SyncShift/sql-correction-env

Updated grader values

B

Fix stagnation penalty floor to 0.01, never 0.0

Fix grader: clamp rewards to strictly (0.01, 0.99)

Fix server.py, expand tasks to 40, add tests, clean observation schema

fix indentation in main loop

fix indentation in main loop

match reference env pattern exactly

simplify tasks in openenv.yaml to match reference format

fix state as method not property, fix abstract class error

fix step crash when current_task is None

fix initial reward from 0.0 to 0.001

fix reset indentation and accept task_id param

fix reset to accept task_id, update tasks endpoint

fix max_concurrent_envs to 1

switch to create_app from http_server

fix step format, add tasks endpoint back

fix inference to handle new reset response format

fix create_fastapi_app call remove env_name

restore SQLTask and SQLReward to models.py

use openenv base classes and fix yaml format to match working submissions

fix openenv.yaml with spec_version, type, runtime, app, port

add grader endpoint

add grader endpoint, fix reward bounds and openenv.yaml

remove version pins to fix dependency conflicts

fix openai version conflict with openenv-core

pin openenv-core version to force reinstall

add grader_fn to tasks endpoint

revert to working server

add openenv-core to requirements

use openenv core base classes

force rebuild

commit missing openenv.yaml and server.py changes

fix grader serialization and tasks endpoint

fix /tasks endpoint format for validator

fix reward range to strict (0,1) exclusive

fix server entry point

Make inference fallback to offline SQL correction

Fix server entrypoint packaging and align environment docs

Fix inference startup and healthcheck compatibility

force clean exit code 0

error handling in inference.py

fix state endpoint, author, and init exports

fix remotes and remove stray folder

update ENV_URL to SyncShift space

fix server entry point

fix openenv validate issues

merge and add pyproject.toml

add pyproject.toml

Fix punctuation in README.md