Skip to content
little big brain

Changelog

little big brain dogfoods itself: we track our own product development as a temporal knowledge graph inside little big brain, and this page is generated by querying that graph through the MCP tool belt. Each dated entry is a shipped feature, stamped with the merge/commit date of the work that landed it — read straight from the graph. See How this page is made.

June 2026

Section titled “June 2026”
  • 26 Jun — SPARQL-subset aggregation. COUNT/SUM/AVG with GROUP BY over entity, scalar-attribute, and calendar date-bucket keys, on a typed value channel — so a per-area or per-month rollup is one server-side query. See SPARQL & SHACL.
  • 25 Jun — Type-closure reasoning. An RDFS subClassOf hierarchy resolved at query time, so a query for a broad type returns its subtypes with no rebuild. See reasoning.
  • 24 Jun — As-of ranged traversal. Time-travel graph traversal pinned to a past snapshot with as_of — walk the graph as it was at any commit.
  • 23 Jun — Additive ontology evolution. Widen a relation’s domain/range or add entity types by name, with ontology-version stamping, preserving every existing record.
  • 21 Jun — A retrieval-quality wave:
    • Typed value channel — typed literals so numbers and dates filter and aggregate server-side instead of matching strings.
    • Graph-aware co-binding — fuse text/vector matches and graph structure into one ranked pass rather than fuse-then-seed.
    • Scored-atom retrieval and retrieval-premise fusion — calibrated, provenance-carrying retrieval premises for answer grounding.
    • Ontology-driven enrichment — expand queries through the ontology’s concepts and relations.
  • 20 Jun — Bounded SHACL-AF inference rules. Derive new edges to a bounded fixpoint and fold them into queries with reason.
  • 15 Jun — Exact rerank over quantized ANN (an opt-in exact-distance pass, wired end to end through the server) and stable search pagination cursors over large result sets.
  • 13 Jun — Incremental BM25 index deltas. Append-only delta runs so an index refresh doesn’t rebuild from scratch.
  • 11 Jun — WAL snapshot compaction. Graph segments plus a node read-through cache to bound object reads on the hot path.
  • 10 Jun — Quantized ANN blocks (4-bit quantized vectors with a norm-corrected L2 estimator), semantic-seeded traversal (resolve seed entities from natural-language text, then follow bounded graph paths), and traversal-expansion optimizations.

Foundations

Section titled “Foundations”

Core capabilities that predate or underpin the dated timeline above:

  • Object-storage-native retrieval — ranged adjacency blocks (sub-block addressing, anchor deltas, hub splitting, append-safe WAL-tail overlay merge), streaming ANN index builds that never hold the whole corpus in RAM, persisted corpus facets, and layered decoded-state / decoded-block / snapshot caches.
  • RDF schema & query surface — SHACL Core / SHACL-SPARQL / SHACL-AF shape bundles, structural domain/range gating on writes, RDFS TBox closure, ontology version stamping, stable content-addressed identity, typed-literal FILTER, and fail-closed query budgets.
  • Hybrid search — lexical + BM25 + vector + ontology + graph-neighborhood fusion with filters, facets, and reciprocal-rank multi-search over one snapshot.
  • Security & platform — short-lived data-plane session revocation and per-stack capacity hardening.

How this page is made

Section titled “How this page is made”

This changelog isn’t hand-maintained — it’s generated from our own product-development graph. We track development as typed facts inside little big brain: each feature is linked to the commit or pull request that implemented it, and those carry the committed_at / merged_at dates shown above. Reading it back through the MCP turns our engineering history into a dated timeline — the migrate-a-knowledge-base pattern applied to ourselves.