# Thesis Defense Prep > Rehearses the viva. Reads the thesis the way a committee reads it, marks the three weakest points in every chapter, builds a question bank of forty to sixty questions in the words examiners actually use, and drills each answer into three parts under sixty seconds with the exhibit to point to, then sizes the defence talk and prepares the room, the concessions and the revision list. Enforces the rule that every answer leads with the direct answer or the concession rather than the defence, and that no answer is rehearsed without knowing where the evidence for it sits. Use this skill when someone mentions a defence, a viva, a proposal hearing, a qualifying exam or a committee meeting, asks what their committee will ask, asks how to answer a particular question, or wants to practise. Trigger also on vague requests such as "my viva is in three weeks", "I am terrified of one of my examiners", "what if they ask about the instrument", or "how long should my answers be". - Skill: `ingridleiria/thesis-defense-prep` (Agent Skill) - Install (CLI): `npx skillmds@latest add ingridleiria/thesis-defense-prep` - Raw SKILL.md: https://api.skillmd.com/api/skills/ingridleiria/thesis-defense-prep/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Research & Search - Author: ingridleiria (https://skillmd.com/u/ingridleiria) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/ingridleiria/thesis-defense-prep --- # Thesis Defense Prep A viva is not won on the day. It is won in the four weeks before it, by a candidate who has already met every question that will be asked and has decided in advance what to concede. What happens on the day is largely a test of whether that work was done. The failure this prevents is specific and it is common in candidates whose theses are perfectly sound. An examiner asks about the weakest link in chapter three, which the candidate knows about and has never had to say out loud. The candidate improvises. The improvised answer is three minutes long, defends a position that is not defensible, and invites a follow-up. The follow-up gets a worse answer, because now the candidate is rattled and is defending the first answer rather than the thesis. Twenty minutes of the viva disappear into a hole that a prepared two-sentence concession would have closed instantly, and the committee's impression shifts from a candidate who understands their work to one who does not know its limits. The other common failure is the opposite: a candidate so rehearsed that they answer the question they prepared rather than the one they were asked. Both are fixed by the same discipline, which is preparing the shape of an answer rather than a script. ## When to use this, and when not to Use it in the weeks before any committee examination: a final viva or defence, a proposal hearing, a qualifying or comprehensive exam, an upgrade or transfer panel, an annual review with a panel, or a mock defence. Use it after a chapter review has produced findings that will not be fixed before submission, because an unfixed weakness that the candidate has prepared an honest answer for is survivable and one they meet cold is not. Use it also for a job-market flyout or a high-stakes seminar where the question period is the assessment, taking the question bank and answer discipline from here and the deck from `conference-presentation-deck`. Do not use it to find out what is wrong with the thesis. That is `thesis-chapter-review`, which reads as an examiner and ranks findings by consequence, and it should run first: this skill takes its findings as an input and turns them into questions and answers. Running this without it produces a question bank built on guesses about where the weaknesses are. Do not use it to decide whether to submit, or what to fix in the remaining time; that is `thesis-advisor`. Do not use it to build the defence talk from scratch, which is `conference-presentation-deck`; this skill sizes it, indexes the appendix to the question bank, and rehearses it. Do not use it to answer referees on a journal submission, which is `response-to-reviewers`, although the ledger discipline for post-defence required revisions is borrowed from there. ## What you need before starting **The submitted thesis, exactly as the committee has it.** Not the working version. Page numbers matter, because an answer that cannot be located in the document the examiner is holding is not an answer. Missing: work from the latest version and mark that page references must be re-checked against the submitted file. **The examination format at this institution.** Public or closed, presentation or no presentation, its length, who chairs, whether the chair participates, how long the examination runs, whether the candidate leaves the room for the deliberation, what corrections categories exist and what each requires. Missing: ask the candidate to find the regulations, which takes twenty minutes and changes the preparation substantially. A viva with no presentation is prepared very differently from one that opens with a forty-minute talk. **The committee, by name, with what each of them works on.** A methodologist, a field specialist and an external examiner from an adjacent discipline ask three different sets of questions about the same chapter. Missing: prepare for all three orientations and say which questions come from which, which is more useful than a single averaged bank. **The findings from a chapter review.** Missing: run `thesis-chapter-review` first. It is the cheapest way to locate the weak points, and a question bank built without it will miss the questions the text is avoiding, which are the ones that do the damage. **The candidate's own list of what they are afraid of.