Agent objective
- Interact with the supplied stateful environment.
- Produce verifier-checkable actions or artifacts.
- Maximise scalar reward under the package contract.
Five research-tier mathematical investigations built around hidden systems with deceptively identical public invariants. Across cellular sheaves, non-normal dynamics, algebraic geometry, quantum channels, and tensor orbits, agents must spend a severe measurement budget, distinguish competing mechanisms, calibrate their posterior, and generalize beyond chosen probes.
Five research-tier mathematical investigations built around hidden systems with deceptively identical public invariants. Across cellular sheaves, non-normal dynamics, algebraic geometry, quantum channels, and tensor orbits, agents must spend a severe measurement budget, distinguish competing mechanisms, calibrate their posterior, and generalize beyond chosen probes.
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 |
|---|---|---|
| SheafGauge | Cellular sheaves, holonomy, graph diffusion | Choose signed, energetic, cycle, and spectral probes that cut across overlapping aliases in a sheaf-valued system. |
| KoopmanMirage | Non-normal operators, resolvents, pseudospectra | Diagnose dynamics with shared eigenvalues using finite-time, transient, resolvent, and Hankel measurements. |
| IdealNavigator | Finite-field algebraic geometry | Infer a hidden polynomial transformation from two compressed algebraic probes, then transfer it to unseen polynomials. |
| QuantumProcessLens | Quantum channels and compressed tomography | Separate ordered coherent and stochastic effects through phase-sensitive, nonlinear measurements under incomplete tomography. |
| TensorOrbit | Tensor invariant theory and group actions | Recover orbit coordinates when standard norms and spectra are invariant, using contractions that retain sign and order. |
A public-prompt Bayesian controller earned 0.8213 mean reward across 40 expert episodes but only 17 strict passes. Held-out prediction averaged 0.9326 while correct identification reached 62.5%, revealing the intended gap between locally good experiments and sufficiently concentrated inference. The package's exact-likelihood reference itself passed only 67/200 expert cases; the release records 61/61 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.
Reported result from the evaluation artifact supplied with this package.
As identified by the supplied artifact.
8 reported runs.
ulam_rlvr_score_report.md
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
84977548ec9ebf5bc456419d633b0502c195963a5dea0b9ba6418c82f71fefdfOne 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.