An impressive demonstration is a starting point for questions. What task was evaluated, what information was available, and what counted as success? A result on a narrow task does not automatically describe performance in a different setting.
Read the comparison as carefully as the score. Were systems given comparable resources? Was the test separate from development? Are failures, uncertainty and costs reported? The HELM research programme illustrates why evaluation can benefit from multiple dimensions rather than a single headline number.
We distinguish what authors report from independent confirmation and from our interpretation. When reading a new paper, look for the methods, the available evidence and the questions the experiment leaves unanswered. A clear limitation is useful information, not a reason to ignore a result.
Sources & further reading
Links point to original reporting or primary research.