** Missing: ask, but ask after the chapter map has been built independently, so the two can be compared. Where they agree, the fear is well founded and the answer must be excellent. Where the map found something the candidate had not worried about, that is the highest-value item in the whole exercise. **Time before the examination, and the candidate's availability in it.** Missing: assume the preparation needs about twenty to thirty hours spread over three weeks, with the last week reserved for rehearsal rather than new work. ## The method 1. **Build the chapter map.** For each chapter, on one page: the question, the design, the main finding with its number, the identification or credibility evidence, the stated limitations, the exhibits, and the link to the thesis-level question. This is the document the candidate revises from in the final week and it is the only thing they should be reading the night before. 2. **Mark the three weakest points per chapter.** The claim least supported by its evidence, the assumption least defended, and the exhibit most likely to draw a question. Draw these from `thesis-chapter-review` where it has run, and from a fresh adversarial read where it has not. These are where the viva will be fought, and the preparation budget should be allocated to them rather than spread evenly. 3. **Profile the committee.** For each member, what they work on, what methods they use, what they have published that touches this thesis, and, where known, what they habitually ask. An examiner who has written on measurement error will ask about measurement error in whichever chapter offers a target. This step converts a generic question bank into a specific one and it takes an hour. 4. **Generate the question bank**, forty to sixty questions in five groups, described below, each written in the words an examiner would actually use rather than in analytical shorthand. Mark each with which committee member is most likely to ask it. 5. **Write every answer in the three-part shape** set out below, in speaking sentences, and mark the exhibit and page for each. Do not write scripts; write the shape, the concession, the evidence, and the boundary, so the candidate can assemble the sentence in the room rather than reciting a paragraph. 6. **Decide the concessions in advance, in writing.** Every thesis has three or four points where the honest answer is that the evidence is weaker than one would like. Write those concessions out, as sentences, and rehearse them until they can be said without discomfort. A concession delivered calmly is the strongest move available in a viva; the same concession extracted after two minutes of resistance is the weakest. 7. **Rehearse aloud, timed, with someone asking the questions out of order.** Silent rehearsal does not prepare speech. Out of order matters because a candidate who has rehearsed in bank order will be thrown by a question that arrives from an unexpected direction, which is how real examinations proceed. Two sessions of an hour each, at least three days apart, beats one long session. 8. **Build or size the defence talk** where the format requires one, and index every appendix slide to a question in the bank so any answer can be shown as well as spoken. 9. **Prepare the room and the day**: what to bring, what the opening sentence is, how the required revisions will be recorded. 10. **Stop new work one week out.** The last week is rehearsal and rest. A candidate who is still running specifications three days before a viva arrives tired and with an answer bank they have not practised, which is a worse trade than any specification is worth. ## The question bank **Thesis-level, ten questions.** Why these chapters together. What is the single contribution, stated in one sentence. What changes in the field if every finding is correct. What unifies the identification strategies, or if nothing does, why that is acceptable. What is the one thing the thesis cannot answer, and why. Which chapter is weakest and why. What would you do differently if you started again. These are asked at the start and at the end, and the answer to the contribution question is the sentence the committee will carry into its deliberation. **Per chapter, eight to twelve each.** Built from the design's attack list in `identification-defense`, from the field-expert concerns a referee would raise, from the data section (how the sample was constructed, what was dropped and why, attrition, measurement), from magnitude interpretation, and from the link between the theoretical mechanism and the estimated result. **Method fundamentals, ten.** The assumptions of every estimator used, stated from first principles without notation. What the estimand actually is and for which population. Why this estimator rather than the two obvious alternatives. How inference was done and why that choice. What the standard errors are clustered on and why. These questions distinguish a candidate who used a method from one who understands it, and examiners use them deliberately for that purpose. **Hostile and broad, five to ten.** Why should anyone care. Is this not already known from the closest paper. What if the main assumption is wrong. Does this generalise beyond this country, this period, this population. Is this a thesis or three loosely related papers. These are asked to see how the candidate behaves under pressure rather than to elicit information, and the correct response is a calm, short, direct answer. **Procedural, three or four.** What revisions do you anticipate. How long would they take. What would you publish first. Which chapter is closest to submission. ## The three-part answer **Part one: the direct answer or the concession, first.** "Yes, that is the main limitation of chapter three." "No, the effect is not identified for that population, and the thesis does not claim it is." "The bias runs downward, so the estimate is conservative." Leading with this is counterintuitive under pressure and it is the single highest-value habit in the whole preparation. Committees read a deflection instantly and respond by pressing harder; they read a concession as evidence that the candidate understands their own work and usually move on. **Part two: the evidence.** The exhibit, the page, the number. "Table 4.6 on page 118 reports the placebo, and the coefficient is 0.004 with a standard error of 0.011." Being able to point at something converts an assertion into a demonstration, and it also gives the candidate something to do with their hands and eyes for three seconds, which is worth more than it sounds. **Part three: the boundary.** What the thesis claims on this point and what it does not. "So the thesis claims the effect for first-time applicants, and it does not claim anything about renewals." Under sixty seconds, all three parts. A long answer signals uncertainty regardless of content, and it consumes examination time that the candidate could be using to demonstrate range on other questions. The discipline is enforced by timing in rehearsal, out loud, with a clock visible. One rule that overrides the shape: answer the question that was asked. If the question is not the one rehearsed, say the direct answer to the actual question first, even if it is "I do not know", and only then reach for prepared material. ## The defence talk Where the format requires a presentation, build it with `conference-presentation-deck` and size it to the programme's rules, commonly twenty to forty minutes. Three things differ from a conference talk. The contribution is stated in the first ninety seconds, in the sentence the candidate wants the committee to use in its deliberation. The limitations are stated by the candidate, in the talk, before any examiner raises them, which removes the possibility of being caught concealing them and consistently improves the tone of the questions that follow. And the appendix is indexed to the question bank, so every prepared answer has a slide, and the candidate can say "that is on slide 34" rather than describing an exhibit from memory. Rehearse the first ninety seconds until it is automatic. Nerves do their worst damage in the opening and an automatic opening carries the candidate past it. ## The day Bring a printed copy of the thesis with the chapter map inside the front cover, tabbed at the exhibits most likely to be needed. Bring water. Bring a pen and a page for the required revisions. Open, where the format allows any opening at all, by stating the contribution in two sentences. Committees form a frame in the first minute and the candidate should supply it. During questions: listen to the whole question without beginning to answer it. Restate it briefly if it was long or multi-part, which buys three seconds and prevents answering only the last clause. Answer directly. Stop. Silence after a complete answer is the examiner thinking, not an invitation to continue, and candidates who fill it talk themselves into trouble. Where an answer is genuinely not known, say so, say how it could be found, and stop. "I do not know, and the way to check would be X" is a good answer. Improvising a result is the only unrecoverable error available in a viva. Where an examiner is wrong on a fact, correct it once, politely, with the evidence, and move on. Where an examiner disagrees on judgement, acknowledge the position, restate the thesis's position once, and stop. Arguing past that point cannot win anything and can lose something. Write down every required revision as it is stated, in the examiner's words. At the end, read the list back and confirm it. This is the single most useful administrative act of the day and almost nobody does it: candidates leave with an emotional impression of what was required and reconstruct the list from memory a week later, which is how corrections are missed. ## After the defence Convert the required revisions into a ledger with the discipline of `response-to-reviewers`: every requirement numbered, quoted, with its disposition, the change made, and the location. Where a requirement is unclear, ask the chair or the internal examiner in writing rather than guessing, and do it within the first week while everyone still remembers. Check the corrections category and its deadline, and diarise the deadline. Where the corrections are substantial, run `thesis-advisor` on the plan, because the same calendar arithmetic applies. ## Worked example **Situation.** A candidate, Hyun-woo Park, had a viva in three weeks on a three-chapter thesis about migration and local labour markets. The format was a twenty-minute presentation followed by roughly two hours of questions, with an internal examiner who was a labour economist, an external examiner known for work on instrument validity, and a non-participating chair. Chapter two used a shift-share instrument. A chapter review two months earlier had flagged the instrument's exclusion restriction as the thesis's weakest point, and the candidate had not been able to fix it before submission. **Task.** Prepare in three weeks, with the instrument problem unfixable, and arrive with an answer that does not collapse. **Action.** The chapter map took a day. The three weakest points per chapter were drawn from the earlier review. The committee profile was built in ninety minutes and it changed the whole allocation: the external examiner had published two papers on the validity of exactly this class of instrument, one of which was cited in the thesis. The reasonable expectation was that a substantial part of the examination would be spent on chapter two's instrument, and the preparation was weighted accordingly, with roughly half the total effort on one chapter. The question bank came to 54 questions. Eleven were on the instrument. **The wrong turn.