Skip to article frontmatterSkip to article content
Site not loading correctly?

This may be due to an incorrect BASE_URL configuration. See the MyST Documentation for reference.

Package and documentation SOTA assessment

The Jaxstro package assessment scorecard is the living graded companion to this ranked investment roadmap. It records current evidence, the deficiency preventing the next grade, and the promotion gate for each assessed dimension.

Here “state of the art” means unusually strong alignment among scientific scope, numerical contracts, JAX behavior, independent evidence, provenance, pedagogy, and maintenance cost—not the largest feature list. Maturity labels are ratified, validated, implemented, experimental, and planned; no numeric score is assigned without an observable rubric.

Delivered strengths

DimensionMaturityEvidence-backed assessment
Scientific breadthvalidatedUnits, constants, coordinates, geometry, numerics, spatial operations, spectra, atmospheres, parameter bridges, and provenance cover recurring lower-level needs.
Ownership disciplineratifiedADRs and architecture keep generic mechanics here and physical interpretation, retry policy, and scientific state downstream.
Numerical robustnessvalidatedRootfinding, interpolation, quadrature, linear algebra, distributions, meshes, and special functions expose boundary and failure contracts in Validation.
ConditioningvalidatedDenominator, slope-floor, covariance, positive-definite, and limiting-kernel checks are explicit.
AD honestyratifiedThe theory index classifies smooth, zero, blocked, surrogate, validation-only, value-first, and certified implicit paths.
JAX transform coveragevalidatedSelected jit, vmap, lax.map, grad, JVP/VJP, scan, and PyTree behavior is tested with named exceptions.
Dimensional safetyimplementedjaxstro.units is canonical; jaxstro.quantity implements exact dimensions and transforms without claiming ecosystem adoption.
API cohesionvalidatedExports, reference docs, typed results, and focused tests give downstream packages stable contracts.
SerializationimplementedScalar quantities have explicit round trips and array rejection; provenance records render deterministically; root telemetry uses fixed-shape NamedTuples suitable for transport, but no public root round-trip/replay serializer is yet validated.
Performance and compilation evidenceimplementedRootfinding and spectra now share a units-explicit evidence envelope; compile/runtime/memory coverage is not yet uniform.
Evidence freshnessvalidatedComputational artifacts, scientific policy, and full provenance-card content resolve through a deterministic class-preserving index.
ProvenancevalidatedRuntime manifests and source-backed cards remain distinct from computational evidence while sharing identity, digest, and freshness infrastructure.
Curriculum qualityimplementedOptional preparation, a complete expert-reviewed conceptual spine, three executable investigations, and instructor assessment all use predict → compute → audit → state the warranted claim; learner validation remains future evidence.
AccessibilityimplementedThe docs gate checks alt text; new figures also use marker and line-style encodings beyond color.
DiscoverabilityimplementedFour explicit homepage routes, foundations, investigations, science patterns, modules, API reference, and validation serve distinct questions.
Downstream reuseimplementedLocal parity fixtures and known sibling use motivate shared primitives, but pinned adoption and compatibility evidence is not yet generated or validated.

High-confidence gaps

Now

1. Add contract-derived foundation figures and learner evidence

Impact. Learners can see parameter-space geometry, derivative maps, removable limits, and evidence classes while maintainers can test whether the material actually improves comprehension and transfer.

Audience and ownership. Research students and courses; Jaxstro owns public-API-derived figures and structural checks, while instructors and learners provide comprehension and accessibility evidence.

Risk. Attractive diagrams can oversimplify conditional derivative and model claims or become manually maintained parallel truth.

Evidence gate. Figures derive from contracts, evidence artifacts, public APIs, or explicit analytic identities; learner checks test stated misconceptions and accessibility rather than page existence alone.

The scientific contract registry, evidence index, foundations spine, executable investigations, and instructor rubric now provide the source material.

2. Generate a JAX transform-contract matrix

Impact. Users can see supported transforms, static arguments, batching cost, and smoothness caveats per public API.

Audience and ownership. Scientific developers; structured contracts live beside exports and resolve to tests and prose.

