ch-colt
10 Reasoning by Structural Repair
The CoLT Architecture
The preceding chapters developed lincs as a theory of structural diagnosis and repair. The same machinery also defines a general architecture for reasoning. A reasoning system maintains a typed model of its commitments, probes that model, records failures of composition, proposes localized changes, and changes its maintained state only when independent evidence admits the proposal. We call an auditable sequence of these operations a Chain of lincs Thought, or CoLT.
11. We write CoLT, with a lowercase “o,” to distinguish the architecture from COLT, the Conference on Learning Theory. ↩
CoLT is not a renaming of ordinary chain-of-thought prompting. A generated rationale is a sequence of linguistic tokens. A CoLT trace is a sequence of typed structural states and evidence-bearing transition records. It need not reveal, imitate, or claim access to a model’s private internal computation. Its purpose is to expose what the maintained system declared, what was observed to fail, what was changed, and why that change was admitted.
A plausible narrative is not yet an auditable reasoning trace. CoLT requires typed commitments, explicit obstruction witnesses, registered repairs, and admission evidence. Fluency may be an observation of a trace, but it is not its semantics.