Langfuse vs GitHub

The signal in Langfuse's recent feed is split: a real product move — promoting Experiments to a top-level feature alongside Datasets, with multi-run comparison and progress tracking — and a smaller LLM-as-a-Judge upgrade adding boolean true/false scores. Everything else surfaced in the changelog is GitHub star milestones and contributor counts, which inflate the feed without conveying any product change.

Langfuse is sharpening the eval surface — Experiments becoming a first-class concept and judges getting boolean outputs both point at making LLM testing more rigorous and decision-grade, not just observational. The community-metric noise dilutes how the actual product cadence reads from the outside, but the substantive cadence is steady on the eval/observability axis.

The next likely move is more depth around Experiments — comparing across model versions, prompt variants, or judges, plus tighter wiring to CI for regression-style LLM testing. Expect more judge configurations (numeric ranges, multi-class) to follow the boolean addition.