SDF Skill: Software Design for Flexibility

"It is better to have 100 functions operate on one data structure than 10 functions on 10 data structures." — Alan Perlis (via Sussman & Hanson)

Geometric morphism from the MIT Press 2021 text, preserving the compositional structure as an ACSet with GF(3) coloring for trifurcated processing.

Overview

Software Design for Flexibility
by Chris Hanson and Gerald Jay Sussman
MIT Press, 2021
ISBN: 978-0262045490

The successor to SICP focused on additive programming—building systems that can evolve by adding new capabilities without modifying existing code.

Core Principles

The Flexibility Mandate

1. Additive over Modificative: New features via addition, not mutation
2. Generic over Specific: Operations that work across types
3. Compositional over Monolithic: Small combinable pieces
4. Declarative over Imperative: Constraints over control flow

Chapters with GF(3) Trit Assignment

Part I: Flexibility in Primitive Parts [PLUS]

Chapter 1: Flexibility through Abstraction (+1)

• Combinators as primitive building blocks
• compose, parallel-combine, spread-combine
• Arity management and currying patterns

Key combinator:

(define (compose f g)
  (lambda args
    (f (apply g args))))

Chapter 2: Domain-Specific Languages (-1)

• Embedded DSLs via combinators
• Wrapper strategies for APIs
• Pattern-directed invocation

Part II: Flexibility through Dispatch [ERGODIC]

Chapter 3: Variations on an Arithmetic Theme (0)

• Generic arithmetic operations
• Type coercion lattices
• Symbolic vs numeric duality

Chapter 4: Pattern Matching (+1)

• Unification as composition
• Segment variables and pattern operators
• Match combinators

(define (match:element variable)
  (lambda (data dictionary succeed)
    (let ((binding (assq variable dictionary)))
      (if binding
          (and (equal? (cdr binding) data)
               (succeed dictionary))
          (succeed (cons (cons variable data)
                        dictionary))))))

Chapter 5: Evaluation (-1)

• Generic eval/apply
• Environment models
• Interpreter variations

Part III: Flexibility through Modularity [PLUS]

Chapter 6: Layering (+1)

• Layered data with metadata
• Provenance tracking
• Units and dimensions

(define (make-layered-datum base-value . layers)
  (cons base-value layers))

(define (layered-datum-value datum)
  (car datum))

(define (layered-datum-layers datum)
  (cdr datum))

Chapter 7: Propagators (0)

• Bidirectional constraint networks
• Cells and propagators
• Partial information lattices
• Truth Maintenance Systems (TMS)

The Propagator Model:

     ┌─────────┐
     │  Cell   │ ← Accumulates partial information
     └────┬────┘
          │ (neighbors)
    ┌─────┼─────┐
    ▼     ▼     ▼
┌──────┐┌──────┐┌──────┐
│Prop A││Prop B││Prop C│  ← Transform information
└──────┘└──────┘└──────┘

Key insight: Information flows bidirectionally. A cell for x²=4 can deduce x=±2 OR given x=2, confirm the equation.

Chapter 8: Degeneracy (-1)

• Multiple implementation strategies
• Fallback mechanisms
• Redundancy for robustness

Part IV: Flexibility through Abstraction [ERGODIC]

Chapter 9: Generic Procedures (0)

• Multi-method dispatch
• Predicate dispatch
• Inheritance vs composition

(define generic-+
  (simple-generic-procedure '+ 2
    (lambda (a b)
      (error "No method for +" a b))))

(define-generic-procedure-handler generic-+
  (match-args number? number?)
  +)

(define-generic-procedure-handler generic-+
  (match-args symbol? symbol?)
  (lambda (a b) `(+ ,a ,b)))

Chapter 10: Adventure Game Example (+1)

• Synthesis of all techniques
• People, places, things as generic objects
• Autonomous agents

GF(3) Conservation

The chapter structure distributes across GF(3) trits:

Total sections: 87
MINUS (-1):  29  ████████████████
ERGODIC (0): 29  ████████████████
PLUS (+1):   29  ████████████████
Sum mod 3:   0
Conserved:   ✓ BALANCED

