10.10 Failure modes and open obligations

CoLT does not solve reasoning by definition. It makes several failure modes auditable:

• an incomplete declaration can omit the obligation that should have failed;

• an insufficient tangent site can miss a fragile inference;

• an overly aggressive quotient can erase semantically material change;

• an incomplete cover can support false localization or false consensus;

• a repair generator can search an impoverished hypothesis language;

• an admission test can reuse proposal data and overstate validation; and

• a certificate can faithfully record an invalid causal or domain assumption.

These are not defects hidden behind a single accuracy score. They are typed open obligations that later applications can test separately.