Every result in a testing record concerns material submitted to a laboratory at a point in time.
Every result in a testing record concerns material submitted to a laboratory at a point in time. The relationship between that material and any individual order depends on sampling and handling, which a record cannot establish.
Composition and safety are different questions requiring different work. A record of what a sample contained says nothing about whether it is safe for any use, and no volume of composition data converts into a safety conclusion.
A long run of results within specification makes the next result more predictable and does not make it certain. Processes drift, suppliers change, and the record shows history rather than a promise.
If only some runs are tested, untested runs are unobserved rather than good. The informative figure is the proportion tested, which is rarely published and is the number that would make the rest interpretable.
Hash-chaining prevents silent revision of past entries. It says nothing about whether an entry was correct when written, and presenting it as a guarantee of accuracy overstates a real but narrower property.
A result hosted on a laboratory's own portal is harder to fabricate than one hosted by a seller. It still depends on the sample having been representative, which the laboratory cannot verify either.
Answering whether a specific claim held on a specific date, and whether a publisher's statements have been consistent over time. Those are narrow questions and they are answered well.
A record presented as proving more than it does invites exactly the scrutiny that would discredit the parts that are sound. Naming the limits is what makes the remainder usable.
A record presented as proving more than it does invites exactly the scrutiny that discredits the parts that are sound. Naming the boundary is what makes the remainder usable, and sources that do it are easier to rely on precisely because they have told you where to stop.
Randomly selected samples taken by someone other than the publisher, a stated proportion of production runs tested, and stability testing over time. Each is checkable, none is present in most published records, and listing them is more useful than asserting rigour.
The claims on this page about the limits of a dated testing archive, stated plainly can be checked against sources outside this site: