Which component do I need?

Start with the scientific task. Repositories are boundaries of responsibility, not five interchangeable products.

I want to analyze data

Interactively: start with Fathom. It operates SPM-Kit Core through visual perspectives.

Programmatically, in notebooks, batch jobs, CI or HPC: start with SPM-Kit Core.

I want to test an algorithm

Against a known numerical surface: generate truth with Phantoms, then execute the comparison through Validation.

Against external software or a published reference: start with Validation and freeze reference independence, parameters and tolerance before execution.

I need a native-format fixture

Use Data Hunter to find candidates, then manually review format, rights and content. A raw-only file can exercise a Core reader; it cannot validate an analysis result.

I want to contribute a dataset

Begin with Data Hunter's evidence taxonomy and lawful provenance. If accepted, design the narrow fixture or Validation campaign separately. Do not upload restricted or personally sensitive data.

I want reproducible analysis over many files

Use Core through Python or `spmkit batch`/`fbatch`. Preserve versions, command, channel, units, preprocessing and input hashes. Fathom remains useful for inspecting representative cases.

I want to add a reader

Implement the versioned `spmkit.plugins.v1` reader contract in Core. Use a lawful native file as a parser fixture, add synthetic edge cases and define a later external validation path.

I want to add a scientific capability

Implement a pure Core function first, create a known recovery case with Phantoms, define an external evidence route with Validation, expose CLI/API, and add a Fathom panel only after numerical behavior is stable.

I want to propose physical validation

Start in Validation. Define the mensurand, calibrated reference, uncertainty, traceability, lawful data handling, blinded or independent design and pass/fail policy. Data Hunter may locate leads but cannot certify them.

No JavaScript required. The branches above are native HTML disclosure controls. If they are unavailable, use the complete table below.

Static decision table

Need	Primary component	Companion	Allowed conclusion	Not allowed
inspect an image or curve visually	Fathom	Core	the configured Core result was inspected interactively	the view is physically validated
Python/CLI/HPC analysis	Core	Fathom for spot-checking	the public function/command produced the result	all formats/models are equally mature
known synthetic answer	Phantoms	Validation	a numerical truth and corruption sequence are known	physical validity
external reference comparison	Validation	Core/Phantoms	the frozen comparison met or failed its rule	universal equivalence
public-data discovery	Data Hunter	human review	a record may support a named role	dataset certification
format fixture	Data Hunter → Core	Validation later	reader behavior can be exercised	analysis correctness
dataset contribution	Data Hunter → Validation	Core	a reviewed case can enter design	automatic acceptance/redistribution
plugin development	Core	Phantoms/Validation/Fathom	a versioned contract can be implemented	stable 1.0 API promise

Still unsure?

Use SPM-Kit Core if the task is a calculation. Use Fathom if the task is operating that calculation visually. Use the three companion repositories only when the task is evidence discovery, known truth or external validation.

Installation matrix · Workflow tutorials