Admin

Wrecks regenerate their AI narratives (What/How/So-What) when scoring or OSINT inputs change. The local Ollama worker processes in batches of five every two minutes.

Pending —

Processing —

Failed —

Total —

Use Queue ALL after a methodology change to rebuild every narrative against the current scoring rules, regardless of when each was last written.

Failed items

Maintenance utilities

Run all monitors

Manually trigger storm + seismic pollers

Process alert queue

Reassess wrecks flagged by environmental alerts

Reprocess images

Re-source images for wrecks missing visuals (Wikimedia)

Approval backfill

Create missing approval docs for legacy users

Trash bin

Review archived wrecks; restore if needed →

LLM diagnostics

Default backend is openai-compat (local Ollama via worker host). Gemini probe below is for the fallback path when LLM_BACKEND=gemini is set.

Gemini probe: —

Preferred model (config): —