Maintenance Log

Sovereign Artifact Rebuild

Use the managed GPU environment and the canonical maintenance entrypoint:

env PYTHONPATH=. /K3D/Knowledge3D.local/envs/k3d-cranium/bin/python \
  scripts/rebuild_sovereign_artifact.py \
  --refresh-feed-source --refresh-build-feed --force-rebuild --verbose

This rebuild is mandatory after:

• foundational star schema changes
• layer, role, or ref wiring changes
• default knowledge signature changes

Operational rules:

• let the rebuild finish; do not interrupt it mid-write
• the rebuild loads the full resident Knowledgeverse, not a selective galaxy subset
• normal warm boot is not the artifact generation mechanism
• after rebuild, normal boot should load in artifact or resident mode
• --refresh-feed-source is the only maintenance step allowed to rehydrate source rows
• the build feed is the authoritative production rebuild input; if it is stale or missing, rebuild fails fast instead of translating source rows on boot

Storage policy:

• gpu_cache/ keeps one active flat_<signature>.npy plus one paired catalog_<signature>.pkl
• heavy timestamped checkpoint families are rotated automatically:
  • keep 2 galaxy_consolidated_*
  • keep 2 sovereign_runtime_bundle_*
  • keep 3 trm_weights_*
  • keep 3 shadow_patterns_*

Cleanup — Local Workspace Reset

• Date (UTC): 2025-09-07T07:58:36Z
• Local path: ../Knowledge3D.local
• Removed: models/* (except root-owned entries), logs/, mr/
• Kept: datasets/ (downloaded data), repos/ (source checkouts), conda_pkgs/ (caches)

Notes

• One or more root-owned artifacts could not be deleted without elevated privileges (e.g., models/world_rssm/rssm.pt). If desired, remove manually via:
  • sudo rm -rf ../Knowledge3D.local/models/world_rssm
• The goal is to start fresh while preserving downloaded datasets to save bandwidth/time.

Next

• Build fresh 50k multimodal GLBs using docs/DATASETS_50K.md.
• Export per-house memory to viewer/public/houses/<K3D_HOUSE_ID>/ going forward.