ASDGF makes agentic swarm governance architecturally explicit. It does not treat a swarm as merely a collection of individually governed agents. It defines the swarm itself as the governed system: the objectives, population, coordination media, topology, authority paths, resolution mechanisms, and intervention structures through which collective behaviour emerges.

The framework begins from a fundamental distinction: an agent can conform individually while the swarm formed by multiple conforming agents can still produce effects that do not belong to any single member. Governance therefore cannot stop at member-level permissions, tool controls, or model behaviour. It must govern the relationships through which agents aggregate effects, compose capabilities, and influence one another.

ASDGF occupies the architectural layer above AAWDF and alongside TAGOF. AAWDF specifies the architecture of an individual agent through ⟨E, L1…L5⟩; TAGOF supplies enterprise governance and agentic-action controls; ASDGF governs the structure that emerges when multiple agent instances interact as a swarm. Conformance to the member-level frameworks is therefore a precondition, not a substitute, for swarm-level conformance.

From these mechanisms, ASDGF defines five governing properties of a governable swarm:

• Bounded Effect — the realizable effect set is enumerable and quantitatively bounded before execution, at swarm scope.
• Topological Legibility — the observed interaction topology can be reconstructed and compared against the topology that was actually authorized.
• Attribution — every effect resolves to a decision path and a named human accountability holder.
• Evidential Independence — the evidence used to govern the swarm remains outside the swarm’s own effect set.
• Interruptibility — the swarm can be halted within a declared and measured bound, with compensating mechanisms for effects already realized.

ASDGF then converts architecture into governance intensity through a computed classification:

⟨K, N, E⟩

where:

K — Coupling measures the reachability and mediation structure between members.

N — Population Mutability captures whether the population is fixed, elastic within a declared ceiling, or capable of self-extension.

E — Effect Reach captures whether effects remain reversible and operator-controlled, become externally observable, or become irreversible.

The important architectural principle is that class is computed from the specification rather than assigned through committee judgement. Governance therefore follows the structure actually designed. A team seeking a lower governance burden must change the design rather than negotiate a lower risk label.

Class then maps to explicit conformance profiles: Profile A — Constrained, Profile B — Mediated, Profile C — Open, together with an out-of-scope state and a structurally prohibited configuration. ASDGF explicitly prohibits ⟨K2, N2, E2⟩: open coupling, self-extension, and irreversible effect reach combined in one system.

The governance requirement architecture is organized into seven families:

• OB — Objective and Evaluation Integrity
• PB — Population Bounding
• IC — Interaction Control
• IA — Influence Admissibility
• RE — Record and Evidential Independence
• IV — Intervention and Compensation
• CA — Conformity and Assurance

Every requirement must carry both a verification method and an evidence artefact. A requirement without a way to verify it and evidence that demonstrates its operation is not admitted into the control catalogue.

ASDGF-2026 v0.3.1 is currently designated as a working specification proposed for expert review, rather than a settled standard.