Trust / Reliability & recovery

Interruptions are expected, not exceptional

A mission that cannot survive a dropped connection or a lost worker is not reliable. Omnacta OS treats interruption as a normal case with a defined recovery path, not an unhandled exception.

In development

Recovery behavior below is demonstrated for local, single-repository reference execution. Distributed, multi-tenant recovery guarantees are in development.

A recovery story

A worker executing one workstream in a multi-step mission loses its connection mid-task. The scheduler detects the expired lease, reassigns the outstanding checkpoint, and the workstream resumes from its last durable state. Unrelated workstreams in the same mission were never paused. If the resumed step still cannot complete, the mission blocks with an explicit reason rather than reporting a false success.

Failure modes and defined responses

Failure modeResponse
Worker crashLease expires; checkpoint is reassigned to a new worker; execution resumes from the last durable state.
Network disconnectClient reconnects and resumes from its event cursor; the mission itself was never paused server-side.
Partial external effectThe mission blocks and reports the uncertainty rather than assuming the effect succeeded or failed.
Conflicting concurrent editVersion-checked commands reject the stale write and surface the conflict instead of silently overwriting.

Uncertain outcomes are reported, not resolved by guessing

When the system cannot determine whether an external effect completed, it surfaces that uncertainty to the user or operator rather than picking an assumed answer. This is a deliberate design choice: an honest "we don't know" is safer than a confident wrong answer.

Current constraints

  • Recovery guarantees are proven for local, single-repository execution today.
  • Cross-repository and multi-tenant recovery coordination are in development.
  • Recovery time objectives are not yet published as a numeric SLA.

See how autonomy is bounded

Responsible autonomy covers the authority levels and human checkpoints that constrain what a mission can do without asking first.