Hard rules.

• Computer-based. All three modules compulsory, taken in order.
• Each module separately timed. Finish early and the time is lost, it does not roll over.
• You cannot return to a module once you have moved on.
• No calculator, no dictionary, no spell-checker. An erasable booklet is provided, and you can ask for more between modules.
• Exactly one correct answer per question, 1 mark each.
• No negative marking. Therefore never leave a blank. Ever.

Scoring. Critical Thinking and Problem Solving are reported separately, each on a scale of 1.0 to 9.0 to one decimal place, statistically scaled so versions of differing difficulty are equated. The writing task is not marked by UAT-UK, but a copy goes to every participating university you applied to, so an Oxford tutor reads it. Treat it as graded. Results land roughly four weeks after sitting.

An argument is reasons put forward as grounds for a conclusion. It is valid if the conclusion follows logically: accept the reasons and you must accept the conclusion.

1. Assume all reasons given are true. Never attack a premise for being false. The only question is whether the reasoning works.
2. The conclusion can appear anywhere, end, middle or opening sentence, with or without a marker. The test is that the conclusion is the statement supported by the rest, not the one doing the supporting.
3. Assumptions are unstated. An assumption is a step left out but required for the conclusion to follow. "She doesn't stand much of a chance. The polar bear is right behind her." Unstated: polar bears are dangerous.

Expect intermediate conclusions: a claim supported by the reasons which itself supports the main conclusion. Direction matters, and getting it backwards is the single most common trap in main-conclusion questions.

UAT-UK does not publish how many of each type appear in a 22 question set. Do not assume an even split or bank on any type being rare. Drill all seven at volume.

1. Identifying the main conclusion

Recognise: which option "best expresses the main conclusion".

1. Diagram reasons and conclusion.
2. Ask what the passage is trying to get you to accept.
3. Check for an intermediate conclusion. If a claim is supported by earlier reasons and also supports a further claim, it is intermediate. The main conclusion sits at the end of the support chain.
4. Match the option that states it without adding or losing scope.

Trap: an evaluative restatement of the reasons that sounds conclusion-shaped but only repackages support already given. Test: does this claim receive support from something else, or only compress what was already stated?

2. Drawing a conclusion

Recognise: what "can be concluded" or "follows", with no conclusion stated in the passage.

1. List every claim the passage asserts. Nothing else exists.
2. For each option ask whether accepting all listed claims forces you to accept it, with no extra assumption.
3. Reject anything needing an unstated assumption, overstating strength (all, always, primary reason, should), or resting on one claim rather than the passage as a whole.
4. If two options both follow, take the weaker, more literal one.

Trap: promoting a correlation or an average into a causal or universal claim. If the passage gives averages, the answer stays at that strength.

3. Identifying an assumption

Recognise: "underlying assumption", "assumes".

1. Diagram, then find the gap between reasons and conclusion.
2. Negate each candidate. If negating it collapses the argument, that is the assumption. If the argument still stands, it is not, however plausible.
3. Reject assumptions stated too strongly, using "the only", "the best", "always" where the argument needs something weaker.
4. Reject options that would be good weakeners but are not necessary for the argument to work.

Trap: a plausible background fact that is not load-bearing. The negation test is the only reliable filter, not plausibility, not topical relevance.

4. Assessing additional evidence

Recognise: "if true, most strengthens or weakens".

1. State the conclusion and the reasons in one line.
2. Identify the specific mechanism linking reason to conclusion, the exact gap the argument depends on.
3. For each option ask whether it touches that mechanism directly, or only a neighbouring fact, a different reason for the same advice, or an analogous case.
4. Pick the one that hits the mechanism, even over a more dramatic parallel.

Trap: an analogous outcome elsewhere mistaken for a direct rebuttal. It is suggestive, not load-bearing, unless the passage tells you the mechanism transfers.

5. Detecting reasoning errors

Recognise: "the flaw in the argument". Full method in section 04.

1. Diagnose before reading the options.
2. State the flaw in your own words.
3. Match your diagnosis to an option. Never read the five options first and pattern-match to whichever sounds most like a flaw.

Trap: an option describing a real flaw the passage does not commit. Getting the letter right without naming the flaw first means you got lucky. Flag it.

6. Matching arguments

Recognise: "most closely parallels the reasoning".

1. Strip content. Assign X and Y to the repeated ideas even where wording differs.
2. Write the skeleton using one of the four conditional forms.
3. If the passage's argument is invalid, the correct match must be invalid in the same way. Do not fix it by picking a valid-sounding option, that is the trap.
4. Ignore topic. Reorder the option back to reasons-then-conclusion before comparing, since the order is often scrambled deliberately.

Trap: an option on the same real-world topic with a different, often valid, structure.

7. Applying principles

Recognise: "which of the following illustrates the principle".

