Project 9 · Representation and retrieval0/5 completeVector Search
A tiny deterministic encoder, cosine similarity, exact top-k, versioned updates, and a semantic evaluation that can name where each retriever wins.
requires artifacts from: lexical-search
- L1Deterministic Encoderdefine the contract · ~14 minstart →
- L2Cosine Similaritypreserve an invariant · ~12 minlocked
- L3Exact Top-kmeasure behavior · ~16 minlocked
- L4Versioned Updatesbound a failure · ~20 minlocked
- L5Semantic Evaluationmake an operational decision · ~24 minlocked
Artifacts this project producesL1 · Deterministic Encoderoutputs: vector-search:l1-deterministic-encoder:artifact
L2 · Cosine Similarityoutputs: vector-search:l2-cosine-similarity:artifact
L3 · Exact Top-koutputs: vector-search:l3-exact-topk:artifact
L4 · Versioned Updatesoutputs: vector-search:l4-versioned-updates:artifact
L5 · Semantic Evaluationoutputs: vector-search:l5-semantic-evaluation:artifact