Whatever the refrigerated figure is, freezing does not simply extend it. minus 20 °C is 25 kelvin below a refrigerator, and below the glass transition of a lyophilised cake the ten-degree rule of thumb stops applying at all — solid-state chemistry is not slow liquid chemistry, it is a different regime, and the failure modes that survive it are mechanical rather than chemical. A frozen solution is not a slow solution: ice excludes solute, so the unfrozen fraction concentrates, the pH of the buffer shifts as one component crystallises first, and the damage happens during the transitions rather than during the hold. "Within specification" also needs a specification: purity, content, or both, and at what limit. Without that the question has no numerical answer at all.
Answer first: the degradation pathways worth knowing are hydrolysis, deamidation, oxidation, aggregation and adsorption, and each has a different trigger and a different mitigation.
Deamidation converts asparagine or glutamine to the corresponding acid via a succinimide intermediate, adding one dalton. It is base-catalysed, accelerates above neutral pH and is the dominant aqueous pathway for many peptides.
Hydrolysis cleaves the backbone, most readily at aspartate-proline and aspartate-glycine sequences, and is acid-catalysed. In a dry solid it barely proceeds at all.
Deamidation via the succinimide intermediate is well characterised, with sequence-dependent rates highest for asparagine-glycine motifs.
At dilute concentrations, suspect adsorption before you suspect chemistry.