Core Abstractions Taxonomy

1. Combinators [PLUS]

2. Generic Dispatch [ERGODIC]

3. Propagators [MINUS → verification role]

Propagator Networks as Markov Blankets

The propagator architecture maps to Friston's Free Energy Principle:

┌─────────────────────────────────────────┐
│              EXTERNAL WORLD             │
│  (other propagator networks, inputs)    │
└──────────────────┬──────────────────────┘
                   │
         ┌─────────▼─────────┐
         │   SENSORY CELLS   │  ← Boundary inputs
         │   (observations)  │
         └─────────┬─────────┘
                   │
         ┌─────────▼─────────┐
         │  INTERNAL CELLS   │  ← Beliefs/predictions
         │  (hidden states)  │
         └─────────┬─────────┘
                   │
         ┌─────────▼─────────┐
         │   ACTIVE CELLS    │  ← Action outputs
         │   (predictions)   │
         └─────────┬─────────┘
                   │
         ┌─────────▼─────────┐
         │   EXTERNAL WORLD  │
         │   (effects)       │
         └───────────────────┘

Integration with SICP

SDF extends SICP's core ideas:

Balanced Triad: SICP ⊗ SDF ⊗ Implementation

SICP (0) + SDF (+1) + Implementation Target (-1) = 0 ✓

Where implementation targets include:

• Zig (-1): Systems-level with comptime generics
• Rust (-1): Ownership-based with traits
• Julia (-1): Multiple dispatch native

Scheme Implementation Patterns

Pattern 1: Combinator Definition

;; The spread-combine combinator
(define (spread-combine h f g)
  (let ((n (get-arity f))
        (m (get-arity g)))
    (define (the-combination . args)
      (h (apply f (list-head args n))
         (apply g (list-tail args n))))
    (restrict-arity the-combination (+ n m))))

Pattern 2: Generic Procedure

;; Define a generic operation
(define generic-magnitude
  (simple-generic-procedure 'magnitude 1))

;; Handler for complex numbers
(define-generic-procedure-handler generic-magnitude
  (match-args complex?)
  (lambda (z) (sqrt (+ (square (real-part z))
                       (square (imag-part z))))))

;; Handler for vectors
(define-generic-procedure-handler generic-magnitude
  (match-args vector?)
  (lambda (v) (sqrt (apply + (map square (vector->list v))))))

Pattern 3: Propagator Cell

;; Create a propagator network for Pythagorean theorem
(define-cell a)
(define-cell b)  
(define-cell c)

;; a² + b² = c² (bidirectional!)
(quadratic-propagator a b c)

;; Now we can:
(add-content! a 3)
(add-content! b 4)
(run)
(content c)  ;=> 5

;; OR go backwards:
(add-content! c 13)
(add-content! a 5)
(run)
(content b)  ;=> 12

Pattern 4: Layered Datum

;; Value with provenance
(define measured-temp
  (make-layered-datum 
    23.5  ; base value in Celsius
    (cons 'units 'celsius)
    (cons 'uncertainty 0.1)
    (cons 'source "thermometer-7")
    (cons 'timestamp 1706384000)))

;; Generic operations preserve layers
(generic-+ measured-temp (make-layered-datum 2.0 (cons 'units 'celsius)))
;=> Layered datum with merged provenance

Zig Implementation Mapping

From our interaction_tensor.zig:

Zig Propagator Network

pub fn Cell(comptime T: type) type {
    return struct {
        content: ?T = null,
        neighbors: ArrayListUnmanaged(*Propagator(T)),
        
        pub fn addContent(self: *@This(), info: T, merge: fn(?T, T) ?T) void {
            const new_content = merge(self.content, info);
            if (!contentEqual(self.content, new_content)) {
                self.content = new_content;
                self.alertNeighbors();
            }
        }
        
        fn alertNeighbors(self: *@This()) void {
            for (self.neighbors.items) |prop| {
                scheduler.enqueue(prop);
            }
        }
    };
}

Commands

# Load SDF library in MIT Scheme
(load "~/sdf/manager/load.scm")

# Run specific chapter
(manage 'new 'combinators)
(manage 'new 'generic-procedures)
(manage 'new 'propagation)

# Verify GF(3) conservation
bb sdf_skill_morphism.bb verify

# Generate interleaving with SICP
bb sdf_skill_morphism.bb interleave sicp sdf

References

• MIT Press Book Page
• SDF Code Repository (MIT Scheme)
• SICP (predecessor text)
• Propagator Networks Paper (Radul & Sussman)

