fix: benchmark fields use None instead of 0.0 when no improvement — verified by exhaustive test

fix: eliminate ALL exact 0.0/1.0 values across entire codebase — baseline_results.json, schemas, reward breakdown, penalty defaults

fix: use 0.01/0.99 bounds — 0.001/0.999 printed as 0.00/1.00 with 2dp format

fix: remap all reward scores to strictly open (0,1) — root cause of Phase 2 failure

fix: ensure no score boundary values (0.0/1.0) anywhere in grading pipeline

fix: clamp task scores to strictly open interval (0.001, 0.999) per Phase 2 validator

fix: rewrite inference.py to match official [START]/[STEP]/[END] stdout format exactly

fix: add [START]/[STEP]/[END] structured output with flush=True for Phase 2 validation

fix: make /reset accept empty body for validator compatibility

fix: inference.py env vars and START/STEP/END log format per hackathon checklist

final changes

improved the ui of carbon saved card

Adding 7 more tasks

fix: ui logic, noise thresholds and identical code bypassing

feat: complete UI rebuild, isolated benchmarks, openenv schema fixes

sync: update openenv.yaml and README with all 8 tasks

fix: add CORS and X-Frame-Options for HF Spaces iframe embedding

fix: remove tmp file

fix: slim deps for faster startup

feat: premium light-mode dashboard with glassmorphism and live task loading

fix: remove incompatible Python 3.12 pycache bytecode

fix: force LF via gitattributes and push missing server dashboard changes

fix: pristine HF space docker compatibility

fix: explicit SDK Docker metadata for HF Spaces build

initial commit