Exam Map, Scoring and Attempt Strategy
This chapter covers exam map, scoring and attempt strategy. Starts by putting the paper in context: the UPSC-conducted NDA/NA written test has two papers — Mathematics and the General Ability Test (GAT).
KarmSakha preparation topics covered
- Understand and apply: Paper Structure: Notification-Fixed vs Observed
- Understand and apply: Scoring Arithmetic: Expected Score Formula
- Understand and apply: Break-Even Accuracy and Operating Target
- Understand and apply: Three-Pass Attempt Strategy and Time Budgets
- Understand and apply: Error-Budget Table and Pre-Exam Checklist
Paper Structure: Notification-Fixed vs Observed
The UPSC notification that announces the NDA/NA examination contains a formal 'Scheme of Examination' table. That table is the controlling document for the written test: it fixes the Mathematics paper at 300 maximum marks with 2½ hours, the General Ability Test (GAT) at 600 maximum marks with 2½ hours, and therefore a total of 900 marks for the written papers. It also fixes the question format as objective type with four alternatives and fixes the penalty for a wrong answer as one-third of the marks assigned to that question. A question left blank is not penalised; the candidate simply receives zero for it.
Notice what the scheme table does not contain. It does not give the number of questions in Mathematics or GAT, and it does not give a split of GAT into English and General Knowledge. Those numbers come from the question papers UPSC actually publishes each year. In recent years the published Mathematics paper has had about 120 questions, each carrying 2.5 marks, and the published GAT paper has had about 150 questions, each carrying 4 marks. These are observed regularities, not notification-fixed guarantees.
Within the observed GAT pattern, Part A-English has about 50 questions and Part B-General Knowledge has about 100 questions, making the familiar 50/100 split. The total marks of 600 is achieved because 150 questions × 4 marks = 600. There is no similar split within Mathematics; all questions are treated as one paper. Even the 50/100 split is a reading of the latest published paper, not a clause in the scheme. Treat it as a strong but revisable assumption.
This distinction is not pedantic. Coaching material routinely states 'Mathematics: 120 questions' and 'GAT: 150 questions' as if they were fixed by the notification. If UPSC changed the number of questions but preserved the total marks and duration, a candidate who had memorised the old question count would be confused on the day. The safer mental model is: the notification controls marks, time, question type and penalty; the published paper controls the question count you prepare against.
The negative-marking rule is applied per question. Since marks per question differ between the two papers, the absolute deduction differs: a wrong answer in Mathematics loses 2.5/3 = 5/6 of a mark, while a wrong answer in GAT loses 4/3 marks. The fraction, one-third, is the same; the absolute deduction is not. This is why any strategy that treats 'one wrong answer costs one mark' is wrong for both papers.
There is a further practical implication: because every question has four alternatives and exactly one correct answer, the answer sheet's options are not equally reliable once you have partial knowledge. The examination does not promise any uniform answer distribution. Do not rely on 'option C is the most frequent' patterns; rely on content knowledge and elimination.
Before you build any other part of your plan, take the current official notice and the latest GAT and Mathematics papers, put them side by side, and reconcile the figures above. If the notice still says one-third negative marking and the paper still shows 120 Mathematics and 150 GAT questions, then the strategy in this chapter can be used as written. If either has changed, the raw numbers change, even though the logical method of reconciling fixed and observed information remains the same.
| Feature | Notification-fixed | Observed on published papers |
|---|---|---|
| Mathematics paper | 300 marks, 2.5 hours | Usually 120 questions × 2.5 marks |
| GAT paper | 600 marks, 2.5 hours | Usually 150 questions × 4 marks |
| Total written marks | 900 marks | — |
| Question type | Objective type, four alternatives | — |
| Negative marking | One-third of allotted marks per wrong answer | — |
| GAT subject split | Not specified in scheme table | About 50 English + 100 General Knowledge |
Scoring Arithmetic: Expected Score Formula
The raw score on a paper can be written as E = m(C − W/3), where m is the marks allotted to each question, C is the number of correct answers, and W is the number of wrong answers. For the Mathematics paper m = 2.5; for GAT m = 4. Because every question in a paper carries the same value of m, the formula is valid for the whole paper. This expression is simply the scoring rule translated into algebra.
The term W/3 appears because each wrong answer deducts one-third of m, not a fixed 0.33 mark. Algebraically, a wrong answer contributes −m/3, so W wrong answers contribute −Wm/3 = −m(W/3). Correct answers contribute +Cm. Adding the two gives m(C − W/3). Blanks do not appear in the formula because they contribute zero; it would be a mistake to put blanks in the W column.
A concrete instance makes the arithmetic clearer. Suppose in GAT a candidate attempts 140 questions, gets 110 correct and 30 wrong, and leaves 10 blank. Then the score is 4(110 − 30/3) = 4(110 − 10) = 400. If the same candidate had left those 30 questions blank, the score would have been 4 × 110 = 440. The 30 wrong answers cost 40 marks: 30 × (4/3) = 40. This is the real price of careless guessing.
There is an important distinction between raw marks and percentage accuracy. Many students compute accuracy as C/(C+W) and assume the score is close to that percentage. That is false because of the W/3 penalty. If p is accuracy among attempted questions, the net mark per attempted question in units of m is p − (1−p)/3. For example, at p = 0.7 the net per attempted GAT question is 4 × 0.6 = 2.4 marks, not 2.8 marks. The penalty eats into every correct answer's surplus.
Now the 20% confusion. A claim that 'guessing is profitable above 20% accuracy' belongs to examinations where the penalty is one-fourth of a question's marks, not one-third. Under a one-fourth penalty, the break-even value is one-fifth, or 20%. NDA uses one-third, and the corresponding value is one-fourth, or 25%. The difference is not trivial: if you plan around 20% on 100 guesses at m = 4, your expected net score is about −26.7 marks, whereas at 25% it is exactly zero.
