Strategy intelligence (V3–V5)

V3–V5 help a collective learn, exchange evidence and synthesize candidate coordination strategies without converting learning into authority.

V3 — bounded local adaptation

V3 keeps outcome measurement, local learning, proposal generation, counterfactual analysis, recommendation, review, catalog transition and execution as separate stages. It measures mission progress, latency efficiency, resource efficiency, recovery quality and safety. Missing evidence is indeterminate; callers cannot inject a scalar reward.

Governance actions are promote, degrade, retire and rollback. Reviews may come from an authorized agent, person or collective. Baseline retirement, self-review, stale recommendations, low confidence and excessive switching fail closed.

V4 — collective intelligence over Agent Mesh

Peers exchange signed, content-free outcome attestations containing bounded identifiers, digests, metrics and dispositions—not prompts, reasoning, credentials, source code or private mission content.

Receiving peers reapply local catalog compatibility, Membership, signature, Trust, diversity and governance policy. Robust bounded priors and convergence results remain advisory. A remote peer cannot approve or execute a local change.

V5 — governed strategy synthesis

V5 turns evidenced catalog gaps into declarative candidates. Generation, reproducible adversarial simulation, independent certification, Blueprint/Trust/Inference eligibility, governance review, experimental canary and catalog materialization are distinct boundaries.

No stage installs arbitrary code. A synthesized candidate remains inert until ordinary local governance later authorizes its use through Morphogenesis.

Safe interpretation

A favorable simulation, collective prior or learned weight is evidence about a bounded experiment—not proof of real-world superiority. Production claims require preregistered operational evidence.