1. State the conclusion and reason in one line.
2. Generalise into an "if X then Y" rule that would justify that move in any case, stripped of topic.
3. Test each option: same rule applied to a different case, or a different rule that looks similar?

Trap: an option with the same moral flavour, say fairness, instantiating a different principle. Two wrongs don't make a right and reciprocity are not the same rule. Check the rule, not the vibe.

Parallel reasoning notation

Why they go wrong. Two failure modes. Label bait: you spot a familiar flaw word in an option, cause, sample, assumes all, and match it to the passage's general shape without checking the argument does it. Skipping the details: flaw passages carry factual sentences that are neither evidence nor conclusion, and you treat them as background when they are the answer key. CT-0887 had both. Option A used the word "cause", and the sentence about the academy figure covering only graduates who played for the first team while the transfer figure covered every player was the flaw, stated outright.

1. Skeleton. Evidence and conclusion, one line each. 3.1 versus 1.4 appearances, so the academy produces better players.
2. Name the gap. What must be true for that evidence to establish that conclusion? For any comparison the default question is whether the two sides are like-for-like: same population, same selection rule, same measure, same period, same conditions.
3. Detail audit. Go through every sentence that is not evidence or conclusion and mark it C if it closes off a flaw by telling you something is controlled, or O if it opens one by revealing a difference. Setters add C-sentences to kill distractors and one O-sentence to plant the flaw. Same match records for both groups is C. Last six seasons and same division are C. Academy figure covers graduates who played for the first team while the transfer figure covers every player signed is O. Squad size and the 1998 founding date are noise. The O-sentence is almost always the flaw.
4. Name it yourself before reading the options, in plain words: compares a success-filtered group against an unfiltered group.
5. Two-test each option. Test (a), does it happen? Point at the words where the argument does what the option says, or eliminate it. This kills most bait. Test (b), is it load-bearing? If the error were fixed, would the conclusion follow, or at least be far better supported? If fixing it leaves the argument just as weak, it is not the flaw.
6. Match the description, not the vocabulary. The correct option often avoids naming the flaw and just describes what the argument does. An option naming a textbook flaw is not more likely to be right for sounding technical.

Causal

Comparison and measurement

Sampling

Conditional logic

Relevance

Structure

Elimination heuristics. Run these on every CT question, after your own diagnosis and before the type-specific attack separates the last two. Kill any option that is:

• Out of scope: imports something the passage never mentions.
• Too strong: says all, never, must or only where the passage supports some or may.
• Reversed: the relationship, or the direction of support, runs the wrong way.
• True but irrelevant: plausible, even correct, but does nothing for this conclusion.
• Partial: answers a slightly different question, or captures half the conclusion.
• Doesn't happen: describes a move the argument never makes, or a problem the passage has explicitly closed off.

Two options usually die on sight. Separate the final two by going back to the exact wording of the stem and passage, never by which sounds better.

Booklet shorthand. → supports, ⇒ main conclusion, IC intermediate conclusion, ¬ negation, C and O on detail sentences, X and Y for repeated ideas in matching arguments. Cross out option letters as they die. Once a type is above 90 per cent untimed, drop the full write-out for that type and keep the one-line skeleton.

Time model. 40 minutes over 22 questions is about 1 minute 50 seconds each. Short main-conclusion and drawing-a-conclusion items should take 60 to 75 seconds, which buys time for flaw, matching arguments and applying principles, which run past 2 minutes. Hard stop: if two options survive after 2 minutes 30 seconds, pick the one that passes test (a), flag it, move on.

Universal routine. Run this on every question before the operation-specific attack.

1. Read the stem first, then the stimulus. Know the target quantity before touching the data.
2. Name the operation. That tells you which attack to run.
3. Circle the constraint words. Most careful-reader errors live there.
4. Estimate, solve, check against the options. If your answer is not among them you have misread a constraint, so do not recalculate the arithmetic first.
5. Ask what each wrong option was built from. Distractors are built from specific mistakes, so landing on one usually tells you which mistake you made.

Operation 1: relevant selection

Recognise: far more data than needed, often a table with several conditions.

1. Write the target and every condition as a checklist.
2. Build a pass or fail grid, one row per item, one column per condition.
3. An item qualifies only if every cell is ticked. Sum only those.
4. Note which wrong option corresponds to including which failed item.

Trap: including an item that passes the condition you checked most carefully but fails another. Every condition gets a column.

Operation 2: finding procedures

Recognise: you have the data but no obvious route.

1. List what you have and what you need, in order.
2. Decompose. Most multi-step items follow a pattern: remove the fixed component, split the remainder by the ratio, apply the separate percentage changes, recombine.
3. Check the sequence before calculating: right base for each percentage, units converted, direction of change.
4. Calculate last. Factor and cancel before multiplying.

