I ran 17 LDBC queries against five engines at two scales. TypeGraph on SQLite wins every point read, often by 10–100x, and the engines built around an in-memory graph index win whole-graph algorithms by three to four orders of magnitude.
Bitemporal recorded time and provenance-backed retraction let a graph say why it believes a fact, recompute what still holds when a source turns out wrong, and replay what it believed at any commit.
Connected components, weighted shortest path, PageRank, personalized PageRank, and label propagation now run as SQL against your store. On an 18-paper citation graph they turn up things a citation count never would.
Two CRM bots stream what they know about the same person into their own belief graphs. One crashes and resumes without a duplicate row, and entity resolution folds both streams into one canonical record.
Roughly 35 performance fixes across bulk writes, point reads, traversals, and search, plus a caching regression I shipped in 0.34 and fixed properly here.
