Recorded SIFT1M evaluation: one million image-descriptor vectors with 128 values each, tested with 200 searches. The approximate index recovered 99.9% of the ten true closest matches on average. This measures how well the index reproduces exact nearest-match results; it does not measure language-model answers or understanding of document meaning.

Based on source review and saved project artifacts. Recorded measurements apply to the stated model, dataset and configuration; they are not claims of a current production deployment.