Two experienced reviewers frequently publish contradictory assessments of the same wine, which readers take as evidence that criticism is arbitrary. Several concrete mechanisms explain most of the gap.
Palates differ physically
Sensitivity to specific compounds varies substantially between individuals, and some tasters cannot detect certain molecules at all. Others register the same molecules at very low concentrations.
Bitterness perception in particular varies widely across people. The same tannin can therefore read as firm structure to one taster and as outright harshness to another sitting beside them.
These differences are genetic and not correctable by training. Experience teaches a taster to describe what they perceive, not to perceive what someone else does.
Conditions are rarely comparable
A wine tasted from a barrel sample, at a producer's cellar, in a competition flight and over dinner will present differently in each setting.
Temperature, glass shape, time since opening and the wines tasted immediately before all shift the reading measurably.
Reviewers who publish their conditions make their assessments comparable. Those who do not leave readers unable to tell which differences are the wine's.
Bottle variation is real
Corks vary in how much oxygen they admit, and two bottles from one case can develop noticeably differently over years of storage. The variation compounds with time.
Cork taint affects a small proportion of bottles and can be present at levels that dull a wine without being identifiable as a fault. A reviewer may simply find the bottle mute.
For older wines this variation grows large enough that reviewers sometimes describe two bottles of the same wine as different wines entirely.
Stylistic preference is a legitimate variable
Critics differ on how much oak, ripeness, extraction or volatile character they consider desirable. Those are aesthetic positions rather than errors.
A reviewer who consistently rewards restraint and one who consistently rewards power will diverge predictably on wines sitting at either extreme.
The useful response is to identify which reviewer's preferences track your own, which makes their scores informative regardless of whether you agree with them.
Timing changes the answer
A wine assessed at release and reassessed a decade later is being judged at different points on its curve.
Reviewers who publish revisions are correcting for that, and the revised note is usually the more reliable one.
Comparing an early score against a later one and treating the difference as inconsistency mistakes a moving subject for a fixed one.