simplify openenv.yaml: pure ASCII, remove all non-essential sections, copy yaml to WORKDIR in Docker

Dockerfile

@@ -24,6 +24,9 @@ RUN pip install --no-cache-dir -r requirements.txt
 # Copy project source
 COPY . /app/payops_env
 
+# Also copy openenv.yaml to WORKDIR so validators can find it at /app/openenv.yaml
+COPY openenv.yaml /app/openenv.yaml
+
 # Both /app (for payops_env.*) and /app/payops_env (for server.*) on PYTHONPATH
 # This matches the openenv.yaml app: server.app:app path and platform example format
 ENV PYTHONPATH="/app:/app/payops_env"

openenv.yaml

@@ -1,132 +1,11 @@
 spec_version: 1
 name: payops_env
-display_name: "PayOps — Payment Operations Incident Response"
-description: >
-  An AI agent acts as a Payment Operations analyst, reviewing financial
-  transactions and deciding how to handle each one. Tasks range from
-  clear-cut fraud to subtle adversarial patterns like APP scams, insider
-  threats, structuring, and KYC violations across 4 difficulty tiers.
 type: space
 runtime: fastapi
 app: server.app:app
 port: 7860
-version: "2.0.0"
-author: "PayOps Team"
-tags:
-  - finance
-  - fraud-detection
-  - compliance
-  - real-world
-  - payment-operations
+version: 2.0.0
 
-# ── Endpoints (OpenEnv required) ────────────────────────────────────────────
-endpoints:
-  reset:   POST /reset
-  step:    POST /step
-  state:   GET  /state
-  health:  GET  /health
-  schema:  GET  /schema
-  tasks:   GET  /tasks
-  grader:  GET  /grader
-  replay:  POST /replay
-  baseline: POST /baseline
-  analytics: GET /analytics
-  leaderboard: GET /leaderboard
-
-# ── Environment variables required at runtime ────────────────────────────────
-env_vars:
-  OPENAI_API_KEY:
-    description: "API key for OpenAI-compatible inference endpoint (takes precedence over HF_TOKEN)"
-    required: false
-    secret: true
-    example: "sk-..."
-  API_BASE_URL:
-    description: "OpenAI-compatible API endpoint for LLM inference"
-    required: true
-    example: "https://api.openai.com/v1"
-  MODEL_NAME:
-    description: "Model identifier to use for inference"
-    required: true
-    example: "gpt-4o-mini"
-  HF_TOKEN:
-    description: "Hugging Face / API key used as Bearer credential (alias for OPENAI_API_KEY)"
-    required: false
-    secret: true
-
-# ── Action space ─────────────────────────────────────────────────────────────
-action_space:
-  type: discrete
-  terminal_actions:
-    - approve
-    - reject
-    - flag
-    - escalate
-    - hold
-  investigation_actions:
-    - inspect          # cost: 0.10 budget
-    - request_docs     # cost: 0.20 budget
-    - verify_kyc       # cost: 0.20 budget
-    - contact_sender   # cost: 0.30 budget
-    - file_sar         # cost: 0.05 budget
-
-# ── Observation space ─────────────────────────────────────────────────────────
-observation_space:
-  type: structured
-  fields:
-    # core transaction
-    - transaction_id
-    - task_id
-    - difficulty
-    - amount
-    - currency
-    - sender
-    - receiver
-    # risk signals
-    - risk_score
-    - ml_confidence
-    - flags
-    - kyc_status
-    - kyc_expiry_days
-    - velocity_1h
-    - velocity_24h
-    - country_risk
-    - account_age_days
-    - previous_sars
-    - counterparty_risk
-    # investigation reveals
-    - inspection_notes
-    - docs_notes
-    - kyc_notes
-    - contact_notes
-    # budget
-    - budget_remaining
-    - action_cost
-    # episode progress
-    - cumulative_reward
-    - chain_step
-    - chain_total
-    - steps_remaining
-    - done
-    - reward
-    - reward_breakdown
-    - recent_decisions
-    - network_graph
-
-# ── Tasks ───────────────────────────────────────────────────────────────────
-# Metadata kept separate so 'count: 30' still appears in file
-task_count: 30
-task_difficulties:
-  - easy
-  - medium
-  - hard
-  - critical
-task_difficulty_weights:
-  easy: 1.0
-  medium: 1.2
-  hard: 1.5
-  critical: 2.0
-
-# tasks: grader field points to a Python callable (module:Class format)
 tasks:
   - id: EASY-001
     name: EASY-001
@@ -249,40 +128,3 @@ tasks:
     difficulty: critical
     grader: server.graders:CRIT006Grader
 
-# ── Budget ───────────────────────────────────────────────────────────────────
-budget:
-  initial: 5.0
-  overspend_penalty_rate: 0.1
-
-# ── Reward ───────────────────────────────────────────────────────────────────
-# Trajectory-based reward (v2): rewards correct intermediate reasoning AND
-# the final terminal decision to provide a dense signal.
-reward:
-  # Terminal-action credit
-  terminal_correct: 0.60
-  # Investigation eligibility bonus (per eligible sub-action used, first use only)
-  investigation_eligible: 0.20
-  # Flag-identification bonus (agent used inspect AND task has decisive key_flags)
-  flag_identification: 0.20
-  # Confidence bonus when agent confidence >= 0.8 and final action is correct
-  confidence_correct: 0.10
-  confidence_wrong: -0.10
-  # Regulatory bonus for SAR filing on regulatory tasks before terminal action
-  regulatory_sar: 0.20
-  # Duplicate investigation penalty (same sub-action on same task, second+ use)
-  duplicate_investigation_penalty: -0.05
-  partial_credit: true
-  passing_threshold: 0.5
-  # Bounds: normalised episode score is strictly clamped to [0.0, 1.0]
-  min_score: 0.0
-  max_score: 1.0
-
-# ── Inference script ─────────────────────────────────────────────────────────
-inference:
-  script: inference.py
-  max_runtime_seconds: 1200
-  requires:
-    - API_BASE_URL
-    - MODEL_NAME
-    - HF_TOKEN
-

server/Dockerfile

@@ -24,6 +24,9 @@ RUN pip install --no-cache-dir -r requirements.txt
 # Copy project source
 COPY . /app/payops_env
 
+# Also copy openenv.yaml to WORKDIR so validators can find it at /app/openenv.yaml
+COPY openenv.yaml /app/openenv.yaml
+
 # Both /app (for payops_env.*) and /app/payops_env (for server.*) on PYTHONPATH
 # This matches the openenv.yaml app: server.app:app path and platform example format
 ENV PYTHONPATH="/app:/app/payops_env"