This is the review map for the 38 method-guide pages in the site. It answers three different questions without conflating them: what a learner can study, what code is implemented, and what evidence supports a bounded claim.
implemented means a public Jaxstro owner exists. experimental means the
public owner exists but is outside the qualified scientific core. planned
means the page teaches a proposed or delegated capability, not an importable
runtime. A passing row supports a numerical contract; it does not establish
scientific-model adequacy.
How to use the matrix¶
Start with a question, make a prediction with explicit units and limiting behavior, choose a row, then follow its method, API, and validation links:
predict -> compute -> audit -> state the warranted claim| Learning family | Runtime status | Canonical owner or boundary | First learning route | Evidence and limitation route |
|---|---|---|---|---|
| Scalar constraints and sensitivities | implemented | jaxstro.numerics roots and implicit roots | Root-finding | Qualified scientific core; value-first roots do not claim implicit derivatives |
| Fixed-step change and optimization | implemented | jaxstro.numerics ODE and objective helpers | Fixed-step ODE integration, Optimization helpers | Validation; adaptive ODE and general nonlinear solving remain delegated |
| Differentiation products | implemented | jaxstro.numerics autodiff helpers | Autodiff products | 1. Classify the transform contract first; route changes and discrete choices are not smooth claims |
| Tables, splines, and regular grids | implemented | jaxstro.numerics interpolation owners | One-dimensional interpolation | Validation; gradients are local to documented smooth cells |
| One-dimensional quadrature and cumulative rules | experimental | jaxstro.quad and legacy numerical helpers | Fixed and weighted quadrature | quad |
| Multidimensional finite-hyperrectangle integration | experimental | jaxstro.quad tensor, cubature, sparse-grid, and QMC methods | Choosing a multidimensional integration method | Multidimensional quadrature evidence; no geometry beyond finite hyperrectangles or universal performance claim |
| Linear structure and special functions | implemented | jaxstro.numerics linear algebra, operators, and bases | Linear algebra helpers | Validation; rank and conditioning boundaries remain explicit |
| Probability, keys, and resampling | implemented | jaxstro.numerics distributions plus jaxstro random helpers | Random computation | Validation; discrete resampling is not assigned an invented physical derivative |
| Grids, meshes, and spatial candidates | implemented | jaxstro.numerics grids/meshes and jaxstro.spatial | Grids and conservative bin transfer | Validation; candidate generation is distinct from physical interaction policy |
| Signals and spectral interpretation | planned | No jaxstro.signal runtime owner | Signal axes, cadence, and units | Future modules and capabilities roadmap; these guides teach required conventions, not a shipped API |
Coverage boundary¶
The generated contract registry has callable-level records only where an owner has registered purpose, transforms, failure behavior, evidence, and limitations. It currently lists 235 unclassified public callables. They may be importable, but this matrix deliberately does not promote them into supported methods and does not establish scientific-model adequacy.
For the small set that is both public and evidence-complete, use the Qualified scientific core. For all public module owners, use the API reference. For development priorities, limitations, and proposed work, use the Development log, Package and documentation SOTA assessment, and Future modules and capabilities roadmap.