The Query Revolution

Review note: This written summary was drafted from on-site product material and may not match every detail in the video. Please verify before publishing.

Summary

Similarity search finds things that look alike. Keyword search finds things that match a string. Neither can express the compound questions that matter in real business data: what was true as of a specific date, what changed between two points in time, which source superseded another, or what is missing from the record entirely.

The query revolution in Evolutionary Neural Memory is about making these questions expressible — because facts are typed, timestamped at claim level, and connected in a graph. Pod-OS answers in one query what would require stitching together multiple systems (or confident guessing) in a conventional stack.

Key Takeaways

• Similarity isn’t enough. Embeddings rank by resemblance; they cannot filter by validity interval, source precedence, or structural change.
• Compound queries need typed, time-stamped facts. As-of state, delta queries, and contradiction detection require predicates that vector and keyword indexes do not support.
• One query, whole picture. Pod-OS traverses Links and applies temporal filters in a single operation rather than across separate stores.
• Expressibility is the proof. The difference shows up in questions the old stack literally cannot formulate.

What You’ll Learn

• Why “find me files about Q3 sales” and “what changed in our top accounts’ behaviour before they churned?” are fundamentally different query classes
• How as-of queries retrieve the state of a fact at a specific point in time
• How delta queries surface what changed between two dates and which source superseded prior claims
• Why vector databases cannot rank for temporal or structural predicates
• Examples of expressible queries that similarity or keyword search cannot answer