Delete server/__pycache__

Update model references: Llama 3.3 70B → Llama 3.1 8B Instant, runtime 19.8min → ~10min

Update greedy baseline numbers from latest test_local run; recalculate lifts

Fix Logs: Updated both Log Outputs

Fix duplicate WebSocket log output; update benchmark numbers to latest run (avg 75.9%, ~10min)

Expose data_lag_days in CityObservation; remove fragile num_districts==6 inference

Reduce rollouts easy=2 medium=3 hard=3 — target ~15min runtime with buffer for judge overhead

Fix 8 issues spotted in code review: Dockerfile path, task descriptions, duplicate function, fallback score consistency, twin constants comment, has_data_lag docs

Update easy rollouts 3→2 everywhere: app.py scores, README

Fix All: Cleaned All Files

Fix all hackathon validator errors: pyproject.toml deps+scripts, app.py main(), inference.py per-task rollouts

Update all dashboard tabs and README with actual test run numbers

Fix orphaned Phase 3 comment text rendering visibly on all tabs

Phase 2: replace live demo with static greedy + LLM+GRPO results from actual test runs

Fix Test Local: Improved to Work Better

Fix Env: Changed Token

Fix easy task seed D0↔D1 so dumb greedy targets wrong district; fix signal thresholds

Remove smart greedy — benchmark is dumb greedy vs LLM+GRPO only, matching judging criteria

Phase 2 dashboard: 3 evals (greedy live + GRPO benchmark + variance check), comprehensive test_local.py

Fix Pylance reportInvalidStringEscapeSequence: replace regex \\b\\w with split/map capitalise

Rewrite app.py: 3-phase judge dashboard, /validate endpoint, live Phase 1 checks

Rewrite app.py: 3-phase judge dashboard, /validate endpoint, live Phase 1 checks

Fix Readme: Corrected Grader Percentages

Fix restriction auto-lift, density-weighted penalty, /info endpoint accuracy, updated README

Fix grader weights hospital=0.40/containment=0.35, fix memory threshold in update_memory

Raise medium seeds so hospital score reflects outbreak severity

Fix: D0 seed above threshold, faster hospital drain, containment as primary metric

Fix treatment/recovery calibration, hospital breach reward, efficiency grader, recalibrate seeds

Fix realism: natural recovery in spread, hospital breach at 10%, linear spillover, correct prompt

Lower spread to 0.08, fix efficiency grader to account for treatment reduction

Swap grader weights: hospital=45% primary constraint, containment=30%, calibrate environment

Lower spread rate to 0.09, fix medium seeds, teach focused allocation strategy

Raise seeds to prevent premature termination on medium and hard

Raise treatment reduction, lower medium/hard seeds, focus prompt strategy

Fix speed score bug, restore partial resource replenishment, raise easy seed

Fix resource replenishment, grader efficiency, seed calibration, prompt strategy

Fix 1-step termination bug, fix done field in trajectory, recalibrate seeds

Fix core environment solvability: add treatment reduction, calibrate seeds and spread rates, fix hospital grader denominator

Professional judge dashboard with 5 tabs, live demo, grader breakdown, baseline scores

Add interactive dashboard with live demo for judges

Calibrate grader: soften hospital breach penalty, grace period for containment, fix speed score

Add /grade endpoint, use real grader scores in evaluator

Reduce hard max_steps to 15 to prevent WebSocket timeout

Fix App tab HTML, remove broken iframe link

Add root HTML endpoint for App tab display

Add README.md

Initial submission: Cascade Containment