docs: blog edits

train: tighten Colab notebook (drop redundant cell, simplify section 5)

train: simplify Colab notebook (drop redundant cells)

ui: clear stale session/transcript when user switches model preset

cleanup: trim verbose comments, drop dead code, fix stale tests, proper Dockerfile + .gitignore

frontend: sync clean source — drop ComparePane + Compare LLMs tab + Which-endpoint copy; ship 3-preset picker

blog: restore plot images + fix unclosed code span in reward-hacking table

blog: add reward-hacking section (3 patched exploits + 3 by-construction invariants)

Dockerfile: bump SPA cache-bust to 10 — force fresh build (drops Compare LLMs tab + stale endpoint copy)

docs: README only lists 3 trained systems; drop tier-3 held-out claims and stale example

docs: drop per-component reward breakdown chart and bullets

docs: drop per-component reward breakdown chart and bullets

W&B disclaimer: point users to their own account, drop hardcoded entity URL

W&B disclaimer: point users to their own account, drop hardcoded entity URL

W&B disclaimer: point users to their own account, drop hardcoded entity URL

notebook: W&B disclaimer points users to their own account, not the author's

notebook: add W&B disclaimer banner before SFT and GRPO subprocess launches

notebook: visible pip install + sys.path refresh + import verify

notebook: pull from Space repo (not stale dataset), default profile=3b

openenv push: sync full repo

ui: default damped_spring, remove Random option

ui: default damped_spring, remove Random option

ui: default damped_spring, remove Random option

ui: default damped_spring, remove Random option

Upload folder using huggingface_hub

blog: clean pass + future scope section

registry: SUPPORTED_SYSTEMS = only 3 trained systems

blog: only 3 confirmed trained systems from W&B

registry: remove tier3 from SUPPORTED_SYSTEMS; mark as TODO

blog: show all 6 systems honestly; GRPO=3, SFT=6, no held-out claims

cleanup: remove writeup.md (superseded by blog.md)

blog: expand system tiers table, add extensibility section

Delete .vscode

Upload folder using huggingface_hub

cleanup: strip verbose comments from train/job_train.py

cleanup: strip verbose comments from physix/models.py

cleanup: strip verbose comments from physix/server/app.py

cleanup: strip verbose comments from physix/server/providers.py

cleanup: strip verbose comments from physix/training/dataset.py

cleanup: strip verbose comments from physix/training/sft.py

cleanup: strip verbose comments from physix/training/prompt.py

cleanup: strip verbose comments from physix/training/reward_fns.py

cleanup: strip verbose comments from physix/training/loop.py

Add SFT loss curve plot

Add blog.md: accurate systems table, SFT+GRPO plots

Update blog link to blog.md

Fix writeup links to absolute HF URLs

Fix links table: absolute URLs for blog, notebook, checkpoint

Sync docs/writeup.md from latest training run

Update plots, fix README table fmt, update training result observations