ora-0158

14.1 Registered repair problems

Let \(p:E\to X\) be the simplicial family of decorated decisions used in Equation 9.6. The base simplex in \(X\) records the compositional history; its lift to \(E\) records evidence, action, consistency, consequence, and warrant data.

Definition 14.1 Registered repair problem

A registered repair problem is a tuple

\[ \mathfrak R=(e_\Lambda ,\sigma ,\mathcal O_{\mathrm{set}}, \mathcal O_{\mathrm{open}},\Omega ), \]

where \(e_\Lambda :\Lambda _i^n\to E\) is the observed decorated boundary, \(\sigma :\Delta ^n\to X\) is its proposed base completion, \(\mathcal O_{\mathrm{set}}\) is a family of obligations that must be preserved, \(\mathcal O_{\mathrm{open}}\) is a family still permitted to change, and \(\Omega \) is an observer of comparison defects. Its admissible repair space is the subspace

\[ \mathsf{Adm}(\mathfrak R) \hookrightarrow \mathsf{Fill}_p(e_\Lambda ,\sigma ) \]

cut out by the settled obligations and the declared safety invariant.

The division between settled and open obligations is the formal analogue of Piagetian assimilation. It says what may be revised without pretending that the rest of the learner’s world has been reconsidered. Accommodation occurs when no filler exists for the current base simplex or when every filler violates a settled obligation.

Repair profile

Homotopy type

Meaning

Required response

Impossible

\(\varnothing \)

No admissible completion has the declared boundary and base semantics

Change evidence, safety constraints, or the declaration

Coherently unique

Contractible

All admissible completions are joined by a coherent system of identifications

Assimilate without arbitrary tie-breaking

Discrete ambiguity

Several components

Qualitatively different repair families survive

Probe for evidence that separates components

Higher ambiguity

Nontrivial loops or higher cells

Repairs agree at the object level but their witnesses do not cohere uniquely

Preserve the mapping space or request higher coherence data

Table 14.1 The repair profile determines the next learning operation. A scalar defect cannot distinguish these four cases.