Universal software factory needs a declared universality axis

Type: kb/types/note.md · Tags: foundations, computational-model, self-improving-systems

An unqualified universal software factory is not a stable technical concept. The word universal has been used for different properties, and evidence for one does not establish the others.

At least four axes must be kept separate:

Axis Question What it does not establish
Target or platform portability Can one development environment support several processors, platforms, or technology stacks? That it constructs new factory machinery or learns family knowledge
Factory-valued output Can a factory or tool factory produce another factory or production tool? That the target factory's production knowledge was acquired rather than supplied
Constructional expressivity Given an adequate description, can a general constructor realize a factory in a declared class? That it can determine which description is adequate for a new demand
Production-knowledge acquisition reach From permitted task and production evidence, can the system determine and retain the family-specific production knowledge needed across a declared demand class? That it can reproduce every possible factory implementation or eliminate all fixed general machinery

The axes are independent. Di Giovanni and Padella called their 1983 environment a universal factory because it could serve different microprocessor families. Greenfield, Tool Factory, and MDSoFa instead establish forms of recursive factory construction. Neither result shows that production evidence determines the family-specific knowledge embodied in the resulting machinery.

Constructional expressivity also becomes weak when the supplied description may contain the complete target. A sufficiently general compiler or interpreter can realize many factories from descriptions while leaving the hard semantic work—choosing the family boundary, representations, decompositions, evaluators, workflows, and variation knowledge—to the description's author.

Conversely, a learning system may acquire useful family-specific production knowledge across a declared class of demands without being able to realize every factory in some implementation universe. Production reach concerns finding adequate machinery for admitted demands, not enumerating all possible producers.

A qualified universality claim should therefore declare:

  • the axis being quantified;
  • the target, factory, or demand class;
  • what descriptions, family knowledge, task evidence, and interaction may be supplied;
  • the adequacy or acceptance relation;
  • the learner and human-intervention boundary;
  • the horizon and resource limits; and
  • how failures, abstentions, and excluded cases are counted.

Without these declarations, universal can collapse into ordinary portability, recursive output, compiler expressivity, or an untestable promise of arbitrary task competence.

Consequences for this research program

The program does not need a registered definition of universal software factory. For its factory-specific branch, the substantive claim can be stated directly: as covered demands widen, computation should increasingly acquire and retain the family-specific production knowledge required for later production rather than receive another bespoke human design.

That branch does not define the program's complete producer. A software house may use one or more Greenfield-style factories or none, and it does not require a declared product family. Evidence for broader capability or acquisition claims must identify the work examined, observation period, supplied inputs, and human production roles. These limits belong to the evaluation; the house's products and responsibilities may develop as new requirements arise.

A fixed general substrate remains compatible with that claim. What matters is whether target-specific production knowledge was already supplied, hidden in a catalog, or determined from the evidence permitted for the assessed demand.

Scope

  • The four-axis list is a working disambiguation, not a complete history of every use of universal factory.
  • The 1983 source establishes one explicit portability usage; it does not show that this was the field's dominant meaning.
  • The note does not claim that broad production-knowledge acquisition is impossible. It requires only that the evidence protocol expose the distinctions the system is expected to recover.
  • A qualified full-coverage claim may still be useful, but its content comes from the declared axis and frame rather than from the word universal.

Relevant Notes: