Disaster Recovery

The asset library is a production dependency. If it goes down — corrupted files, lost metadata, deleted storage — every page that references an asset displays a broken image. Disaster recovery planning ensures that the library can be restored from any failure state.

Three recovery scenarios. Scenario 1 — Individual Asset Loss: a single file is corrupted or deleted. Recovery: regenerate from the stored prompt in the metadata JSON. Same template, same variables, same style version. The regenerated asset will not be pixel-identical (generative models are stochastic) but will be visually consistent with the style specification. Redeploy to all locations listed in deployed_to.

Scenario 2 — Metadata Loss: the asset files exist but the metadata is gone. Recovery: this is why the naming convention encodes critical information. From hero-agent-teams-a-1920x1080-v2.0.png, you can reconstruct: type (hero), subject (agent-teams), variant (a), dimensions (1920x1080), style version (v2.0). The prompt and quality scores must be reconstructed from generation logs.

Scenario 3 — Full Library Loss: everything is gone. Recovery: regenerate from the template library and the CSV inputs. The templates and CSVs are the source of truth — if they are backed up (and they must be), the entire library can be regenerated. The style specifications, templates, and input CSVs are the three artifacts that must be backed up redundantly.