Run Model Reference
| Field | Meaning |
|---|---|
id | Random run_ plus 32 lowercase hexadecimal characters. |
source | chat, openai_chat, alert:<provider-type>, telegram_bot, mattermost_bot, change_preview, or change_apply. |
actor | api-key, anonymous, alert-source:<name>, or a bounded bot actor. |
workspace | Named workspace when applicable. |
model | Akritas facade model ID. |
status | running, succeeded, or failed. |
started_at, completed_at | UTC timestamps. |
error | Bounded host error category. |
metadata | Sanitized bounded string map. |
events | Ordered audit events with one-based sequence numbers. |
Final statuses are terminal. Duplicate starts, out-of-order events, and repeated finishes make history invalid.
Alert Run metadata links the Run to incident_id, event_id, and job_id.
The validated investigation object is stored as a separate append-only record
and restored with Run detail after restart.