Add loss curve plot (required for automated validation)

Update blog + Run 2 real data

Submission ready: narrative README, real reward plot, clean notebook, remove internal files

docs: add competitive analysis of ~250 OpenEnv hackathon repos

Submission: clean public repo - remove internal docs, add notebook and results

Submission: notebook, README, blog post, reward plot

V4: steep reward slope - full-episode eval + stronger milestones + progress bonus

fixes : rewards and training

Fix GRPO parse collapse for notebook training

Fix reward-stationary reset controls and notebook training setup

fix file structure

V4: fix 5 critical + 8 major + 2 moderate issues

Update prompt.md

Add evaluation results and warnings to prompt.md

Add GRPO V3 training guide: explains bug, fix, and testing steps

Add detailed training issues and fixes documentation

fix: rebalance rewards for GRPO slope - remove efficiency baseline, add milestone bonuses

docs: capture colab validation run and hf plan

docs: prep training intake and handoff workflow

Add comprehensive RL book for beginners (16 chapters, docs/internal/rl_book)

fix: add root route so HF Space doesn't show 404

merge PR#15 (S1-S7), integrate extra_info into hack_info+ROADMAP, reset reward CSV

Merge branch 'Roopalgn:main' into main

[Push 8 Pre-25] S1-S7: action space docs, onsite checklist patch, eval key rename, 7b lora_r fix, dry-run outputs

Create extra_info.md

pre-25th R1-R8: add innovation argument, training-failure fallback, notebook validation outputs

hack_info: restructure with proper markdown; roadmap: add pre-25th task list

roadmap: add competitor analysis, validate 6 code gaps, add P0 fix plan for onsite

Update hack_info.md

docs: consolidate 28->16 md files, refine README, add eval report

fix: train_colab.ipynb - correct API, DRY_RUN, MODEL_PRESETS, reward handling; mark Suyash tasks done

prep: onsite artifacts - kaggle notebook, checklist, templates with [FILL ONSITE] placeholders

docs: add pre-onsite checklist + update KnowledgeBase

fix: remove all ROADMAP contradictions - training is onsite only

Update ROADMAP: mark branch merge complete, update Phase B checklist

Update ROADMAP: re-enable pre-training on Kaggle, Push 8 becomes H100 refinement

Add internal resources document

[Push 7] Roopal: fix HF Space card metadata, align port to 7860, add training_log.md

Push 7 Phase A: grounding + kaggle notebook + docs

Update ROADMAP: pre-training on Kaggle before onsite, Push 8 becomes H100 refinement

Polish repo for judges: move internal docs, clean winner refs, improve dashboard

[Push 7-8] Add G16-G22 post-merge gaps, Push 7 (pre-onsite) and Push 8 (onsite training), updated checklists and Definition of Done

post-merge: add .gitignore, update project status with integration test results

[Push 6] Roopal: storytelling assets, pitch notes, mini-blog finalized, ARCHITECTURE.md finalized, docs proofread

[Push 5] Roopal: reward tuning, adaptive difficulty spec, dashboard UI, improved prompts

[Push 4] Roopal: training runbook, Colab notebook, evaluation template, mini-blog draft

[Push 3] Roopal: curriculum policy, verification spec, benchmark protocol, dashboard metrics, phase scoring

[Push 2] Roopal: reward spec, scenario cards, milestone map, shaping function

[Push 1] Roopal: enhance all docs with explicit winner-inspired patterns (KubeSRE, Bio, VRAM)

[Push 1] Roopal: rewrite KnowledgeBase as progressive textbook, add change log to project status