Definitions of new symbols merely abbreviate expressions in logical frameworks, and no new facts (regarding previously defined symbols) should hold because of a new definition. In Isabelle/HOL, definable symbols are types and constants. The latter may be ad-hoc overloaded, i. e. have different definitions for non-overlapping types. We prove that symbols that are independent of a new definition may keep their interpretation in a model extension. This work revises our earlier notion of modeltheoretic conservative extension and generalises an earlier model construction. We obtain consistency of theories of definitions in higher-order logic (HOL) with ad-hoc overloading as a corollary. Our results are mechanised in the HOL4 theorem prover. Const c ty 1 # Const d ty 2 def = c = d ∨ ty 1 # ty 2. A user may introduce (non-built-in) types and constants by theory extension, as described in Section 2.3. For types and constants we generally say symbols. Built-ins We abbreviate Bool for Tyapp «bool» [ ] x → y for Tyapp «fun» [x; y] Equal ty for Const «=» (ty → ty → Bool) s === t for Comb (Comb (Equal (typeof s)) s) t Any type with any of these type constructors at the top level is built-in, as is the constant Equal. These are the only symbols which are not user-defined. A formula is a term of type Bool.