Agent objective
- Interact with the supplied stateful environment.
- Produce verifier-checkable actions or artifacts.
- Maximise scalar reward under the package contract.
A long-horizon wildfire-command environment for planning under partial observability, uncertain weather, failing roads, limited resources, and competing public-safety goals. Centralized or cooperative policies coordinate engines, bulldozers, helicopters, scouts, and evacuation buses across procedural maps while balancing containment, evacuation, assets, ecology, responder safety, equity, cost, and forecast shifts.
A long-horizon wildfire-command environment for planning under partial observability, uncertain weather, failing roads, limited resources, and competing public-safety goals. Centralized or cooperative policies coordinate engines, bulldozers, helicopters, scouts, and evacuation buses across procedural maps while balancing containment, evacuation, assets, ecology, responder safety, equity, cost, and forecast shifts.
Enough detail to understand the intellectual terrain; generated instances, hidden mechanisms, and solution paths remain inside the private package.
| Environment | Mathematical or technical frontier | Adaptive research problem |
|---|---|---|
| Fast-moving fire | Urban interface, canyon wind, crosswind shift | Anticipate spread, protect routes and assets, and reposition before partial observations or forecasts become stale. |
| Evacuation pressure | Gridlock, hospitals, vulnerable populations | Coordinate buses, scouts, and protective units while optimizing survival equitably under smoke and road constraints. |
| Fragmented incidents | Lightning clusters and river crossings | Allocate scarce aerial and ground resources across simultaneous fronts and terrain-separated regions. |
| Degraded awareness | Sensor blackout and night operations | Spend limited capacity on sensing while acting safely under stale fire, smoke, road, and visibility information. |
| Resource and asset conflict | Scarcity and critical infrastructure | Balance water, fuel, fatigue, budget, recovery, containment, and high-value facilities without abandoning public safety. |
A frozen observation-only heuristic averaged 0.6095 across 60 private episodes, beating the bundled greedy response by 42% and winning 54/60 episodes. It greatly improved containment and evacuation while preserving safety, assets, and ecology—but used roughly twice the operational cost and retained a hard worst-case tail. That tradeoff makes the environment useful for multi-objective command rather than simplistic fire suppression. The technical beta reports 17/17 tests.
We publish aggregate behavior and task structure, while withholding generated instances, hidden labels, exact successful probes, private checks, and solution trajectories.
Shown with its provenance and limitations; it is not a performance guarantee.
Private safe-summary result across 12 scenarios. This is a bundled baseline, not a language-model rollout.
As identified by the supplied artifact.
60 reported runs.
crisisgrid_safe_results_bundle.zip:private_controls_summary_safe.json
Machine-readable provenance and the exact displayed metric are available in results.json.
The paid ZIP will live in a private R2 bucket. Vercel authorizes the buyer and issues a 2–5 minute object URL; R2 serves the bytes directly.
Authenticated buyer + entitlement check
+ private R2 object + 2–5 minute signed URL
= direct, auditable download
Package SHA-256
8c32eaed3589fa4a9e632eda191a73e521cbe36b03d3835a44228af39e3fa30fOne purchase licenses this identified item to one legal organisation for worldwide, perpetual commercial model training, evaluation, research and development. Redistribution and resale of the package are not permitted.