Summary

• extend the Tier-1 benchmark runner with BEIR FiQA, TREC-COVID, and SciFact plus staged corpus limits
• speed up SQLite batch FTS ingest by skipping FTS deletes for newly inserted nodes
• add public scale guard workflow and scale report covering FiQA 57k and TREC-COVID 171k runs

Scale results

• FiQA 57,638 docs build improved from 577.4s to 58.4s; search 1.4s over 10 queries
• TREC-COVID staged/full smoke: 50k build 49.3s/search 1.3s; 171,332 build 370.4s/search 5.2s

Tests

• uv run --extra dev ruff check
• uv run --extra dev ruff format --check
• uv run --extra dev pytest tests/test_tier1_benchmarks.py tests/test_backend_sqlite.py tests/test_backend_sqlite_graph.py tests/test_real_scores.py tests/test_memory_operating_layer.py::test_graph_mutations_record_memory_events tests/test_memory_operating_layer.py::test_graph_add_can_skip_memory_event_for_bulk_loads -q
• CI-like local suite earlier in this branch: 1474 passed, 5 skipped