flat index

**Flat index** is the **exact vector-search index that compares each query against every stored vector without approximation** - it provides perfect nearest-neighbor recall at the cost of high computational expense. **What Is Flat index?** - **Definition**: Brute-force nearest-neighbor search over the full vector corpus. - **Accuracy Property**: Returns true exact top-k results given the selected similarity metric. - **Complexity Profile**: Query cost scales linearly with corpus size. - **Benchmark Role**: Serves as ground-truth reference for evaluating ANN recall. **Why Flat index Matters** - **Gold Standard Accuracy**: Needed when maximum retrieval correctness is required. - **Evaluation Baseline**: Essential for measuring approximation error of ANN methods. - **Small-Corpus Fit**: Practical for low-volume datasets or offline analytics workloads. - **Debug Utility**: Simplifies retrieval diagnostics by removing ANN approximation effects. - **Calibration Anchor**: Helps tune ANN parameters against exact search outcomes. **How It Is Used in Practice** - **Ground-Truth Runs**: Generate exact recall benchmarks for candidate ANN configurations. - **Hybrid Deployment**: Use flat search for high-value subsets and ANN for large tails. - **Capacity Planning**: Estimate compute requirements before scaling to larger corpora. Flat index is **the exact-reference method for vector retrieval** - although computationally expensive, it is indispensable for benchmarking, validation, and small-scale high-precision search workloads.