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 about scientific identification under uncertainty. Agents choose costly, noisy experiments across generative models, symmetry, topology, neural geometry, and conformal uncertainty, then transfer the inferred mechanism to sealed conditions—a compact test of mathematical reasoning, planning, and judgment.
A five-environment mathematics suite about scientific identification under uncertainty. Agents choose costly, noisy experiments across generative models, symmetry, topology, neural geometry, and conformal uncertainty, then transfer the inferred mechanism to sealed conditions—a compact test of mathematical reasoning, planning, and judgment.
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 |
|---|---|---|
| DiffusionLens | Analytic probability and score-model geometry | Choose probes across locations and noise scales to distinguish generative mechanisms, then predict behavior under sealed queries. |
| SymmetryHunter | Finite-group representations and equivariance | Audit subtle symmetry breaking using transformations and compressed sensors whose individual nullspaces can hide the defect. |
| PersistentAtlas | Persistent homology and multiparameter filtrations | Balance coarse invariants against localized slice measurements before transferring inferred topology to unseen directions. |
| TropicalCircuit | ReLU arrangements and tropical geometry | Reverse-engineer piecewise-linear structure from line probes complicated by cancellation, sign loss, and overlapping boundaries. |
| ConformalSleuth | Finite-sample conformal prediction | Diagnose latent scoring effects through thresholds, coverage, set sizes, and compressed membership behavior on selected panels. |
A GPT-5.6 Pro tool-assisted, truth-blind exact-Bayes policy earned 0.8784 mean reward but only 3/5 strict passes on untouched expert episodes. It achieved 0.9415 held-out prediction, yet misses on SymmetryHunter and PersistentAtlas show that prediction alone cannot substitute for identification, calibration, truth mass, experiment design, and stopping. The package reports 61/61 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.
Reported result from the evaluation artifact supplied with this package.
As identified by the supplied artifact.
5 reported runs.
5Math-3-results.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
952a492e9d26c3a6aff02ff3dfc791d934433aa6354c200cf721443c80bfb5d7One 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.