Block Library

FunctorFlow ships with a library of reusable diagram builders.

Foundational blocks

These are the basic categorical patterns used throughout the package:

ket_block for left-Kan aggregation
completion_block for right-Kan completion
db_square for commutativity and obstruction losses
gt_neighborhood_block for graph-transformer style neighborhood aggregation

Planning and repair

FunctorFlow includes block patterns inspired by categorical planning:

basket_workflow_block
rocket_repair_block
basket_rocket_pipeline
democritus_assembly_pipeline
topocoend_block
horn_fill_block
higher_horn_block
bisimulation_quotient_block

These builders support symbolic examples directly and can also be compiled to Lux-backed models where appropriate.

CATAGI block guide

The CATAGI-style surfaces now have a clearer documentation path:

Block family	Symbolic surface	Trainable surface	Best walkthrough
BASKET / ROCKET planning	`basket_workflow_block`, `rocket_repair_block`, `basket_rocket_pipeline`	`build_basket_rocket_lux_model`	`18-neurosymbolic-pipelines`
Democritus assembly	`democritus_assembly_pipeline`	symbolic-only	`18-neurosymbolic-pipelines`
TopoCoend relation learning	`topocoend_block`	`build_topocoend_lux_model`, `RelationInferenceLayer`	`23-topocoend-triage` then `07-lux-neural-backend`
Horn and higher-horn consistency	`horn_fill_block`, `higher_horn_block`	`build_horn_lux_model`, `build_higher_horn_lux_model`	`18-neurosymbolic-pipelines` then `07-lux-neural-backend`
Bisimulation quotienting	`bisimulation_quotient_block`	`build_bisimulation_quotient_lux_model`	`24-bisimulation-quotient` then `07-lux-neural-backend`

A good reading order is: 18-neurosymbolic-pipelines for the symbolic overview, 23-topocoend-triage and 24-bisimulation-quotient for concrete focused examples, then 07-lux-neural-backend for the differentiable versions.

Tutorial libraries

To expose curated subsets of the block registry, use tutorial libraries:

using FunctorFlow

lib = get_tutorial_library(:planning)
diagram = build_tutorial_macro(lib, :basket_rocket_pipeline)

Key exported libraries include:

FOUNDATIONS_TUTORIAL_LIBRARY
PLANNING_TUTORIAL_LIBRARY
UNIFIED_TUTORIAL_LIBRARY

Lux-backed helpers

For neural execution, FunctorFlow provides:

compile_to_lux
KETAttentionLayer
DiagramDenseLayer
DiagramChainLayer
RelationInferenceLayer
build_ket_lux_model
build_db_lux_model
build_gt_lux_model
build_basket_rocket_lux_model
build_topocoend_lux_model
build_horn_lux_model
build_higher_horn_lux_model
build_bisimulation_quotient_lux_model

These helper constructors are especially useful for moving from symbolic diagram design to trainable models.