Use this page when you know what kind of help you need and want the shortest connected route through Jaxstro’s concepts, APIs, and evidence.
This site is not one compulsory linear sequence. Every route uses the same habit: predict -> compute -> audit -> state the warranted claim, but starts from a different need. Choose the smallest route that answers today’s question, then follow links outward when the audit exposes a missing concept or contract.
The routes remain separate because they answer different questions. Foundations asks what an idea means. A method page explains the finite algorithm and its boundaries. The API reference records the exact supported surface. Validation links a bounded claim to independent evidence. Keeping those roles visible prevents a successful execution from being mistaken for scientific validation.
Complete onboarding¶
Use this route if JAX or differentiable scientific computing is new to you.
Begin at Start here for package setup, precision, and ownership.
Read Why JAX? for the benefits, costs, and claim boundaries.
Work through JAX from first principles and run the executable map.
Use Choose your foundations path to reactivate any background you need.
Complete How to learn with Jaxstro: predict, compute, audit with your prediction written before you run the code.
Research-question first¶
Start from the scientific object you want: a root, integral, interpolation, distribution, coordinate transformation, spectrum, or other observable. Use the theory section to identify the mathematical and algorithmic contract, the API reference for the callable surface, and validation for executable evidence. Return to Foundations: the ideas we will not assume only when a term or relationship is blocking the question.
Background recovery¶
Use Choose your foundations path as a research router. Select by the idea you need today: units and scales, models and parameters, linear maps and derivatives, probability and inference, or conditioning and identifiability. Rejoin the research page as soon as the missing connection is active again.
API lookup¶
If you already know the method and need its signature, inputs, outputs, or failure state, go directly to the API reference reference. Follow its links back to theory before relying on a derivative, boundary policy, or approximation whose meaning is not already clear.
Method comparison¶
Open the relevant theory pages side by side. Compare the mathematical target, executed algorithm, transform support, failure behavior, telemetry, and audit requirements, not only speed or a successful return value. Use the research investigations for worked comparisons where available.
Result audit¶
Begin at Validation. Identify the exact claim, then find the analytic identity, limiting case, independent numerical check, convergence study, source card, or runtime manifest that supports it. If evidence is absent, weaken the claim or add the missing audit before treating the output as a result.
Group or mentor reading¶
Assign roles around one page: one reader states the prediction, one traces the executed method, one challenges the audit, and one records the strongest warranted claim. Compare predictions before running code so discussion is not shaped by knowing the answer in advance.
Contributor maintenance¶
Start with the architecture and decision records before changing an owner or public contract. Update source, tests, theory, API reference, and validation anchors together. Run focused tests first, then the strict documentation gate; keep generated registries fresh and route-manifest entries unique.
Whichever route you choose, keep the distinction between a plausible value, an executed numerical result, and a scientifically audited claim visible.