Spaces:
Runtime error
Runtime error
|
Download README.md from Majen/recalltrace-openenv: direct link, hf CLI and curl.
- Browser
- Download file 2.93 kB
-
https://huggingface.co/spaces/Majen/recalltrace-openenv/resolve/main/README.md
- Command line
-
hf download hf://spaces/Majen/recalltrace-openenv/README.md
-
curl -L -o README.md https://huggingface.co/spaces/Majen/recalltrace-openenv/resolve/main/README.md
2.93 kB
metadata
title: RecallTrace OpenEnv
emoji: π¨
colorFrom: red
colorTo: blue
sdk: docker
app_port: 7860
π RecallTrace OpenEnv
RecallTrace is a real-world AI environment designed for product recall tracing and precision containment.
It simulates how companies handle:
- contaminated product recalls
- supply chain tracing
- selective quarantine decisions
This environment evaluates agent reasoning + decision-making, not just correctness.
π§ What This Environment Does
Given a recall notice (e.g., "Lot A is contaminated"), the agent must:
- Trace where the product went
- Identify affected nodes (warehouses, stores)
- Handle relabeling / transformations
- Quarantine only unsafe inventory
- Avoid blocking safe stock
- Notify affected entities
- Finalize with correct containment
π― Why This Is Important
This is a real industry problem seen in:
- food recalls
- pharma defects
- logistics failures
Challenges include:
- Graph traversal
- Partial observability
- Lot transformations
- Mixed inventory reasoning
- Precision decision-making
π§© Tasks (Scenarios)
πΉ Easy β Direct Recall
- Single contaminated lot
- Straight supply chain
- Goal: trace and quarantine correctly
πΉ Medium β Relabeled Inventory
- Lot gets renamed (LotA β LotA1)
- Goal: track transformations and quarantine
πΉ Hard β Mixed Inventory
- Contaminated + safe stock mixed
- Goal: isolate unsafe quantity without over-blocking
βοΈ Action Space
| Action | Description |
|---|---|
| inspect_node | View inventory at a node |
| trace_lot | Follow product lineage |
| quarantine | Block unsafe stock |
| notify | Inform affected nodes |
| finalize | End task |
π¦ Observation Structure
Each step returns:
- recall_notice
- inventory
- action history
- trace results
- inspection data
π Reward & Grading
Reward System
- Correct tracing
- Correct quarantine
- Correct notification
- β Wrong node
- β Over-quarantine
- β Missed unsafe stock
Final Score
Range: 0.0 β 1.0
Based on:
- accuracy
- precision
- efficiency
π§± Project Structure
recalltrace-openenv/
β
βββ env/ # Environment logic
β βββ env.py
β βββ __init__.py
β
βββ scenario/ # Scenario generation
β βββ scenario.py
β
βββ grader/ # Evaluation + reward
β βββ grader.py
β
βββ inference/ # Agent simulation
β βββ inference.py
β
βββ config/
β βββ openenv.yaml
β
βββ Dockerfile
βββ requirements.txt
βββ README.md