# PROJECT_STATE.md

Project: OTB Cloud
Version: v1.1.0-alpha4
Updated: 2026-04-20
Location: /opt/otb_cloud

## Current State
OTB Cloud now has a working multi-profile, multi-GPU video processing pipeline integrated into the tenant storage platform.

### Confirmed Working
- Portal-branded OTB Cloud dashboard
- Device creation and browsing
- File-ID based workshop staging
- Device-specific Video Workspace access from dashboard
- Video Workshop page
- Multi-profile processing selection:
  - default
  - compress
  - hq
- Manual rotation override option:
  - auto/default behavior when unchecked
  - selectable 90 / 180 / 270 override when enabled
- Job queue API and job listing API
- File-ID based source resolution
- Output routing to:
  - devices/<device>/video/
- Profile-specific output filenames
- Completed job actions:
  - View
  - Send to LTS
  - Download Output
  - Delete
- Failed job delete action
- LTS routing by file type:
  - lts/video
  - lts/archived
  - lts/pictures
- Health page
- Lifetime processing metrics retained after visible job deletion
- Intel + AMD GPU processing both in service
- GPU time accounting active in Health page
- Global video jobs route exists in codebase
- Processed video section exists in device browser flow

### Processing / GPU Behavior
Current live behavior:
- both GPUs take jobs
- AMD prioritizes heavier / HQ work first
- Intel handles lighter work
- workers continue taking suitable jobs from the queue batch as available

### Latest Proven Health State
Health page currently shows stable cumulative values including:
- uploaded file counts and space
- LTS counts and space
- archive counts and space
- total jobs
- completed jobs
- failed jobs
- cumulative GPU time that does not zero out when workshop cards are deleted

### Current Storage Layout
- originals remain in device originals tree
- processed outputs go to:
  - devices/<device>/video/
- LTS destinations include:
  - lts/video/
  - lts/archived/
  - lts/pictures/

## Known Remaining Improvements
- README is now being realigned to actual live state
- Global video jobs page should be fully wired into UI navigation and polished
- Dashboard template still contains some mixed button class styles that should be normalized
- Health page can be expanded with per-processor breakdown later
- Processing metrics can be refined further into Intel/AMD/CPU buckets if desired
- Output browsing UX can still be improved further with richer previewing and filtering

## Recommended Next Step
Proceed after alpha4 with:
1. global video jobs page polish and filters
2. per-processor GPU metrics split (Intel / AMD / CPU)
3. scheduler documentation and/or scheduler UI visibility
4. processed output browsing improvements in device view