There is not an unqualified direct connection between a binding and a mapping.
A binding carried in a child archetype may not be appropriate at template -level e.g. we do not use or want LOINC mappings, or they may not have been formally Q/Ad and published.
Even when published, it is not uncommon for local mappings to be required. As an example I am doing Covid-19 mappings and will need to use a local UK SNOMED CT code in some cases, even though an international equivalent is available - there is good, practical reason for this ( temporary!). Also getting consensus is sometimes impossible.
In some cases we might want to record a plain text default value but have that mapped into SNOMED-CT, though I see that as largely going away in ADL2 when I would expect in most cases to use atCoded template-level codes. (some iDCodes as well).
We also may need to be able define other mappings attributes like ‘match’ and ‘purpose’ as it is going to be important o easily cross-query for SNOMED-CT codes recorded as amappings but with an equivalent match or perhaps flagged as ‘exact synonyms’ so that they can be picked up whether record as defiing_code or as a mapping (perhaps with a convenience function in AQL that can look in both places). This will be a very common issue, where depending on the exact application, there will be disparity between primary use of SNOMED-CT vs. recorded as a mapping.
I guess one possibility might be to expand ‘bindings’ to include the other TERM_MAPPPING attributes and to regard bindings carried explicitly in templates as being mapping directions, which could optionally pull-through bindings in underlying archetypes.