[Lojhan]

Dynamic mapped lists currently widen every result property to SQL-nullable unknown, including an ordinary WHERE id IN (...) or bulk INSERT ... VALUES ... RETURNING. Preserve the selected grammar's inferred results when its AST proves that every repeated range supplies scalar inputs. PostgreSQL, MySQL, and SQLite certify scalar IN lists in top-level SELECT predicates and scalar VALUES rows. Other placements retain the conservative fallback.

The compiler passes SQL ranges through the optional dialect analysis context and keys cached analysis by those ranges as well as SQL text. Identical representative SQL from an input list and a repeated JSON projection therefore cannot share incompatible evidence. RETURNING column locations also map through expanded SQL, preserving editor source navigation.

Delimiter checks now consume each list's closing token explicitly and distinguish punctuation from a closing-parenthesis string literal. Nested calls no longer steal an outer closing token or make a missing delimiter appear valid.

Validation: full pnpm verify passed, including coverage, packed consumers and editor type checks, docs contracts/build, performance budgets, and soak checks. The shared three-grammar corpus asserts result types, ordered parameter patterns, diagnostics, and source spans in both compilation modes. Regressions cover both cache orders, actual runtime plan selection at cardinality three, absent evidence, repeated RETURNING projections, literal delimiters, and the existing JSON cardinality counterexamples. Includes package Changesets and documentation.