rushi07ai/email-triage-openenv

Update README and openenv.yaml with real scores and 20-email episodes

Add 10 more emails (E011-E020), harder grading, 20 email episodes

Fix [END] format: task= score= steps= with finally block

Final fix: all rewards strictly 0.1-0.9, no 0.0 or 1.0 anywhere

Reinstate score clamping to 0.05..0.95

Merge pull request #1 from Ram-coded/feature

New Changes

Force score defaults and ranges within (0,1)

Harden score clamping to 0.05..0.95

Clamp scores to 0.01..0.99 to satisfy validator

Clamp all scores strictly within (0,1)

Clamp scores within (0,1)

Fix inference.py: clamp all rewards in [START][STEP][END] logs - never 0.0 or 1.0

Fix penalties field - never exactly 0.0

Clamp rewards in inference.py stdout logs - never 0.00 or 1.00

Complete rewrite: scores naturally in (0.05, 0.95) - never 0.0 or 1.0

Clamp all reward and score fields in API responses - never 0.0 or 1.0

Clamp all reward and score fields in API responses - never 0.0 or 1.0

Update baseline agent

Fix graders indentation and update deps

Fix all scores strictly between 0.01 and 0.99 - no exact 0.0 or 1.0

Fix Dockerfile: use python:3.11-slim-bookworm for stable builds

Fix grader scores to be strictly between 0 and 1 (exclusive)

Fix /reset to accept empty body - default task_classify

Fix HF_TOKEN to have no default value per submission checklist

Clean up structure - tests in tests/ folder

openenv validate passing - ready for submission

Initial commit

initial commit