Agent objective
- Interact with the supplied stateful environment.
- Produce verifier-checkable actions or artifacts.
- Maximise scalar reward under the package contract.
A five-environment mathematics suite in which agents uncover hidden mechanisms through budgeted experiments across optimization dynamics, kernel geometry, attention operators, Boolean Fourier structure, and entropic transport. Noisy compressed observations and sealed transfer checks reward informative measurement choices and calibrated uncertainty rather than static answer recall.
A five-environment mathematics suite in which agents uncover hidden mechanisms through budgeted experiments across optimization dynamics, kernel geometry, attention operators, Boolean Fourier structure, and entropic transport. Noisy compressed observations and sealed transfer checks reward informative measurement choices and calibrated uncertainty rather than static answer recall.
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 |
|---|---|---|
| OptimizerFingerprint | Convex quadratics, dynamical systems, preconditioning | Design complementary trajectory measurements that reveal hidden training dynamics without wasting the experiment budget on redundant regimes. |
| KernelCipher | RKHS geometry, kernels, Gaussian-process uncertainty | Select kernel functionals that separate subtly different latent geometries when simple pairwise measurements leave explanations aliased. |
| AttentionMicroscope | Softmax operators and harmonic analysis | Vary content, position, temperature, and compressed spectral readouts to infer a hidden positional mechanism without seeing the full attention map. |
| BooleanSpectrometer | Boolean Fourier analysis and active learning | Construct restricted sampling experiments that expose hidden high-order structure through noisy low-order observables. |
| TransportDecoder | Entropic optimal transport and Sinkhorn geometry | Choose marginals, regularization regimes, and coupling statistics that distinguish latent transport costs concealed by aggregate measurements. |
This suite separates adaptive mathematical reasoning from answer recall. In the supplied 40-episode expert evaluation, a tool-assisted exact-likelihood policy reached 0.8418 mean reward but only 22/40 strict passes; performance ranged from 8/8 passes on AttentionMicroscope to 2/8 on KernelCipher. The spread shows that its five domains exercise materially different experiment-design skills. The release passed 60/60 automated 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.
The report identifies this as a source-informed or executable-policy result, not a clean black-box benchmark.
As identified by the supplied artifact.
40 reported runs.
5Math-1-run.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
39bb5cbcc3dd6059d96a871090adc5b6441bb8ff240218957bfc4d0a633611f6One 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.