrics, cross-boundary effects, feedback, and comparator tiering in AI deployments.

Layer-2 descriptive grounded

L2-AI-01 — Model Metrics Compress Reality

Statement

Model evaluation metrics compress diverse behaviors into tractable scores that can distort optimization.

Primitives

P10 P4 P6 P5

Composites

C3 C4

Notes

Benchmarks are comparators; optimizing them can miss deployment context.

L2-AI-02 — Effects Can Cross Product and Organizational Boundaries

Statement

AI system effects can manifest outside the product boundary and outside internal ledgers.

Primitives

P1 P5 P9 P6

Composites

C5

Notes

Boundary-accounting misalignment makes downstream effects invisible.

L2-AI-03 — Comparators Drift Under Incentives

Statement

Operational comparators tend to drift toward what is easiest to measure and most rewarded.

Primitives

P7 P10 P6

Composites

C11 C3

Notes

Hidden comparators become sacred; metric capture becomes governance reality.

L2-AI-04 — Local Optimization Can Alter System-Level Outcomes

Statement

Subsystem optimization can shift system-level outcomes through misaligned objectives and coupling.

Primitives

P1 P6 P10 P7

Composites

C10

Notes

Local KPI wins can degrade global behavior.

L2-AI-05 — Feedback Delays Distort Risk Perception

Statement

Effects may appear with delay, biasing governance toward what is immediate.

Primitives

P6 P10 P5

Composites

C9 C10

Notes

Delayed feedback increases oscillation and misattribution.

L2-AI-06 — Attribution Simplifies Systemic Causality

Statement

Outcomes are often attributed to individuals or single causes despite recursive system structure.

Primitives

P6 P7 P10

Composites

C8 C4

Notes

Blame narratives replace loop-level diagnosis.

Use this pack to map real artifacts (policies, configs, incidents) into the spine. Then run a gap-check: what grounding effects are missing?