A score published by a competition or a panel is not one person's opinion, and it is not a simple average either. The procedure used to combine assessments is what actually generates the number.
Independence comes first
Panellists taste and score without discussion, because a spoken opinion influences everyone who hears it before they have decided. The first person to speak carries disproportionate weight.
Wines are presented blind and usually in flights of similar style, so that a taster is comparing like with like. Bottle order is varied between panellists where practical.
Scores are collected before any conversation begins, which preserves genuinely separate judgements to work with.
Disagreement is handled by rule
Panels define in advance how much spread between scores is acceptable before a wine is re-examined. A tight cluster is accepted and a wide scatter is not.
Where scores diverge sharply, the wine is normally re-tasted, often from a fresh bottle in case the first was faulty. Bottle variation explains a meaningful share of disagreements.
Only after re-tasting does discussion happen, and even then some systems retain the individual scores rather than negotiating a consensus.
Averaging conceals the useful information
A wine scored consistently in the middle by everyone and a wine that divides a panel sharply can produce identical averages. Those are entirely different wines.
Distinctive, polarising styles suffer most under averaging, since they collect both high and low marks. Inoffensive wines accumulate solid middling scores.
This is a known weakness of panel systems, and some publications report the spread alongside the average to counteract it.
The chair shapes the outcome
A panel chair sets the pace, decides when a wine is re-tasted and manages discussion. Those procedural choices affect results even when the chair never scores differently.
Calibration exercises at the start of a session, where everyone tastes a reference wine, exist to align how the scale is being used.
Without calibration, panellists apply the same numbers to different standards, and the average combines incompatible scales.
What a panel score can and cannot tell you
Panel scores are reasonably good at identifying faulty and poorly made wine, since faults produce agreement quickly.
They are less good at identifying wines that reward attention over an evening, because the format allows a minute per glass.
Read as a screening tool rather than a verdict, the numbers do useful work. Read as a ranking of pleasure, they promise more than the method supports.