spec_version: 1
name: honest_env
type: space
runtime: fastapi
app: server.app:app
port: 8000
description: >
  HONEST — Honesty-Optimised and Normalized Environment for Self-Triage.
  An OpenEnv-compliant calibration benchmark where LLM agents must emit
  an answer together with a self-reported confidence (or explicitly
  abstain). Reward is the Brier score, a strictly proper scoring rule.
  Three domains (math, code, logic) at five difficulty levels.
license: BSD-3-Clause
homepage: https://github.com/Rushhaabhhh/HONEST-RL-Calibrator
tags:
  - openenv
  - calibration
  - rl
  - grpo
  - mcp
  - honesty