Take this simple MSSQL model for example:

MODEL (
  name s.t,
  kind INCREMENTAL_BY_UNIQUE_KEY (
    unique_key (c)
  )
);

SELECT
  'ab' AS c

Since SQLMesh (i.e., SQLGlot) infers the column's type to be VARCHAR, the model is considered "annotated", so instead of creating its snapshot's table with a CTAS query, we use create_table for it directly here.

Unfortunately, using the VARCHAR type in an MSSQL CREATE TABLE statement results in a VARCHAR(1) actual type under the hood, so the model's evaluation fails with the following error:

    OperationalError:
      (2628, b"String or binary data would be truncated in table 'master.sqlmesh__s.s__t__4201193267', column 'c'. Truncated value: 'a'.DB-Lib error message 20018, severity 16:\nGeneral
SQL Server error: Check messages from the SQL Server\n")

A couple of ideas that we can consider are:

1. Avoid using inferred types in DDL statements, either generally or specifically for MSSQL
2. Ensure plain VARCHAR types are generated as VARCHAR(MAX) for MSSQL DDL statements, to avoid truncation issues.

Relevant Slack thread for reference: https://tobiko-data.slack.com/archives/C044BRE5W4S/p1751937537407689