Quality resources
Most quality statements describe an intention. This one describes a mechanism, because an intention cannot be tested and a mechanism can.
A guaranteed analysis lives on the website, in the formulation handbook a co-manufacturer works from, in the label copy a packaging designer sets, and in the specification behind all three. Left to proofreading, those four copies diverge — and the divergence is invisible, because each document is internally consistent.
So they are not left to proofreading. The copies are compared automatically, and a disagreement stops the change rather than reaching a label. Reasoning documents carry no figures at all: two copies of a number drift, and two copies of an argument do not.
What is checked
Each of these traces to a defect that occurred here and was caught. They are not hypothetical failure modes.
The statement on the specification and the statement on the panel are compared word for word and in order — not merely for the same ingredients, but the same sequence, because descending order by weight is a regulatory requirement rather than a formatting preference.
A minimum above what the formula delivers cannot hold on assay, and a maximum below it cannot either. The formula's actual output is recorded alongside the guarantee so the comparison is made mechanically rather than remembered.
As fed, dry matter, per 1,000 kcal, and as a share of energy. A figure can satisfy one basis and breach another — a calcium maximum of 1.7% as fed is 1.889% on a dry-matter basis — so all of them are computed, not the convenient one.
Calories per serving must equal energy density multiplied by the weight of that serving. A feeding table computed against the wrong density understates every amount on it, and nothing else in the document disagrees.
A complete food carries a life-stage adequacy statement and names the profile it meets. A treat carries none, and instead carries the intermittent-or-supplemental-feeding statement its exemption depends on.
No specification may carry language that would make the product a drug, and no livestock product may describe itself as acting through the immune system. The boundary is checked in the copy, not left to whoever writes it.
Every check added here is verified by breaking the thing it guards and watching it go red, then restoring. That step is not ceremony: checks have been written here that could never have failed — one threw before reaching its assertion, another compared a value in a way that silently stopped matching.
A check nobody has seen fail is trusted anyway, which makes it more dangerous than an absent one. The same discipline governs what is published: figures shown before production batches exist are development targets and are labelled as such, because a guaranteed analysis is derived from assay of finished batches and cannot be claimed before there are batches to assay.
