Wrong answers are not random — they repeat in patterns, and the pattern is personal. One candidate bleeds marks to Not Given traps, another to word-limit violations, a third to questions never reached. This lesson installs the five-column error log that finds your pattern within a week of practice, and converts it into a fix you can rehearse.
Why scoring a test isn't reviewing it
Most candidates "review" like this: check answers, sigh at the score, glance at the wrong ones, feel briefly enlightened, do another test. The score goes nowhere, because the information in the mistakes was never extracted. A wrong answer is a data point about a repeatable failure; ten wrong answers reviewed properly are a diagnosis. Reviewed casually, they are just disappointment.
The rule: a practice test is not finished when it is scored. It is finished when every wrong answer has a row in the log.
The five columns
Keep it anywhere — notebook, spreadsheet, phone note. One row per wrong answer (and per lucky guess: a mark you got without knowing why is a miss wearing a disguise).
| Column | What goes in it | Example |
|---|---|---|
| 1. Where | Test and question number | 2026-03 T5, Q27 |
| 2. Type | Question type | TFNG / headings / MCQ / completion / matching / YNNG |
| 3. Mine vs right | Your answer vs the correct one | NG vs FALSE |
| 4. Cause | One of the five cause codes below | TRAP |
| 5. Fix | The specific rule or drill that would have prevented it — written as an instruction to yourself | "Point at the contradicting words before writing NG" |
Column 4 is the engine. Use a closed set of five codes — free-text causes ("wasn't careful") produce nothing trainable:
- LOC — located the wrong sentence, or never found it. Keyword chosen badly, scanned past it, or gave up early.
- PARA — found the right sentence but missed the paraphrase link; couldn't see that the question phrase and passage phrase matched.
- TRAP — saw the evidence, judged it wrong: an extreme word ignored, a half-truth option verified only halfway, NG/F boundary crossed.
- TIME — never reached the question, or answered in panic during the final scramble.
- MECH — mechanical: word limit broken, spelling miscopied, answer form changed, wrong box on the answer sheet.
If a row seems to need two codes, take the earlier failure in the chain — a trap you fell into because you were rushing is TIME, not TRAP. The chain always runs LOC → PARA → TRAP, inside a budget set by TIME, executed by MECH.
Reading the log after one week
After three or four logged tests (30+ rows including lucky guesses), stop and count. Two tallies:
- By cause. One code almost always owns 40%+ of your rows. That code is your band ceiling, and each has a known prescription: LOC → redo location discipline and Drill B of speed drills; PARA → the daily paraphrase-pair drill; TRAP → the trap tables in TFNG and multiple choice, with fault-naming made mandatory; TIME → checkpoints and the circuit breaker from the time budget; MECH → the two mechanical checks from completion, spoken aloud until boring.
- By type. If one question type contributes disproportionate rows within your dominant cause, your next three practice sessions target that type specifically — do that type's sets from two or three different tests back-to-back rather than sitting whole fresh tests.
Then rehearse the fix, don't just know it. Column 5 exists to be executed: take the exact questions you missed, cover your old answers, and re-answer them applying the written instruction. Getting a question right the second time, for the stated reason, is the moment the log converts into marks.
Keeping it alive
- Log within an hour of scoring. Next day, you no longer remember why you chose NG, and column 4 becomes fiction.
- Two minutes per row, maximum. The log is a tally, not a diary. Long remorseful entries kill the habit by day four.
- Re-tally weekly. Your dominant code will change — TIME usually falls first, then LOC, and most candidates discover their endgame is a stubborn PARA or TRAP profile. Each shift redirects the next week's training.
- Fold it into your plan. Our study plans schedule test-days and review-days separately; the log is what review-days are for. A plan that only schedules tests is a plan to repeat the same mistakes at higher volume.
Your drill
Thirty minutes: one test, one seeded log.
- Sit Reading 2026-03 Test 5 under full exam conditions — 60 minutes, checkpoints, circuit breaker.
- Score it, then build the log immediately: one row per wrong answer and per lucky guess, all five columns, cause codes from the closed set only.
- Tally column 4. Even one test usually shows a lean. Write one sentence at the bottom: "Provisional diagnosis: ____. This week I train it by ____."
- Later this week, log Reading 2026-01 Test 4 the same way and re-tally across both tests. Then open the study plans and slot your prescribed drill into the review days.
The log costs ten minutes per test. It is the difference between practising — doing tests and hoping — and training: doing tests, finding the leak, and sealing it before the next one.