Labs

Labs contains seven experimental capabilities:

1. Run Checks (Proof) verifies a recorded run against operator-controlled assertions.
2. Safety Preview (Sentinel) evaluates deterministic invariants before an operation.
3. Temporary Access (Capability Tokens) grants short-lived, scoped authority.
4. Restore Points (Rewind) snapshots selected local files and previews restoration.
5. Portable Runs (Capsules) exports and verifies redacted run bundles.
6. Compare Approaches (Counterfactuals) runs bounded plans in copied candidate workspaces.
7. Smart Routing (Darwin) scores model routes from recorded outcomes.

Every Lab is experimental and off by default in v0.4.0-beta.3. Automatic improvements is a separate bounded feature and is not an eighth Lab.

Inspect gates

odinn experimental status
odinn experimental help proof

The console shows the same startup gate state. Enabling a feature in configuration requires restarting an already-running gateway before its UI and API become available.

Enable deliberately

odinn experimental enable proof --confirm-impact

Enable only the feature you are actively testing. Each gate is stored separately; enabling one does not grant another.

Shared boundary

Labs use the durable run ledger and audited tool boundary. They do not create:

• an operating-system sandbox;
• deterministic model behavior;
• guaranteed reversal of external effects;
• tamper-proof storage against an attacker who controls local state.

Restore and selection actions default to previews where applicable. Full capsule replay and applied restoration retain additional checks.

Choose a Lab

• Need evidence that a run achieved an operator-defined result? Start with Proof.
• Need deterministic pre-operation policy? Use Sentinel.
• Need one-use authority? Use Capability Tokens.
• Need local file recovery? Use Rewind.
• Need a portable evidence bundle? Use Capsules.
• Need to compare bounded candidate workspaces? Use Counterfactuals.
• Need transparent model-route scoring? Use Darwin.