Snowflake Data Scientist Interview Experience — Built a Streamlit App Instead of LeetCode, Rejected After Onsite

Got in through a referral.
Hiring Manager interview — the whole thing was just chatting about my project, it went pretty well, and I moved forward to the next round.
Technical round — I'd read a report on the forum saying they usually ask 3sum and some fairly hard SQL, but it turned out to be building a small data application with Streamlit instead. It really did require partition SQL, plus being comfortable with Python for plotting and other data-analysis work. The interviewer was very nice — if there was something I didn't know, he just told me to go ahead and google it. Had to share my screen the whole time.
Onsite — the data science round had three interviewers total: someone fairly senior who covered my data science background, a data engineer who covered data modeling, and a data scientist who covered a product case.

Just got the rejection email. It was probably the product case round that I failed — I probably didn't articulate the metric trade-offs well enough, including the statistical reasoning behind choosing one metric over another.
A little disappointing, but not really surprising. The HR person was very efficient and friendly.
Actually, if I'd passed that round there would still have been a presentation, plus a cross-functional interview, and a hiring manager interview.
The rejection letter said they went with other candidates.