WHAT IT SHOWS
Where reranking helped—and where it destroyed a correct answer.
The article, row explorer, and recorded output expose the same frozen experiment at different levels. A live GPU run can test whether its selections still agree.
LAB 05 / MEDICATION RESOLUTION
A frozen benchmark follows one resolution pipeline from plain search to medical-LLM reranking—and records why the first model-assisted version was rejected.
Research artifact. Not clinical guidance.
01 / THE LARGER CAPSTONE
These field notes cover step 3 only. Input realism, confirmation, interaction retrieval, and grounded output belong to the later capstone.
02 / INSIDE STEP 3
Frozen benchmark: 100 products / 603 queries. Each result determines the next experiment.
FIELD NOTE 1 / V0 → V2
Plain Typesense retrieval scored 506 / 603. Always-on MedGemma reranking scored 495. The useful result is in the 11 fixes and 22 regressions between them.
WHAT IT SHOWS
The article, row explorer, and recorded output expose the same frozen experiment at different levels. A live GPU run can test whether its selections still agree.
WHAT IT DOES NOT SHOW
The evidence covers one model, prompt, Top-5 policy, 100-product catalogue, and synthetic-noise benchmark. Downstream interaction grounding is not yet built.