TMUA October 2026 Recap: What the Recall Data Actually Shows About Form Overlap, Repeats and Results Day

TL;DR

• Every October form came from one shared international question bank, and overlap between forms was extremely uneven: about 35%, about 10%, and one form that overlaps no other form at all (0%). Your region does not decide your paper; your form does. This one fact explains most of the same-day arguments: two candidates sitting the same morning were given questions that partly differed from the start.
• Cross-year repeats are real but small: only about 3-7% of the October paper traces back to the 2025 paper, and genuinely new questions matched zero of all past papers from 2016-2024. If you are eligible for the January sitting, expect at most a handful of repeats — anyone selling “this year’s paper” is selling something useless to you.
• The difficulty consensus is converging on “about the 2022/2023 level”: Paper 1 was hard and polarising, Paper 2 leaned logical and was more doable — and friendlier than the community mock papers most candidates had been drilling.
• Results are expected in mid-November, accessed through your UAT-UK account. Until then, ignore the raw-to-scaled conversion tables and “answer keys” circulating in the community — both are far less reliable than they look. Practical advice is in section three.

Everything below comes from cross-checking the recall documents that circulated after each sitting. We deliberately do not reproduce any question or name any source — strip those away and the aggregate numbers still hold.

1. What Actually Happened on Test Days

The Paper 1 / Paper 2 split dominated the experience. Every sitting followed the same pattern: Paper 1 hard and time-pressured; Paper 2 logical and more doable — with huge individual variation. That variation is not noise; section two explains it as the inevitable result of candidates on the same day receiving different question combinations.

What the papers tested. Paper 1 leaned on functions — exponentials and logs, modulus graphs, and “how many real solutions” questions were the most-named across sittings. Paper 2 was almost uniformly logical in flavour: spotting the flawed step in a proof, multiple necessary-and-sufficient judgements, and counterexamples. If you are sitting in January, train to that shape — every form we could check follows it.

Community mocks ran high — much higher. Multiple candidates who consistently score 18-20 on official past papers reported only 14-15 on community mock papers (JZ and others). The recurring verdict: community mocks are inflated in difficulty. If a mock score knocked your confidence, this sample says your real score was probably higher, not lower.

Difficulty has no single answer — and that is itself the finding. On overall difficulty, the consensus is converging on “comparable to 2022/2023”; several candidates said exactly that. At the same time, one UK candidate called it “harder than every official past paper”, and another said “a bit easier than 2022, nothing like the JZ set”. Both can be true — because their forms differed.

Loud leak claims, thin evidence. Tracing the “the paper leaked beforehand” claims leads back to a handful of sources, none confirmed. What can be verified is that recalled “answer keys” are unreliable: in one thread, candidates split three ways on a single trigonometric integral and two ways on an infinite series; and at least one widely shared recall compilation changed a question’s wording in a later revision, flipping its answer. Treat any circulating key as a coin flip, not an estimation tool.

2. What the Data Says: How the Forms Were Assembled

One bank, unevenly drawn. All forms came from one international shared pool. This sitting, one UK vs China/Hong Kong pair of sittings shared about a third of their questions, another pair about 10%, and one form overlapped no other form at all. 2025 showed the same pattern (UK vs CN: 10%). Your region does not decide your paper; your form does.

Cross-year repeats are small. About 3-7% of October questions trace back to the 2025 paper (two independent reference banks). The January sample is higher, about 16% — winter sittings recycle more old questions than October. Of the 88 questions used in 2025, only about 12.5% reappeared across the two 2026 cycles. Against all past papers from 2016-2024, October’s genuinely new questions scored zero matches — the bank is being topped up with new material, not recycled. A tiny fixed core exists (one quadratic vertex question has now appeared in five sittings across two cycles, and a few more span three cycles), but one question is not a paper.

No evidence of per-candidate random variants. A popular claim holds that each candidate gets the same questions with different numbers. We could not confirm a single case. Every seemingly “variant” question pair fell apart into recall noise once checked.

Practical takeaway: single-digit percentages of a new paper come from recent papers. Buying a “leaked October paper” for the January sitting buys you almost nothing — and under UAT-UK’s rules on exam content, it could cost you your application.

3. What to Watch Before and After Results

Before results (expected mid-November, via your UAT-UK account; timing and how to check: the October 2026 results page):

1. Ignore circulating raw-to-scaled conversion tables. They contradict each other — 32/40 becomes 6.5 on one table and 7.0 on another — and scaling is done separately per form, so outside UAT-UK nobody can convert precisely. We do not apply the 2025 model to 2026 either: our conversion page covers 2016-2025, and once results are out we will publish a table that tracks the actual 2026 distribution. What the historical tables can and cannot tell you before results day is set out in our pre-results estimate guide. There is no reason to panic over any single-number table.
2. Do not treat a community mock score as a forecast. Use section one’s directionality in reverse: candidates who steadily score 18-20 on official past papers often reported only 14-15 on community mocks. If your mock score looked bad, your real score probably sits closer to your official-paper level.
3. Wait for results, not for “answer keys”. Recalled answers are unreliable — candidates split three ways on one question, and a revision flipped an answer — and every pre-release conversion is only a reference. Your 1.0-9.0 score is computed by UAT-UK per form; it never passes through any third-party table.

After results:

1. First understand where the number comes from. The October median is fixed at 4.5 and the October top-10% boundary at 7.0; those two points pin the whole conversion line, so your score is directly your position within the October field. Our scoring explainer covers the full mechanics.
2. A score only matters for your course. TMUA has no universal admissions cutoff: Cambridge, Imperial, LSE and Warwick — joined for 2027 entry by Oxford Mathematics and Computer Science and others — each decide how to use it. Panic or euphoria against national distributions gets you nowhere.
3. If you are eligible for the January sitting, prepare for the real January. The data shows January papers recycle more old questions than October (about 16% in our sample), but spread across 40 questions that is still only a handful — your preparation should rest on the official past papers, not on hoping for repeats. See October vs January results for how the two sittings compare.

Method and Limits

Recall documents are written from memory after the exam and carry sampling bias: the people who post, and the sittings that get reposted, are over-represented. All percentages are aggregated per sitting and do not describe any individual candidate’s paper. Each figure was cross-checked against multiple independent reference banks and the directions agree; once UAT-UK publishes this cycle’s technical report, we will re-check the aggregates against the official parameters.