Trap: applying a percentage to the wrong base, or splitting a fixed fee by a ratio. With your maths background the extra risk is over-engineering, reaching for algebra where a ratio or inspection does it in 20 seconds.

Operation 3: identifying similarity

Recognise: the same information shown two ways, bar against pie, table against graph, net against solid, or a spatial manipulation.

1. Read each representation independently before comparing.
2. Find the structural constraint that must hold for them to agree, for example two segments summing to half the total.
3. For rotations, reflections and folds, fix one anchor point or edge and track only that.

Trap: matching on surface appearance rather than the numeric relationship.

Constraint words, read literally

Maths assumed

Simple fractions, place value, percentage concepts. The four rules, percentage operations, everyday calculations, the mean. Time and the calendar, money, measures. Conversions required: 1 km = 1000 m, 1 m = 100 cm, 1 cm = 10 mm, 1 kg = 1000 g. Other units may appear but their conversions are given. Area, perimeter, volume. Extracting information from graphs, charts and tables. Nothing above GCSE and no complex fraction or decimal work, so if a problem seems to need heavy algebra you have missed the intended shortcut.

Genres to drill

Percentages, ratios, proportion. Rates: speed, distance, time, work rate, filling. Money: best value, discounts, bills, hire charges. Counting and basic combinatorics. Area, perimeter, volume and scaling. Spatial reasoning: nets, folding, rotations, shortest routes, dice. Timetables, calendars, scheduling, date windows. Logic-grid deduction. Table, chart and graph reading. Number puzzles and sequences. Grouping, where the greatest number of identical groups is the highest common factor.

No-calculator technique and timing

Estimate first, since the options are usually well spread and a rough figure often kills three or four. Back-solve by substituting options when forward working is slow. Round then correct. Factor and cancel before multiplying. Convert units early. Hunt the structural insight before grinding. Sanity-check magnitude.

Under 2 minutes each. First pass banks everything that yields in under 90 seconds. Flag anything with no traction after about 30 seconds. Second pass takes the flagged set, hardest last. Final 60 seconds: fill every blank. With no negative marking a blank is strictly worse than a guess.

Three prompts, answer one. Each typically gives a short statement and asks you to explain what you think is meant by it, give a reasoned argument against it, and say to what extent you agree. Max 750 words, 40 minutes, no subject knowledge required. Official example: democratic freedom means there should be no restriction on what may be said in public.

All three parts must be clearly and separately addressed. Missing part two, or treating part three as a summary, is a structural failure however good the prose.

1. Choose, 1 to 2 min. Pick the prompt where you already have a defensible thesis and two concrete examples, not the most familiar topic.
2. Define, in the plan. Find the one or two words carrying the weight: democratic freedom, no restriction, in public. Part one lives or dies on reading these precisely and noting where they are ambiguous.
3. Plan, 5 to 8 min. Thesis in one sentence, a real position with a stated limit, X except where Y. Strongest case against, two or three distinct points from different angles rather than one restated. Your case, two or three arguments each with a load-bearing example. The single best objection to your view, and your reply.
4. Write, 27 to 30 min. Part one, one paragraph: interpret charitably, identify the ambiguity, set up the tension. Part two, one or two paragraphs: make the argument against as strong as its best defender would. Part three, two or three paragraphs: state your position, argue it, answer the strongest objection, and refer back to part two explicitly rather than re-summarising. Conclusion, two or three sentences, decisive, no new material.
5. Review, 4 to 5 min. Word count, grammar and typos, since there is no spell-checker, and check all three parts are visibly present.

The bar for PPE: a dialectically mature 750 words. A non-trivial thesis, the strongest opposing case engaged, load-bearing examples, controlled prose. Your economics and philosophy help only when the content serves the argument, never as a substitute for reasoning.

1. Official first. The Content Specification and Question Guide, plus the two practice tests on the Pearson VUE landing page, which mirror the live format and software. Save both for full timed conditions close to test day.
2. Your own bank and site, tara90.app. 1,000 CT and 1,000 PS items, timed drilling, mocks, error log, retest queue. This is your main volume.
3. TSA past papers, Oxford, 2008 to 2024. CT and PS styles are near-identical.
4. BMAT Section 1 papers, which UAT-UK flags as sharing a common history with the TARA. Filter to the seven CT types and three PS operations, since some BMAT styles do not appear.

Error log categories. Use these consistently so the retest queue targets the skill, not just the question. Misread stimulus. Missed the conclusion. Assumption not tested. Scope error. Too-strong option. Reversed relation. Label bait, matched flaw vocabulary without checking the argument does it. Detail not audited, skipped the sentence that opened or closed the flaw. Arithmetic slip. Unit or conversion error. Wrong figure selected. Constraint word misread, boundary day, ceiling against floor. Billing-period model, calendar days against elapsed blocks. Method not found. Time pressure. Careless.