Matching information ("Which paragraph contains…") and matching features or names ("Which researcher said…") share one defining property: the questions do not follow passage order. That single fact breaks the sequential habits that work everywhere else, and rewards a different procedure — work the question list, not the passage. This lesson installs that procedure.
Why these types punish sequential reading
For most reading questions, question 4's answer sits below question 3's. Locate, answer, move down. These matching types shuffle the deck: question 1 may live in paragraph F, question 2 in paragraph B, question 3 back in F again. A candidate who reads the passage top-to-bottom hunting for question 1 will re-read the same paragraphs over and over — once per question — and the time budget dies quietly.
The rule: for out-of-order types, you make one organised pass over the passage and many cheap passes over the question list — never the reverse.
Two further rule changes to absorb before anything else, because they are printed in the instructions and routinely ignored:
- "NB You may use any letter more than once." When this note appears, a paragraph or a person can be the answer twice. "I've already used D" is not an elimination.
- Not every paragraph is used. Matching information does not guarantee each paragraph contains an answer. Some exist purely as search noise.
Matching names and features: index first
When questions attach statements to people (researchers, historians, named organisations), the anchors are gifts — names are priority-1 keywords in the location hierarchy. Capitalised, un-paraphrasable, visually loud. Use them like this:
- Index pass (60–90 seconds). Scan the whole passage once and circle every occurrence of every listed name. Most names appear in one cluster; a few appear twice — mark both, because second mentions ("Haynes later revised this view…") are favourite answer sites.
- Work name by name, not statement by statement. Go to a name's circled territory, read the two or three sentences it owns, and then run down the statement list asking "which of these did this person say?" Reading one location against six statements is cheap; reading six locations for one statement is expensive.
- Match by meaning. The statements are heavy paraphrases of the passage's reported claims. "X doubted the reliability of early data" will appear in the passage as something like questioned whether the original measurements could be trusted. Same claim-attribution care as in Yes / No / Not Given: make sure the view belongs to that person, not to a critic quoting them.
Matching information: anchors, then triage
"Which paragraph contains a reference to…" has no names to index, so the anchor comes from the statement itself. Each statement usually contains one concrete, scannable element — a number, a technical noun, a date, a comparison. Extract it before you search:
| Statement says… | Scan for… |
|---|---|
| a reference to the cost of the project | currency symbols, numbers, million, funding, budget |
| an example of the technique used outside medicine | scan for domain nouns: engineering, agriculture, sport |
| a comparison between two time periods | dates, earlier, previously, by contrast, than |
| mention of an unexpected finding | surprisingly, contrary to, unexpectedly — the test loves attitude adverbs |
Then triage the statement list itself:
- First pass — hard anchors only. Answer every statement with a number, name, or rare noun first. These fall in under a minute each and shrink the search space.
- Second pass — abstract statements. "A reason why the theory was abandoned" has no scannable surface; it needs paragraph-purpose knowledge. If you did matching headings first, or made your structural skim, you already know which paragraph holds reasons and which holds examples — go there directly.
- Anything still open after two passes gets the 90-second circuit-breaker treatment from the time budget: best guess from your structural map, mark it, move on.
The detail-level trap
Matching information statements point at details, not main ideas — the mirror image of headings. "A mention of early failures" can be answered by a single subordinate clause buried mid-paragraph in a paragraph whose main idea is success. Two consequences:
- Do not reject a paragraph because its topic seems wrong. The reference you need may be one clause long.
- Do not choose a paragraph because its topic seems right. "The paragraph about costs" may never actually state a cost figure — the statement needs the specific reference, not the general subject.
Confirm the way you always confirm: put your finger on the words in the paragraph that are the reference. No finger, no answer.
Order of operations on a full passage
When a passage carries a matching set plus ordered questions (completion, TFNG), do the ordered types first if you already know the passage, or do the matching names index pass first — it doubles as a structural skim. Whichever you choose, never alternate between an ordered set and a shuffled set question-by-question; finish one set, then start the other. Context-switching between ordering regimes is how answers end up in wrong boxes.
Your drill
Twenty minutes, two passes.
- Open Reading 2026-05 Test 4 and find a matching set (information or names). Before answering anything, do the index pass: circle every name, or extract and write down one anchor per statement.
- Answer hard-anchor statements first, abstract ones second. Time the whole set — note how many total passes over the passage you made. Target: one organised pass plus short returns, not one pass per question.
- Check. For each miss, diagnose: wrong anchor extracted, second mention of a name missed, or detail-level trap (right topic, wrong paragraph)?
- Repeat with Reading 2026-02 Test 2, enforcing the triage order strictly.
When the index-first habit is in place, the "hard" shuffled types become bookkeeping — and bookkeeping is trainable.