posts.text_is_promotional -> text_flagged and
posts.vision_is_real_cat -> vision_accepted across models,
schemas, pipeline, and REST responses, with a data-preserving
alembic migration (and downgrade).

@@ -20,6 +21,10 @@ The posts API's `reviewer_notes` field is now `auto_reject_reason`. The old name
 
 ---
 
+The two AI verdict fields in the posts API have new, topic-neutral names: `text_is_promotional` is now `text_flagged`, and `vision_is_real_cat` is now `vision_accepted`. Since the classification prompts are configurable, the API should not assume the curated topic is cats — a book-curation instance flags text for opposite reasons, and "accepted" describes what the vision stage decided without naming the subject. Existing databases keep their data through the rename via a migration; only code reading these field names needs updating.

@@ -20,6 +21,10 @@ The posts API's `reviewer_notes` field is now `auto_reject_reason`. The old name
 
 ---
 
+The two AI verdict fields in the posts API have new, topic-neutral names: `text_is_promotional` is now `text_flagged`, and `vision_is_real_cat` is now `vision_accepted`. Since the classification prompts are configurable, the API should not assume the curated topic is cats — a book-curation instance may flag text for entirely different reasons, and "accepted" describes what the vision stage decided without naming the subject. Existing databases keep their data through the rename via a migration; only code reading these field names needs updating.

@@ -395,0 +455,4 @@
+
+    @test("Downgrade restores the original AI verdict column names")
+    def test_downgrade_restores_names() -> None:
+        """Downgrading from head puts the cat-era column names back with data intact."""