Practical Workplace English Assessment for Office, Service and Support RolesNot whether your grammar is right. Whether the other person does the right thing.

Methodology: Thirty-four original exercises across five formats: fourteen keyed single-choice exercises on real workplace messages, six select-every-that-applies exercises, six binary claims, four matching exercises and four ordering exercises in which the steps of a written instruction have to be put in the order the instruction gives. One response instruction covers the whole form and it is a knowledge instruction: what does this message actually say, and what would be the most effective thing to do about it. Construct statement: it measures practical English at work - whether somebody extracts the binding detail from a written message, carries out a written instruction in the order it gives including its conditions, writes so that a busy or second-language reader acts correctly, and can tell when a request is too incomplete to act on. It does not measure spoken English, accent, pronunciation, listening, vocabulary size, grammar in the abstract, general intelligence or suitability for any job, and it is not a visa, university or immigration test. Placement is referenced to the published Council of Europe framework levels and that framework is cited, not licensed: there is no accreditation, no endorsement and no affiliation with any language test provider, examination board or publisher, and no employer or institution is obliged to accept the result. Scoring is a reliable change index, which is the point of the product. The current placement is reported with its standard error of measurement, computed from the corrected score and a stated assumed test-retest reliability, and the report then prints the threshold a second sitting would have to cross for the difference to exceed measurement error at ninety-five per cent: RCI = (post - pre) / (SD * sqrt(2 * (1 - r))). Publishing that threshold on a FIRST sitting is unusual and it is deliberate - it is the number that tells somebody whether a course was worth taking, it costs nothing to compute, and a retake product that does not print it is selling the same test twice. Every corrected figure is corrected against an authored answer prior per option rather than against a uniform random draw. Those priors are a model, they are printed, no percentile appears anywhere and none could be claimed until real sittings replace them. Reliability is reported as McDonald's omega estimated from the item count and a stated assumed inter-item correlation; a strand whose omega will not carry a number is given a three-way placement and the refusal is printed rather than performed quietly. Constructs and sources: the Council of Europe Common European Framework of Reference and its 2020 companion volume, cited as a public framework; functional and task-based views of language ability (Bachman and Palmer 2010) and the argument that a language test should predict what somebody can do rather than what they know; readability and sentence length as the dominant lever on second-language comprehension (Crossley, Allen and McNamara 2011); the reduced-processing account of how non-native readers handle complex syntax (Clahsen and Felser 2006); instruction-following and the cost of conditional clauses in procedural text (Dixon 1982); plain-language effects on compliance and error in workplace documents (Kimble 2012); the reliable change index and its distinction from a minimally important difference (Jacobson and Truax 1991; Christensen and Mendoza 1986); test-retest reliability as the basis of an individual change statistic (Charter 1996); implementation intentions and the effect size of a specific if-then plan (Gollwitzer and Sheeran 2006); and score-report design for a lay reader (Zenisky and Hambleton 2012). All items are original works. No published test's items, task types, scale names, band descriptors or report layouts are reproduced or implied.