Status: Accepted (draft v0.1)
Date: 2026-07-04
Decision: T0 = zero accountable humans; T1 = zero humans + independent agent corroboration; T2 = exactly one accountable human (either role); T3 = two distinct accountable humans (author ≠ reviewer).
Rationale: the original taxonomy (T0 AI/none, T1 AI/AI, T2 AI/human, T3 human/human) was authorship-centric and incomplete — human-authored/unreviewed code had no level (caught in the pre-spec review pass). Rederiving from P2 produced a complete, principled scale: every authorship × review cell maps, and the ordering follows from counting independent accountability rather than from opinions about AI code quality.
Consequence accepted deliberately: human+unreviewed and agent+human-reviewed both map to T2 at the tag level. Policies needing the distinction consume the full provenance vector in the attestation (ADR-003).
Rejected: bolting extra levels onto the original four (ad hoc, ordering disputes); making human-unreviewed T1-adjacent (equates one accountable human with zero).
Revisit trigger: empirical evidence on T1 efficacy (spec §12.1) could justify re-pricing, not restructuring.