Labs that really run

Change the chunk strategy and size, and see how often the answer survives in one piece.

Paste text and see how a model splits it into tokens, and why sizes are counted in tokens.

Ask a question and see keyword and vector search, the prompt, and the answer.

Replace or patch a changed document and see which chunks go stale.

Compare two sentences by similarity, and see one question scored against every chunk.

Keyword, vector and hybrid search side by side, with the rank of the answer in each.

Pick chunks, a token budget and an order, and see the prompt, where the answer lands and the cost.

Recall, precision, MRR and nDCG for each retriever over all 47 golden questions.