Retrieval, Embeddings and Evidence Tracking
Check representations, source versions and what a retrieved result actually supports.
Check the representation
Start with embedding compatibility before changing your search model. Equal vector dimensions do not mean the query and document vectors share a coordinate system.
Keep the index current
CloudVec explains how database changes reach a vector index after a crash. Graph memory asks whether explicit relationships improve on ordinary metadata retrieval. Assistant memory adds ownership and time validity.
Keep evidence attached
For retrieval augmented generation, retain the passage and version behind each answer. Autoresearch, Xanadu and OpenAbstract cover citation history, stable passage links and corrections. SL5 adds nested discussion attribution.
Articles in this reading path
- Embedding Compatibility: Models, Vectors and SearchRecord model revisions, pooling and normalisation beside embeddings, then compare retrieval methods without mistaking equal dimensions for compatibility.
- CloudVec: Vector Index Freshness with a SQLite OutboxFollow updates, deletion and crash recovery through a SQLite outbox, with freshness and replay checks kept separate from semantic search quality.
- Graph Based Research Memory with Source ProvenanceCompare graph and metadata retrieval on versioned sources, keeping source history visible and accounting for cases where the simpler baseline matches it.
- Local AI Memory: Retrieval and Temporal ValidityCompare current and historical assistant memory using ownership, validity and recording time, with a fixed word budget and bounded synthetic results.
- Autoresearch: Citation Tracking for Research AgentsTrack exact citations through revised sources without rewriting earlier answers, and distinguish citation preservation from evidence supporting a claim.
- Building Technical Knowledge Bases for TroubleshootingOrganise a technical knowledge base around the next useful check, with source evidence, device time and receipt time kept distinct.
- A Practical Workflow for Independent AI ResearchTurn an AI research question into a baseline, measurable test and evidence record without letting reading lists and architecture ideas replace the question.
- Organising Research Papers, Books and Reading NotesKeep selection, acquisition and reading separate in a research collection, with source versions and the question behind each saved paper or book.
- Xanadu Inspired Links to Versioned Document PassagesKeep document versions, passage spans and link identities intact through export and import, including Unicode text and historical source references.
- OpenAbstract: Versioned Claims and CorrectionsFollow a claim correction without erasing its history, separating replay correctness, support labels and the rules for the default reading view.
- SL5: Thread Aware Retrieval and Quote AttributionPreserve nested quote attribution, source timestamps and thread context, while separating exact attribution tests from forecast scoring and retrieval quality.