Payroll and Pay-Run Accuracy Assessment for Payroll, HR Operations and Shared ServicesEverybody can find the error afterwards. The question is what it costs to stop the run for it.

Methodology: Thirty-five original exercises across six formats: ten pay-run sweeps, eight keyed questions, six binary claims, five quantity estimates, three matching exercises and three ordering exercises. The declared response instruction is applied judgement throughout - what would you do with this run - and one instruction covers the whole form. Construct statement: it measures whether somebody can tell a line that must stop a pay run from a line that is worth a note and a line that is ordinary, whether they can follow a figure from gross to net, whether they know what a starter, leaver or mid-month change does to a record, and which control catches which failure. It does not measure knowledge of any country's tax law, any named payroll system, or speed of keying, and it is not a test of integrity. Scoring is a decision-cost matrix rather than an accuracy percentage. Every line in the ten sweeps carries a class - stops the run, worth a note, nothing wrong - and a published cost for each of the two actions against each class: holding costs nothing on a stopping line, two on a note and three on an ordinary line; releasing costs six on a stopping line, one on a note and nothing on an ordinary line. The costs are a value judgement about which error hurts more, not a measurement, so they are printed on the report rather than left in the code. The total is normalised against a modelled typical reviewer drawn from the authored answer priors on each line, so zero means indistinguishable from somebody reviewing the way people review. The two fixed habits are calibrated against explicitly: holding every line and releasing every line both cost more than the typical reviewer and both land below zero, and the report prints what each of them would have scored on the reader's own sweeps. The number of stopping lines varies from one to three across the ten sweeps, so counting to a fixed number of holds is not a strategy either. Every other strand is corrected against its own authored priors: per option on the keyed and select-every questions, per claim on the binary ones, and as a declared distribution of the answers people actually give on each quantity, because a slider has no options and a uniform draw across its range is not a null anybody produces. Each of the four parts carries enough exercises for a placement and not enough for a precise figure, so each is reported as a three-way placement with its reliability printed beside it and the refusal stated. Constructs and sources: the internal-control literature on segregation of duties and independent authorisation, and the finding that self-review is the control weakness most often present in payroll fraud and error cases; the auditing distinction between an existence control and a completeness control, which is why a control total and a variance report catch different things and neither catches an error repeated from the previous period; signal-detection and decision-theoretic treatments of inspection, where the operating point is set by the relative cost of a miss and a false alarm rather than by accuracy; Swets, Dawes and Monahan on publishing the cost structure behind any screening decision; the human-factors result that error rates in repetitive checking fall sharply when exceptions are listed rather than searched for; the overpayment-recovery literature showing recovery rates fall steeply with elapsed time and with the recipient's income; the protected-earnings principle common to statutory deduction regimes, stated as a principle rather than as any one country's rule; and Haladyna, Downing and Rodriguez on plausible distractors and cue control. All exercises are original works; no commercial instrument's items or name is reproduced, no payroll product is named, and no affiliation is implied.