Fix: Immediate task switcher visibility and hidden sandbox during runs

Fix: Custom prompt orchestration, metadata flickering, and task switcher visibility

fix: sample prompts directly run custom-only mode; root cause was missing customOnly flag

fix: custom prompts run single task only, not all 3 default tasks

fix: auto-fill custom prompts, keep prompt box visible during runs, run user prompt not default tasks

UI: Place overlay buttons in the same row

style: remove logo and title from startup overlay

fix: make custom prompts overlay scrollable and original size

feat: add custom command prompts to startup overlay

feat: add custom command prompts

fix: allocate PTY via script for HF Space live container logs

fix: remove tee pipe from start.sh to fix HF Space Container log streaming

docs: evidence proof above graphs, overlay fix, M2M section, logs fix

Update blog training hardware wording to Google Colab T4

Update notebook Cell 1 hologram setup instructions

Expand blog intro with safety and local model narrative

Add persistent logs endpoint for HF Spaces runtime logs

Clear stale UI when HF Space sleeps or backend disconnects

Fix local dev instructions: correct model path, add full local setup steps

Fix docs: rename Start Simulation references to Start inference.py

Add explicit print() logging for HF Space container logs visibility

Redesign Incident Summary card; fix container log visibility with unbuffered python

Fix RCA execution trace: escape pipe chars that broke markdown tables

Add task descriptions to start overlay; verify all README/BLOG links

Remove train_kaggle.py

Remove updated_blueprint.md, kaggle_training_notebook.py, kaggle_visualize_training.py

Fix RCA showing same for all tasks: reset per-task state on task transition

Show all three task labels (Easy/Medium/Hard) on the start overlay

Fix model path in openenv.yaml to match HF Space persistent storage (/data/models/)

Rename Start Simulation button to Start inference.py

Remove upload guides from docs; add GitHub repo link to Submission Resources

Document all frontend panels, OpenEnv log lines, and add GitHub/HF upload guides

Fix UI scroll/layout, SLA/budget tracking, counterfactual, sandbox status; add demo guide and local model sections to docs

Run all 3 tasks sequentially (easy/medium/hard), fix RCA tables, clean refresh state, update repo structure

Fix graphs: broadcast reward events; move FinOps Summary to full-width bottom bar

Add training plots, update Submission Resources with HF file links in README and BLOG

Add Start Simulation overlay with blur, re-shows on Clear

Add BLOG.md

Add kaggle training/visualization scripts and remove outdated docs

Add inference logs to HF Logs tab and fix idle timer/budget drift

Fix: unset PORT for llama-cpp subprocess to prevent port 7860 collision

Fix: add libgomp1 required by llama-cpp-python

Fix MODEL_REPO_ID default to aryxn323 (actual HF username)

Fix: strip newlines from MODEL_REPO_ID and MODEL_FILENAME secrets

Speed up build: use pre-built CUDA wheel for llama-cpp-python, switch back to runtime image

Fix: use CUDA devel image so nvcc is available for llama-cpp-python build

Initial Space deployment with llama-cpp + React dashboard