Agent objective
- Interact with the supplied stateful environment.
- Produce verifier-checkable actions or artifacts.
- Maximise scalar reward under the package contract.
Five adaptive test-time-computation environments in worlds that never pause. Every action consumes latency before interpretation, allowing evidence, rules, utilities, network conditions, or identities to change while the agent reasons. Across memory, planning, sensor fusion, network coding, and exact-cover assembly, success depends on what to inspect, stabilize, validate, and commit before a solution becomes stale.
Five adaptive test-time-computation environments in worlds that never pause. Every action consumes latency before interpretation, allowing evidence, rules, utilities, network conditions, or identities to change while the agent reasons. Across memory, planning, sensor fusion, network coding, and exact-cover assembly, success depends on what to inspect, stabilize, validate, and commit before a solution becomes stale.
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 |
|---|---|---|
| MemoryVault | Lossy streaming memory | Choose which records to inspect or preserve as evidence expires, corrupts, aliases rotate, and later questions target history. |
| NormStorm | Dynamic temporal/deontic planning | Build a compliant schedule while priorities, expiries, and amendments invalidate earlier reasoning. |
| SignalSiege | Bayesian sensor fusion under drift | Select and calibrate sensors while incidents cycle, groups spoof, and response utility changes. |
| PacketAlchemy | GF(2) network coding | Balance telemetry, path search, redundancy, and reservation as topology, latency, and capacity revise. |
| MosaicFlux | Dynamic exact cover | Assemble all pieces while aliases rotate, seam signatures mutate, clues activate, and cached placements stale. |
A supplied controller scored 27.41 with 20% exact success across 100 public standard episodes, beating greedy by 6.45 points, but fell to 13.26 on stress. Fixed extra thought harms the greedy baseline, dropping it from 20.96 to 14.11 after four short-think actions. A delayed oracle passes all standard cases, establishing solvability under the same mandatory latency. These remain public development results; the release reports 67 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.
Public development standard suite. Agent success rate was 20%, versus 4% for the baseline.
As identified by the supplied artifact.
100 reported runs.
redqueen-run-bundle.zip:score-summary.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
bf670375e73b8c7f77bd11208427824b9c62f2459ba093ec947c8893cec7bc41One 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.