Reflective theory builder

Type: kb/types/definition.md · Tags: foundations, self-improving-systems, learning-theory

A reflective theory builder is a theory builder that performs reflective theory refinement on a theory of its own behavior-determining organization, where the organization in question is the machinery that determines how its theories are built, tested, and revised. The self-theory must be a causally connected self-representation in the sense of reflective system: changes in the machinery update the theory, and revisions of the theory can change the machinery. A builder that holds a description of its machinery and consults it without that two-way connection is not reflective in this sense. The KB needs the term because a readable self-theory doing causal work is what separates a methodology-driven builder from systems that retain content without one.

The self-theory is a tentative theory. Its claims about the machinery, including its interpreters, are decided by evidence, not by being read, and its own acceptance of a machinery change does not make the change warranted.

Work on external theories is both useful work and a test of the current machinery. When that work exposes a machinery limitation, the builder may revise its self-theory, change the implicated machinery, and test the change by returning to external theory-building work. Reflection therefore serves external theory-building competence; it is not a requirement to build theories only about the builder itself.

Evidence

Reflection itself is evidenced by the two-way causal connection, which internal records can establish: machinery changes that updated the self-theory and self-theory revisions that changed the machinery, as the Commonplace boundary case below shows. Whether a reflective episode paid, in performance the builder did not assess itself, is a further claim evidenced under the externally tested case by one connected path:

  1. Externally assessed work exposes a possible machinery limitation. The outcome signal alone does not locate the fault.
  2. Diagnosis revises an identified commitment in the self-theory, using evidence about the machinery's actual operation.
  3. That revision guides a machinery change, and the installed change and its operation update or correct the self-theory, preserving the causal connection in both directions.
  4. Later theory-building work uses the changed machinery, and external work consumes the resulting product and tests its consequences under the declared evaluation protocol.

The record holds the versions, consumption traces, predicted effects, and resulting outcomes that connect those steps. Matched interventions on the retained commitment or its consumption path strengthen the attribution that the self-theory guided the change; they test the causal connection and do not replace the definition's two-way requirement. A machinery change followed by better outcomes does not by itself establish that attribution. Establishing the reflective path and establishing that it improved performance are separate claims.

Scope

  • Independent of autonomy. A reflective builder may depend on a person for diagnosis or machinery changes; see autonomous theory builder.
  • Independent of extension. Reflection concerns whether machinery changes pass through a causally connected self-theory; extension concerns whether retained changes supply capability beyond the seed baseline under a budget. A reflective revision can fail to improve capability, and a capability gain can arise without a causally connected self-theory, including through a change to an existing procedure or prompt. PAST-Bench draws the same line from outside: it measures retained experience changing later behaviour and leaves improvement of the mechanisms by which a system stores, retrieves, verifies, and updates to future work.
  • Inside the externally tested case. Internal diagnosis and targeted probes do not move the builder outside the externally tested case while the consequences of the change still face external assessment.

Exclusions

  • A builder that holds and consults a description of its machinery with no two-way causal connection.
  • A machinery change with no retained account of why; that may be an extension, but it is not reflection.
  • A standard of faithful interpretation. The condition does not require one; the self-theory's claims about interpretation are decided by evidence like any other.

Misuse Cases

  • Calling a builder reflective because it retains prose about itself.
  • Reading better outcomes after a machinery change as evidence that a self-theory guided the change.
  • Treating a self-theory's own acceptance of a machinery change as warrant for the change.
  • Treating a self-description, a successful machinery change, or a computational proof as sufficient on its own to establish reflective theory refinement.

Boundary cases

  • FORTE is not reflective: its theory is external and its machinery is not represented in any theory it refines.
  • Commonplace today, with the operator inside the boundary, is reflective: the KB's methodology notes, type specs, and ADRs are a theory of its own theory-building machinery, and revisions of them change the validators, skills, and contracts that build later theories. The evidence for the two-way connection is in Commonplace as a reflective system.
  • The Gödel machine is a reflective system, since its axioms describe its own software and its rewrites are causally connected to them. Whether it is a reflective theory builder turns on whether that self-representation is revised against evidence rather than serving only as a premise of proofs; that classification stays open.

Relevant Notes: