Retrieval evaluation | Tayab

evidenceTwo benchmarks, three pipeline configurations, and every failing case left visible. Scores are computed in your browser by running the same pipeline the assistant uses.

Retrieval, measured in public

suitepipeline

MRR

1.00

Precision@1

1.00

nDCG@5

1.00

Pass rate

1.00

MRR by pipeline stage · golden setcomputed live · k=5

Dense only

1.00

+ RRF fusion

1.00

+ Reranker

1.00

full pipeline. Each bar is produced by running the bundled corpus through the same runTrace function the live assistant uses — there are no recorded numbers on this page.

cases · + Reranker12/12 passing

Run your own query

retrieval trace · live pipelinecorpus loaded · ready

The adversarial suite scores materially worse than the golden set, and that gap is the honest part. How the pipeline works →