Risk. Check boxes could hide conditional domains.

Evidence gate. Every claimed cell links to a test and limitation; unverified cells remain explicitly unclaimed.

3. Ratify or reject jaxstro.quantity adoption

Impact. Resolve dual-surface maintenance and either unlock ecosystem-wide dimensional safety or establish a stable non-adoption boundary.

Audience and ownership. All sibling packages; each owns scientific parity while Jaxstro owns quantity mechanics.

Risk. Premature migration could destabilize mature science.

Evidence gate. Downstream parity, serialization, performance, ergonomics, and migration-cost reports precede the ADR.

4. Generate downstream adoption evidence

Impact. Reuse claims become observable, and abstractions are judged by real cross-project demand.

Audience and ownership. Ecosystem maintainers; downstream projects retain scientific acceptance.

Risk. A manual matrix would immediately age.

Evidence gate. Generate records from pinned imports, compatibility tests, and last-verified revisions.

Next

1. Visualize removable singularities

Impact. Teach why correct forward branches can have wrong parameter derivatives.

Evidence gate. Curves and limits derive from public APIs, analytic limits, and central FD; color is not the only encoding.

2. Visualize the AD contract taxonomy

Impact. Make smooth, blocked, value-first, validation-only, and certified implicit claims distinguishable at a glance.

Evidence gate. First deliver the planned transform-contract or maturity registry; then require every visual class to map to that registry and an executable example.

3. Add validation-triangle figures

Impact. Separate analytic, AD, FD, convergence, and provenance support.

Evidence gate. Figures are generated from artifacts and fail their declared tolerances when stale.

4. Standardize compile, JAXPR, runtime, and memory reports

Impact. Transform-heavy APIs can be assessed on developer cost as well as values.

Evidence gate. Reports separate compile, warm runtime, evaluations, graph primitives, and peak memory.

5. Define public API maturity metadata

Impact. Reference pages can distinguish ratified, validated, implemented, experimental, and planned surfaces.

Evidence gate. One registry is checked against exports, docs, and required validation anchors.

6. Build provenance and ownership-flow figures

Impact. Trace source artifact → preparation → JAX kernel → validation → claim.

Evidence gate. Every node resolves to a real record, module, or evidence page.

7. Add science-facing exercise sequences

Impact. Turn roots, interpolation, integration, distributions, spatial methods, and spectra into research-onboarding units.

Evidence gate. Each sequence has objectives, prerequisites, tested code, misconception checks, and a bounded claim rubric.

Later

1. Broaden certified implicit primitives after adoption evidence

Impact. Vector roots or fixed points could support differentiable equilibria.

Evidence gate. Two consumers share the contract and validate uniqueness, conditioning, residual error, and linear solves first.

2. Evaluate adaptive integration ownership

Impact. Error-controlled workflows could extend fixed-step coverage.

Evidence gate. Demonstrate cross-project demand for a narrowly defined primitive, compare against established specialized solvers as the default owner, and show that the proposed boundary avoids duplicating their stack.

3. Evaluate sparse iterative operators

Impact. Larger inverse problems may need matrix-free solves.

Evidence gate. Demonstrate cross-project demand, conditioning, convergence telemetry, AD semantics, and dependency fit.

4. Evaluate higher-dimensional mesh primitives

Impact. Conservative methods could serve more simulation domains.

Evidence gate. Define topology, conservation, capacity, JAX shapes, ownership, and two real consumers before implementation.

Evidence required

Accept a future SOTA claim only when the relevant layer includes:

  1. a mathematical or semantic contract;

  2. a public API with typed boundary and failure behavior;

  3. supported JAX transforms and cost semantics;

  4. independent validation rather than self-consistency alone;

  5. reproducible artifacts where metrics matter;

  6. accessible teaching that states what the evidence does not prove;

  7. downstream adoption evidence when reuse justifies ownership.

The highest-impact direction is not adding the most algorithms. It is making the existing breadth uniformly inspectable from equation to execution to claim, while admitting new primitives only when reuse and evidence justify them.