Run evidence
Observed
Captured from a specific run or operating period with known versions and conditions.
Proof model
A claim about an agent or swarm is useful only when the system boundary, evidence class, scenario, run conditions, review, and decision are clear.
Evidence flow
Each link can be inspected and challenged independently.
State exactly what the system, agent, control, or workflow is believed to do.
Define representative inputs, context, tools, authority, constraints, and expected boundaries.
Capture the versions, state, messages, calls, policy decisions, cost, latency, and outcome.
Apply automated checks and named human judgment using explicit criteria.
Proceed, narrow, remediate, accept risk, defer, or stop—and record the conditions.
Classification
Evidence is labeled by what it can support.
Run evidence
Captured from a specific run or operating period with known versions and conditions.
Repeatability
Repeated under controlled conditions with materially consistent results.
Human judgment
Assessed by a named reviewer against stated criteria and authority.
Analysis
A reasoned conclusion from available evidence, explicitly marked as an inference.
Future state
A design, target, or control that has not yet been implemented or observed.
Gap
A material question that remains unresolved and affects the decision.
Swarm-level proof
Multi-agent systems can produce a plausible outcome through unsafe, wasteful, brittle, or irreproducible coordination.
Start with a bounded decision
A readiness sprint or evaluation engagement can turn an informal claim into a decision-ready proof chain.