feat: solid 3D mesh rendering with mirrored back + edge faces

fix: correct seg mask rendering in inspection view by mapping cropped frame to full-frame mask coordinates

fix: 3D point cloud rendering — normalize depth scale + fix bbox label NaN

fix: resolve detector name from segmenter for attention maps in segmentation mode

fix: remove bbox overlay from inspection view — show object only

fix: clean up video display — remove track ID labels, smaller fonts, masks-only for segmentation

fix: bundle Three.js locally to avoid CDN failures on HuggingFace Spaces

fix: handle degenerate bbox and unhandled exceptions in inspection frame endpoint

feat: side-by-side inspection/chat layout + 360° spin interaction

feat: crop inspection to selected object bbox with 20% padding

chore: add .claude/ to gitignore

fix: forensic code trace fixes across all inspection modules

fix: enable multi-GPU round-robin for all inspection modules

feat(inspection): add super-resolution and 3D point cloud (Phase 3)

feat(inspection): add depth analysis and attention heatmap endpoints (Phase 2)

Merge branch 'worktree-agent-aabfbe78'

feat(inspection): add on-demand SAM2 mask generation for detection mode

feat(inspection): add frame extraction and mask retrieval API endpoints

feat(inspection): add canvas renderers for seg, edge, depth, attention, super-res

feat(inspection): persist RLE masks during segmentation inference

feat(inspection): add panel controller with mode switching and lifecycle

feat(inspection): add RLE mask encode/decode module

feat(inspection): add frame extraction and cropping module

feat(inspection): add panel HTML structure and CSS styles

feat(inspection): add inspection API client module

feat(inspection): add inspection state sub-object

fix: merge redundant track card badges into a single unified status badge

fix: prioritize mission-relevant tracks in card ranking and make list scrollable

fix: use actual video FPS for frame index calculation to prevent track card disappearance

fix: keep last known tracks when backend returns empty frame data

fix: assessment flicker, pause/resume toggle, chat icons, overlay bbox scaling

fix: add chat.js script tag, remove missing features.js and demo data references

feat: add mission analyst chat panel with track context injection

refactor: replace canvas overlay with SVG-based bounding box system

debug: add console.log to overlay render for track highlight debugging

fix: comprehensive overlay highlight fix for track selection

fix: ensure overlay canvas is sized and tracks preserved for highlight

fix: highlight selected object on video when clicking track card

feat: pause video when clicking track card or bounding box

fix: add cache-busting version params to all frontend assets

fix: streaming track poll used wrong summary key, causing NaN frame index

fix: show track cards during MJPEG live streaming

fix: track cards not displaying during streaming and replay

docs: add planning docs policy to CLAUDE.md

chore: remove plan/design docs from tracking

feat(isr): wire mission param through API/background pipeline, add openai dep

feat(isr): store latest frame in writer thread for ISR cropping

feat(isr): add latest-frame storage for ISR cropping

feat(isr): create async assessor loop with verdict merging

feat(isr): create ISRAssessor with GPT-4o-mini vision integration