fix(hf): avoid eager 27GB dataset download on /api/cases (#46)

fix(audit): comprehensive P1-P4 audit fixes (#45)

fix(arch): comprehensive architecture audit fixes (#44)

fix(security): address P1 findings from post-refactor audit

refactor(data): use standard datasets.load_dataset() with neuroimaging-go-brrrr

fix(arch): Config SSOT, reproducible builds, and data pipeline documentation (#41)

fix: address P1/P2 bugs from audit (BUG-005/006/008/010/011) (#40)

fix: apply CodeRabbit feedback (security + lint)

fix: comprehensive frontend-backend integration fixes

fix(security): apply CodeRabbit security feedback (#38)

fix(ci): format code and add ErrorBoundary tests to pass CI

fix(api): add mypy ignore for optional torch import

fix(api): validate case_id and log volume computation errors (BUG-011, BUG-012)

feat(api): add ephemeral storage warning to results (BUG-005)

fix(api): anchor CORS regex and add CORP middleware (BUG-002, BUG-003)

feat(api): async job queue with comprehensive test coverage (#36)

fix(deploy): CORS and proxy-headers for HF Spaces (#35)

feat(api): FastAPI REST backend for React frontend

feat(frontend): React + Vite + NiiVue frontend (replaces Gradio) (#32)

feat: Gradio Custom Component for NiiVue (#29)

fix(ui): use head= for NiiVue loading, not dynamic import in js_on_load (#24) (#28)

fix(ui): load NiiVue directly in js_on_load, remove head= dependency (#24)

feat: implement prediction mask binarization for improved visibility in slice comparison

fix(ui): simplify NiiVue loading, add diagnostic logging (#24)

fix(ui): add gr.set_static_paths() to fix HF Spaces Loading bug (#24) (#26)

fix(ui): use head_paths for NiiVue loading on HF Spaces (#24) (#25)

fix(ui): vendor NiiVue library to bypass HF Spaces CSP (#24) (#24)

fix(ui): prediction overlay invisible, race condition, thread safety (#23) (#23)

fix: remove silent failures and redundant checks

fix(ui): replace base64 data URLs with Gradio file serving (#19)

fix(ui): NiiVue viewer re-initializes after segmentation completes (#21)

fix(ui): use js_on_load for NiiVue viewer instead of script tags (#20)

fix(docker): use subprocess bridge for Python version isolation (#18)

debug: add logging to diagnose DeepISLES path discovery

fix(docker): relocate app to /home/user/demo to preserve DeepISLES modules (#17)

fix(data): bypass load_dataset() to fix HF Spaces streaming hang and OOM (#16)

fix(data): use streaming mode to fix HF Spaces dataset hang (#15)

fix(ui): lazy dataset loading to prevent HF Spaces timeout (#14)

fix(ui): use lazy imports to avoid RuntimeWarning on startup (#13)

fix(pipeline): copy input files to results_dir and add UI cleanup

fix(pipeline): prevent unbounded disk usage from HuggingFace temp files

feat(data): integrate HuggingFace dataset as primary data source (#11)

fix(inference): correct DeepISLES Docker volume mounts and CLI args

fix: resolve technical debt (P2/P3) with TDD validation (#9)

feat(hf-spaces): complete deployment infrastructure for Hugging Face Spaces (#8)

feat(phase-5): polish, observability, and documentation (#6)

feat(phase-4): Gradio UI with NiiVue visualization (#5)

feat(phase-3): End-to-end pipeline with metrics and CLI

feat: Phase 1A + Phase 2 - Local data loader and DeepISLES Docker wrapper (#3)

fix(phase-1): implement URL materialization and add requests dep