Fix onboarding skipping embedding generation due to micro-ingestion

mitchins · mitchins · commit 52ee7efa692c · 2025-08-06T21:04:06.000+10:00
Root cause: Onboarding only ran full ingestion if indexed_count == 0,
but partial embeddings from previous runs caused it to skip to
micro-ingestion, leaving embeddings incomplete.

Result: First search had to generate missing embeddings (3-5s lag)

Fix: Check embedding_count vs indexed_count to detect incomplete
embeddings and force full ingestion when needed.

Now 25-30s embedding time moves back to onboarding where it belongs.
diff --git a/src/fuzzyshell/tui/onboarding.py b/src/fuzzyshell/tui/onboarding.py
@@ -116,7 +116,10 @@ def setup_delegate():
             advance_to_step(2)
             advance_to_step(3)
             indexed_count = fuzzyshell.get_indexed_count()
-            if indexed_count == 0:
+            embedding_count = fuzzyshell.command_dal.get_embedding_count()
+            
+            # Run ingestion if no embeddings OR embeddings don't match command count
+            if indexed_count == 0 or embedding_count != indexed_count:
                 added_count = fuzzyshell.ingest_history(use_tui=False, no_random=True)
                 if added_count and added_count > 0:
                     # Wait for database to be fully ready for searches
