Questions to examination standard. Then solved.

The method

PLEW, the method underneath

The engine is built on PLEW, a reading method developed in our classrooms in 2019 and since patented. PLEW extends the Toulmin model of argument to the examined passage: every passage is decomposed into four structural elements. The Purpose sentence carries the main claim. The Logic is the inferential structure connecting claims. The Evidence is classified by type and by its logical relationship to the claim. The Weaknesses are the concessions and qualifications the writer admits. A student trained on PLEW reads an unfamiliar passage the way an examiner reads it — knowing what each sentence is for. The same analysis, run in software, is what lets the engine write questions that behave like real ones.

The decomposition is not an exercise in labelling. Every examined question type tests one of the four elements, and the mapping is exact:

PLEW is in commercial use. Korean education companies are building integrations on it; the most recent delivery is the solutions sheets produced for A Dot's Dshare platform.

The indicators

What makes a question hard, measured

PLEW is the first of two layers. The second is a set of reasoning indicators derived from the Thinking Skills Assessment — the reasoning test developed by Cambridge Assessment for Oxford admissions — and adapted for examined English. Six operations are scored on every passage and every question: assumption identification, stated and unstated; parallel-structure recognition; the distinction between cause and correlation; counter-argument handling; analogy evaluation; and the logical distance between evidence and claim.

In full, the difficulty model runs on twenty-two indicators across two dimensions, passage and question, combined into a single predicted answer rate. On the passage side: sentence complexity, vocabulary tier, logical chain length, idea density, assumption depth, and whether the main idea is stated, implied, or distributed across the passage. On the question side: distractor design, the number of inference steps between passage and correct answer, and the paraphrase distance between the passage's language and the language of the options.

The illustrative passage above, profiled on six of the twenty-two indicators.

The taxonomy is published; the calibration is proprietary. The weight each indicator carries was not set theoretically — the weights are derived from the responses of the TARA research cohort, more than 5,000 students assessed on the same reasoning framework. Cambridge Assessment and Pearson publish their construct definitions and withhold their item parameters. The engine operates on the same principle.

The functions

The functions, as a specification

The engine's format descriptions follow the convention of the tests themselves.

Coverage: 수능 and 내신 English, TARA format, LNAT format. 10,000+ questions produced to date.

The paraphrase levels are defined by logical distance — whether the inference chain survives the rewrite — not by how different the surface text looks.

The measure

Using it

For students. For companies.