Someone might ask whether omitted answers hurt because they lower the attempted count. No, the scoring formula has no term for blanks. However, the opportunity cost of a blank is real: if you leave a question blank that you could probably solve, you lose the chance to add m marks and also lose the chance to earn the surplus that a confident attempt would give. That is a strategic cost, not a scoring penalty. The formula tells you blank is zero, but strategy must treat avoidable blanks as negative opportunities.
Always read the exact wording in the notification. The formula above is built on 'one-third of the marks assigned to a question'. If a future notice changes the fraction to one-fourth or changes the marking scheme to no negative marking, the formula changes. The method, translating the notification into algebra before using any shortcut, remains the same. Never import a formula from another examination without checking its penalty fraction.
Break-Even Accuracy and Operating Target
The break-even accuracy is the probability of being correct at which attempting a question gains nothing in expectation. Let p be the probability that your answer to a particular question is correct. If you get it right, you add m marks; if you get it wrong, you lose m/3. The expected gain from attempting that question is therefore p × m + (1−p) × (−m/3).
Set this expected gain to zero. Then pm − (1−p)m/3 = 0. Divide by m and rearrange: p − (1−p)/3 = 0, so 3p = 1−p, so 4p = 1, so p = 1/4. Therefore any answer you give should have at least a 25% chance of being correct. For a completely blind guess among four equally attractive options, p is exactly 1/4, so the expected gain is zero.
This p is not an overall score target. A blind guess is expectation-neutral only if all four options are equally likely and you have no partial knowledge. In a real NDA question, the options differ, the distractors have designed flaws, and your preparation makes some options impossible. The moment you can discard one option, your p rises to at least 1/3, and the expected gain becomes m/9. If you can discard two, p becomes at least 1/2 and the expected gain is m/3. This is why elimination is the only way to turn guessing from a risk into a small but real advantage.
The 20% figure deserves a direct warning. Many guides repeat 'break-even is 20%' from examinations that deduct one-fourth of the allotted marks. For such exams, break-even is p = 0.25/1.25 = 0.2. For NDA, with a one-third penalty, break-even is p = 0.25. A candidate who thinks a 20% correct-guess rate is acceptable will be operating below break-even on every guessed question. On 100 GAT-mark guesses, that error costs about 26.7 marks in expectation, while break-even would cost nothing.
Now the operating target. Break-even is only a floor; it is not a comfortable place to plan from because at 25% expected gain is zero. On a high-stakes examination, you should be far above the floor, not near it. A realistic planning target is 65–70% accuracy among attempted questions. Put the number in the formula: if you attempt questions with m = 4 and your overall accuracy is 70%, the expected net per attempted question is 4(0.7 − 0.3/3) = 4 × 0.6 = 2.4 marks. At 65%, it is about 2.13 marks. At 25%, it is zero.
A worked contrast makes the value of accuracy concrete. Suppose you attempt 100 questions at m = 4. At 70% accuracy, your expected net score is 100 × 4 × 0.6 = 240 marks. At 50% accuracy, it is 100 × 4 × (0.5 − 0.5/3) = 133.33 marks. At 25%, it is zero. The jump from 50% to 70% accuracy is worth more than 100 marks on just 100 questions. This is why the three-pass strategy directs you to attempt known questions first and reserve guesses for questions where elimination has raised p well above one-quarter.
Finally, break-even is a floor, not a goal. Suppose you guess on 30 questions with exactly 25% accuracy. In expectation you get nothing from those 30 questions, but you have spent time, introduced variance, and perhaps exposed yourself to pattern-based traps. If you can eliminate one option, p rises to one-third and the same 30 questions become profitable in expectation. The planning target of 65–70% should therefore be checked in mock tests: if your attempted-question accuracy is below 65%, revise rather than expanding the number of guesses. At break-even you are not scoring; you are merely not losing.
Three-Pass Attempt Strategy and Time Budgets
Three-pass strategy is not a suggestion about the order in which to read questions; it is a filter that matches effort to marks. Because every correct answer earns the same marks, a question that takes three minutes is not worth three times as much as one that takes thirty seconds. The first pass is a scanning pass: answer only questions where the answer is clear immediately, and mark the rest without letting your eyes linger. The second pass is the working pass; here you do the straight calculation or derivation that you identified as needing time. The third pass is a decision pass, reserved for questions you have narrowed down by elimination. This order makes the scoring formula work for you: it fills your answer sheet with cheap correct marks before you spend time on expensive ones.
In the first pass — the known-question pass — act as if the only instruction is to find the question you could answer in your sleep. A formula question from previous years, a direct GK fact, or a vocabulary item belongs here. For mathematics, keep your mental threshold high: if the solution is not immediate, do not begin at once. Put a small dot or a question number on a rough sheet and move on. For GAT, the same rule applies; a grammar correction or a factual question you know should be answered as you meet it. This pass should be quick and almost mechanical. Its purpose is to bank the marks that cost almost no time so that every minute you spend later is buying marks you cannot get more cheaply elsewhere.
Pass two — the workable-question pass — is where the paper is actually won or lost. Take the list of marked questions and solve the ones you recognise as workable: a calculation that needs two steps, a reading-comprehension question with a clear supporting line, a general-science question you can reason out. Work them directly on paper. Do not reread the whole stem twice before starting; solve. If a question does not start yielding within the first working attempt, leave it. The common mistake is to treat pass two as another reading pass: you revisit a question, reread it, and then put it aside because no promising method comes. That is not an attempt; that is a time debt.
