Commitment Requires Authority
Binding consequences demand a named decider
No L2 or L3 derivation becomes operational ground truth without crossing an explicit commitment boundary where a named authority accepts responsibility.
Before the commitment boundary, the system computes — scores, recommendations, draft states. These artifacts inform; they do not bind. After the boundary, something in the world has changed because a named authority decided on known evidence, under known policy, at a documented moment. Technically, writing to a log is trivial; architecturally, the boundary is the transformation of epistemic status from inference to principled fact.
This principle connects truth and accountability. Truth requires that committed facts be immutable; accountability requires that committed facts be attributable. A model score landing in an operational table without a commitment event violates both. So does “the system decided” with no human or certified process named in the record.
Reversal is always forward — new L0 events that supersede or compensate — never silent rewrite of what was committed. The commitment boundary pattern operationalises this principle; accountability without commitment names the failure mode when consequences exist but no authority ever crossed the boundary.