Skip to content

Primitive Families

The Planetary Kernel is organized around primitive families.

A primitive family is a canonical execution class with stable semantics, versioned identity, deterministic behavior, and profile-driven specialization.

Primitive families are not marketing labels and not route names. They are the computational building blocks of the Kernel.


Canonical Model

Each distributed execution request resolves through:

  • primitive family name
  • primitive family version
  • profile
  • args
  • seed behavior
  • policy
  • reducer / output semantics

Example:

json
{
  "op": {
    "name": "mc",
    "version": 1,
    "profile": "eta.v1"
  }
}

This means the API remains stable while the Kernel expands.


Current Family Map

mc@1

Probabilistic simulation and deterministic Monte Carlo execution.

Used for:

  • ETA simulation
  • insurance loss modeling
  • climate ensembles
  • portfolio stress
  • compound risk scenarios

Reference:


ensemble@1

Reducer-driven multi-run aggregation and confidence-producing execution.

Used for:

  • consensus surfaces
  • confidence fusion
  • instability detection
  • multi-model aggregation
  • scenario comparison

Reference:


graph@1

Graph-oriented execution for traversal, relationship expansion, routing, dependency analysis, and structural reasoning.

Used for:

  • network traversal
  • path analysis
  • dependency propagation
  • graph scoring
  • connected-system evaluation

Reference:


search@1

Search-oriented execution for retrieval, ranking, candidate generation, scoring, and search-time distributed evaluation.

Used for:

  • distributed candidate retrieval
  • ranking pipelines
  • relevance scoring
  • search-time expansion
  • discovery under uncertainty

Reference:


media@1

Distributed media transformation and media-aware analysis.

Used for:

  • transcode
  • segmentation
  • frame extraction
  • pre-processing for analysis
  • large-scale media normalization

Reference:


tensor@1

Dense numeric and tensor-oriented execution.

Used for:

  • matrix operations
  • tensor transforms
  • embedding workloads
  • inference plumbing
  • distributed numeric pre/post-processing

Reference:


Families vs Profiles

A primitive family defines the computational class. A profile defines the workload specialization within that family.

Example:

  • family: mc@1
  • profiles: eta.v1, insurance.loss.v1, risk.compound.v1

This keeps the Kernel extensible without route explosion.


Families vs Adapters

Primitive families are execution-native.

Adapters are domain translators that:

  • normalize inputs
  • build canonical Kernel envelopes
  • interpret outputs for domain workflows

The Kernel remains the execution authority. Adapters remain the translation layer.