A factor of a hundred and forty seven
On this page11
Split the recorded weights by material and the pattern is obvious
The 55 glass rows run from 0.1288 to 2,002 pounds with a median of 7.9975. The 117 PET rows run from 0.02 to 28.83 with a median of 0.0544. The 77 polypropylene rows run from 0.01 to 21.01 with a median of 0.0456.
The glass median is a hundred and forty seven times the PET median
For containers covering the same capacity range on the same shelf. Glass is denser than PET and a glass wall is thicker, which puts the same vessel at roughly two to three times the mass. Not a hundred and forty seven.
So the unit of the column correlates with the material
Which is a supplier side pattern rather than a physical one. Materials do not determine whether a figure is per unit or per case; a data entry convention does.
The plastic halves of the column look internally consistent
A median of 0.0544 pounds is about 25 grams and 0.0456 is about 21. Both are credible single container weights for a small jar, which suggests the plastic rows are mostly recording one unit.
And the glass half mostly is not
A median of eight pounds cannot be one small glass jar. The 0.1288 row and a handful of others sit at plausible unit weights while most of the glass rows sit two orders of magnitude above them.
Both plastics also have rows that break the pattern
The PET maximum is 28.83 pounds and the polypropylene maximum is 21.01. So the convention is not even consistent within a material, which rules out correcting the column by a per material factor.
Exactly one row on this shelf exceeds fifty pounds
The 2,002 pound figure. Which means a filter that excluded implausible weights above fifty pounds would remove one row and pass every other affected one, because eight pounds is not obviously absurd.
That is the uncomfortable part and it is worth stating plainly
The worst row is the easy one to catch. The ones that cause harm are the rows that look reasonable and are a case weight, and no threshold separates those from a genuine unit weight.
Which is why the answer is not to repair the column
It is to stop using it. Four of the 253 rows record nothing at all, and a blank is a better state than a plausible wrong figure, because a blank cannot be mistaken for an answer.
And nothing on these pages depends on it
Every other figure quoted across this property comes from a field that states what it is: a capacity in milliliters, a height in inches, a case pack as a count, a price in cents.