GameAgent-Recovery-100K

A deterministic synthetic benchmark of 100,000 game-agent failure and recovery transitions.

Each record contains:

state ? failed action ? failed state ? recovery action ? recovered state

Summary

• 100,000 records
• 10 failure families
• 100,000 unique IDs
• 100,000 unique fingerprints
• deterministic recovery oracle
• no copyrighted gameplay
• no player recordings
• no PII
• no LLM-generated ground truth

Failure families

• wrong_target
• navigation_loop
• missing_prerequisite
• premature_interaction
• resource_exhaustion
• low_health_attack
• locked_door
• missing_tool
• failed_extraction
• wrong_sequence

Why this dataset

Autonomous agents need to do more than select good actions.

They must detect when progress has failed and recover.

This dataset explicitly represents failure transitions and the corrective action required to restore progress.

Example

{
  "state_before_failure": {
    "has_key": false,
    "door_locked": true
  },
  "failed_action": "open_door",
  "recovery_action": "pickup_key",
  "labels": {
    "failed_action_correct": false,
    "recovery_action_correct": true
  }
}
Validation
Independent deterministic validator:
- 100,000 / 100,000 PASS
- unique IDs checked
- fingerprints checked
- recovery action recomputed
- recovered state recomputed
- failed/recovery action separation checked
- provenance checked
Ground truth
Correct recovery actions are produced by explicit state-machine rules.
An LLM does not determine the correct recovery.
Intended use
- autonomous game agents
- agent recovery evaluation
- reinforcement learning
- hierarchical planning
- failure detection
- imitation learning
- reward-model evaluation
- embodied AI
- agent regression tests
Commercial customization
Need failure/recovery data for your own environment?
Custom datasets can be adapted to:
- your action API
- your simulator
- your state representation
- known production failures
- recovery chains
- tool-use errors
- navigation failures
- inventory/prerequisite systems
- private held-out evaluations
Typical delivery:
Your environment ? synthetic failures ? recovery oracle ? validator ? QA report
License
MIT.

---

## RegalFire — Custom / Private Dataset Work

RegalFire builds custom AI datasets, evaluation sets and data pipelines for:

- AI agents
- computer-use systems
- multimodal models
- world models
- RAG systems
- code agents
- enterprise AI

Available services include:

- synthetic data generation
- private evaluation datasets
- agent trajectories
- failure / recovery datasets
- multimodal RGB / segmentation / state-action data
- web data acquisition
- cleaning and deduplication
- structured dataset packaging
- continuous dataset production

For custom or private work:

**Email: ootiris@gmail.com**

Hugging Face: **RegalFire**