- Shipped
- August 10, 2026 at 12:35 AM UTC
- Author
- Kamo
- Commit
- d366580
The drift diff earned its place. KamoInitializerService builds the schema from JPA entities and produced 599 tables; production carries 714. The 115 difference are tables whose entities were deleted but whose tables never were -- abandoned pos_/hte_/ht_ schemas plus billing_* and ocr_document_*. legacy_ddl.py recreates them from SHOW CREATE TABLE, translating :::TYPE annotations, inline indexes, ASC in constraints, NOT VISIBLE, FAMILY and ON UPDATE. More importantly the column diff found 25 columns present in production but absent from the entity schema, including 11 on accounts (first_name, last_name, phone, address_line1, city, postal_code, ...) and 4 on leads. Those are 100% NULL in production so nothing would have been lost there -- but **************** and .is_licensee are populated in all 9 rows and would have been silently dropped by a load that trusted the entity schema. All 25 re-added. Both sides now report 714 tables / 8706 columns.