L2 — Compound Aggregate

L2 denotes a compound aggregate: the output of multi-step computation over lower-level artifacts — sums, joins, model scores, rule-engine evaluations, LLM drafts. L2 artifacts may be committed or uncommitted; most operational risk lives in treating uncommitted L2 output as if it were L0 ground truth.

The commitment boundary exists precisely because L2 informs decisions but does not bind until an authority crosses it. Model output, credit scores, and recommendations stop at the boundary; only boundary-crossing operations emit L0 events. Inference-as-fact is the failure mode when this separation collapses.

Epistemic uncertainty compounds across L2 processing steps. Each probabilistic step multiplies distance from ground truth. Class B consumers may use L2 with appropriate reservation; Class A invariant enforcers require L0 or low-uncertainty L1 — never silent promotion of L2 to operational authority.