[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"article-body-sampling-was-a-concession-not-a-method":3},"\nAudit sampling has a birthday. It was born the day populations\noutgrew ledger-paper arithmetic, and it was never a method. It was\na concession. Examining everything was impossible, so the\nprofession built a rigorous discipline around examining a\ndefensible fraction. The rigour was real. But it was rigour about\na constraint, and somewhere across the decades the constraint\nbecame invisible and the concession got promoted to a method.\n\nThe constraint is now dead. Transactions, approvals, master-data\nchanges and access grants are all events a machine can test\nagainst a control rule, continuously, at population scale. The\nhabit survived, and the habit has started to need defending.\n\n## What a sample cannot see\n\nBe fair to sampling first. Against systematic failure, a designed\nsample works, because a control that fails everywhere fails in any\n25 items you pick. But mature control environments rarely fail\nsystematically. They fail episodically (e.g. an override used\neleven times in one bad week, or\n[an automated control that silently regressed](/insights/control-automation-went-backwards)\nin March and was fixed in May). A sample's odds of catching an\nepisode are roughly the episode's share of the population, which\nis to say close to zero. The workpapers will still say \"no\nexceptions noted\". That sentence is true and unhelpful at the same\ntime, and it is all a sample can leave behind.\n\n*Test the population. Explain the sample.*\n\n## The mechanics of testing everything\n\nPopulation testing is not a bigger sample. It is a different\nthing. The control is written as a rule the data must always obey.\nEvery payment matches an approved order, within tolerance, and no\none both requested and approved it. Every access grant has an\nowner and an expiry. The rule runs against every event as it\nhappens. Exceptions surface as they occur, each with an owner and\na clock, while the millions of conforming events\n[route to nobody, by design](/insights/millions-of-events-in-a-handful-of-decisions-out).\n\nThe evidence changes shape with it. Instead of \"we tested 25 and\nfound nothing\", the file says \"we tested the population; here are\nthe eleven exceptions, their dispositions and the dates\". One is\nan inference about what probably holds. The other is a record of\nwhat did.\n\n## What auditors do when machines test everything\n\nPopulation testing does not shrink audit judgement; it relocates\nit. Someone must write the rules, and writing them is control\ndesign, the discipline's actual centre. Someone must judge the\nexceptions, which is where experience earns its keep. And someone\nmust decide where sampling still belongs, because it does,\nwherever the question cannot be written as a rule (e.g. estimates\nand fraud hypotheses). The inversion is only this. The sample\nbecomes the justified exception, not the default described as\nrigour.\n\nProphesee runs continuous assurance the way the argument implies.\nControls live as always-on rules over one shared stream of events,\nexceptions arrive with owners and deadlines, and the audit trail\nis generated by the testing itself rather than assembled for the\nvisit. Assurance as a by-product of operating is the part periodic\ntooling cannot imitate, whatever its testing calendar says.\n[Start here](/contact).\n",1786984937178]