Beyond Delegation Depth: Audit Reach, Enforcement Reach, and Management Coverage in Recursive AI Delegation

Recursive AI delegation is commonly constrained by hop counts, but equally deep structures can differ in whether authority remains attributable, auditable and stoppable. This paper separates path-level management-substrate coverage from contiguous audit reach, structural enforcement reach and timely enforcement reach, all relative to an explicit assurance context. Non-monotone reach is reported as a set, and graph-level denial is evaluated against the effective authorization support and revocation objective. A timed support-closure identity connects authorization topology to the deadline of effective denial. Strict effect-boundary enforcement is distinguished from residual exposure bounded jointly by expiry and aggregate execution controls. An indistinguishability argument shows why sample evidence alone cannot guarantee an unsampled branch property, for either audit or enforcement; the operational distinction is between population evidence and universal prevention. Further elementary results identify a depth-only decision limitation and conditions under which actual propagation delay increases revocation-race risk. Worked calculations are logical witnesses, not empirical results. The resulting admissibility framework incorporates impact, reversibility, coverage, reach, support topology, timing and evidence freshness. A prospective evaluation contract separates pre-action certificates from held-out outcomes, includes component ablations, request and follow-up accounting, competing hypotheses and full assurance costs, and prevents absence of a guarantee from being counted as execution failure. The contribution is this operational composition, not the underlying delegation, revocation, sampling or complete-mediation mechanisms. Note on Version 2.0: this version revises the registered v1.0 (about 12,100 to 16,000 words). It adds a timed support-closure result and temporal composition of closure, worked logical cases and a conditional timing result, certificate evaluation with bounded exposure, and a rebuilt Section 9 covering the measurement contract and minimal test design, external measurements and transfer conditions, discriminating predictions with competing explanations, and a joint test of support, timing and current evidence. No empirical result is claimed: the supporting archive records 53 exact analytical checks on cited summaries, external evidence records, a public result recalculation, a novelty audit, a change log and a validation record with SHA-256 checksums. Files: the v2.0 manuscript and that supporting archive. This paper derives from the author's broader Information Ecosystem Theory, which remains unpublished at the time of this release; it is written to stand alone.