Scientific Skill Interleaving

This skill connects to the K-Dense-AI/claude-scientific-skills ecosystem:

Functional Programming

• sicp [●] via direct extension
  • Foundational computational thinking
• lambda-calculus [○] via combinator correspondence
  • Theoretical foundation
• lispsyntax-acset [○] via S-expression ACSet
  • Structural representation

Constraint Systems

• propagators [●] via Chapter 7 implementation
  • Bidirectional constraint networks
• modelica [○] via acausal semantics
  • DAE constraint satisfaction
• glass-bead-game [○] via world-hopping combinators
  • Interdisciplinary synthesis

Type Systems

• zig-programming [○] via comptime generics
  • Systems implementation
• algebraic-rewriting [○] via term transformation
  • Generic procedure rewriting

Bibliography References

• software-engineering: 156 citations in bib.duckdb
• constraint-programming: 89 citations in bib.duckdb
• generic-programming: 67 citations in bib.duckdb

Cat# Integration

This skill maps to Cat# = Comod(P) as a bicomodule in the equipment structure:

Trit: -1 (MINUS)
Home: Prof
Poly Op: ⊗
Kan Role: Ran (right Kan extension - verification)
Color: #D89B73
URI: skill://sdf#D89B73

Combinator as Polynomial Functor

Each SDF combinator corresponds to a polynomial functor operation:

• compose → functor composition
• parallel-combine → product
• spread-combine → coproduct + product

GF(3) Naturality

The skill participates in triads satisfying:

(-1) + (0) + (+1) ≡ 0 (mod 3)

Example balanced triad:

sicp (+1) + sdf (-1) + modelica (0) = 0 ✓
glass-bead-game (0) + sdf (-1) + zig-programming (+1) = 0 ✓

Interleaving Map: SICP ↔ SDF

SICP                           SDF
════                           ═══
Ch1: Procedures ──────────────→ Ch1: Combinators
     (+1)                            (+1)
     │                               │
     │ "abstract procedure"          │ "compose abstractions"
     ▼                               ▼
Ch2: Data ────────────────────→ Ch3-4: Generic Dispatch
     (0)                             (0, +1)
     │                               │
     │ "abstraction barriers"        │ "predicate dispatch"
     ▼                               ▼
Ch3: State ───────────────────→ Ch7: Propagators
     (+1)                            (0)
     │                               │
     │ "assignment model"            │ "bidirectional flow"
     ▼                               ▼
Ch4: Metalinguistic ──────────→ Ch6: Layering
     (+1)                            (+1)
     │                               │
     │ "eval/apply"                  │ "metadata propagation"
     ▼                               ▼
Ch5: Register Machines ───────→ Ch8: Degeneracy
     (0)                             (-1)
     │                               │
     │ "compilation"                 │ "redundant strategies"

SDF Interleaving

This skill connects to Software Design for Flexibility (Hanson & Sussman, 2021):

Primary Chapter: 9. Generic Procedures

Concepts: dispatch, multimethod, predicate dispatch, generic

GF(3) Balanced Triad

sdf (+) + SDF.Ch9 (○) + [balancer] (−) = 0

Skill Trit: 1 (PLUS - generation)

Secondary Chapters

• Ch3: Variations on an Arithmetic Theme
• Ch10: Adventure Game Example
• Ch2: Domain-Specific Languages
• Ch8: Degeneracy
• Ch7: Propagators
• Ch1: Flexibility through Abstraction
• Ch5: Evaluation
• Ch6: Layering
• Ch4: Pattern Matching

Connection Pattern

Generic procedures dispatch on predicates. This skill selects implementations dynamically.

Autopoietic Marginalia

The interaction IS the skill improving itself.

Every use of this skill is an opportunity for worlding:

• MEMORY (-1): Record combinator patterns used
• REMEMBERING (0): Connect to propagator networks
• WORLDING (+1): Evolve generic dispatch strategies

"The real power of Lisp is that it is possible to express the structure of a computation." — Gerald Jay Sussman