01
Understandable
The system boundary, roles, state, memory, tools, authority, and evidence can be explained and inspected.
Long-Term Intelligence Framework
A framework for machine intelligence that remains understandable, governable, and useful
Machine intelligence is durable when the organization can understand, constrain, evaluate, recover, and improve the system even as models, providers, tools, and teams change.
EnterpriseGovernmentPartners
Framework
Six durability dimensions
A weakness in one dimension can undermine the rest of the system.
02
Governable
Policies, permissions, budgets, approvals, exceptions, and shutdown are executable and owned.
03
Evaluable
Representative behavior, coordination, cost, latency, safety, and recovery can be measured and compared.
04
Recoverable
Failures can be detected, contained, reconstructed, corrected, replayed, and learned from.
05
Portable
Contracts, memory, evidence, controls, and operating knowledge survive provider and framework change.
06
Capability-building
The system improves organizational judgment and reusable operating capability rather than creating opaque dependence.
Self-review
Run a local self-review
Use one bounded workflow and answer with evidence, unknowns, and owners.
Choose the workflow
Name the decision, users, agents, tools, actions, and consequence boundary.
Score each dimension
Use strong, partial, weak, or unknown—never average away a critical weakness.
Link evidence
Attach architecture, evaluation, control, incident, and operating records.
Name the blocked decision
Identify what cannot credibly be approved, released, expanded, or funded.
Select the smallest intervention
Choose a readiness, architecture, evaluation, implementation, or operating step.
Start with a bounded decision
Use the framework to narrow the next decision.
A weak or unknown dimension is a signal for evidence—not automatically for a larger project.