** The candidate's first prepared answer on the exclusion restriction ran two minutes and forty seconds and defended the instrument. It cited three papers that used similar instruments, argued that the setting made the concern less likely, and presented an overidentification test. Rehearsed aloud in front of a colleague playing the external examiner, it produced exactly the outcome it was designed to avoid: the colleague asked a sharper follow-up, the candidate defended the defence, and four minutes had gone with the position weaker than at the start. The answer was rebuilt into the three-part shape and the first part was changed from a defence to a concession. The new opening: "The exclusion restriction cannot be tested directly, and the most serious threat is that the same national demand shocks affect local wages through channels other than migration. I cannot rule that out." Then the evidence: the two robustness exercises in Table 2.7 and Appendix Table B4 that bound how large such a channel would have to be to overturn the result, with the actual numbers. Then the boundary: what the chapter claims under the assumption and what it does not claim. Fifty-two seconds. Rehearsed eleven times over two weeks, in different orders, until it could be delivered without hedging. Three other concessions were written and rehearsed the same way, including one on chapter three's external validity that the candidate had not previously regarded as a weakness and which the chapter map had surfaced. The talk was rebuilt to state the contribution in the first ninety seconds and to state the three limitations explicitly at minute sixteen, before any examiner raised them. Nine appendix slides were indexed to specific questions in the bank. **Result.** The instrument came up in the fourth question and occupied about twenty-five minutes of the examination, which was expected. The prepared concession was delivered as the first sentence of the answer. The external examiner's recorded response was that this was the right way to think about it, and the discussion moved to how large the violation would have to be, which was the ground the candidate had prepared with the bounding exercise and appendix slide 31. Of the 54 questions in the bank, 19 were asked in some form. Six questions were not in the bank at all, of which one was substantial: whether the results were sensitive to a national policy change in the middle of the panel. The honest answer was that it had not been checked. The candidate said so, said how it would be checked, and it became a required correction rather than a failure. Outcome: pass with minor corrections, four of them, all recorded in the candidate's own handwriting during the examination and read back at the end. The item that mattered most was the two-minute-forty defence being abandoned. On the colleague's account of the rehearsal, that answer would have cost the candidate the tone of the entire examination. ### A second scenario, where it goes differently A proposal defence at the end of the second year, where no results exist. Almost everything reorients. There are no findings to defend, so the question bank collapses on the thesis-level and method-fundamentals groups and expands in a direction the final viva does not have: feasibility. Can this be done in the time. Is the data access secured or hoped for. What happens if the main design fails. What is the fallback. Those questions carry most of the weight, and a candidate who prepares only the theoretical contribution will be caught by them. The concession discipline changes shape. At a final viva, a concession is about the evidence. At a proposal, it is about the plan, and the strongest version is a stated contingency: "If the parallel trends assumption does not hold in the pre-period, the fallback is a design restricted to the subset of regions where it does, which costs me about a third of the sample and I have checked that it is still powered." The talk changes too: no results means the identification section and the data access section carry the presentation, and the single most important slide is the timeline, which the panel will read as a proxy for whether the candidate understands what they are proposing. One thing does not change. The panel is assessing whether the candidate knows the limits of their own plan, which is the same assessment as at a final viva, made on a different object. ## Output **The chapter map**, one page per chapter: | Chapter | Question | Design | Main finding, with number | Identification evidence | Stated limitations | Weakest three points | **The committee profile** | Member | Field and methods | Relevant own work | Questions likely to come from them | **The question bank** | # | Question, in an examiner's words | Group | Likely asker | Answer, three parts | Exhibit and page | Appendix slide | **The concession list**, three or four sentences, written out verbatim and marked as rehearsed or not. **The talk**, where required, with the appendix indexed to the bank. **The day sheet**: opening two sentences, what to bring, the rules for listening, restating, stopping, and the blank page for required revisions. **After the defence**, the required revisions ledger: | # | Requirement, in the examiner's words | Disposition | Change made | Location | Due | ## Failure modes **Rehearsing silently.** Recognise it when the candidate has read their answers but not said them. Speech under pressure is a different skill from composition and it is the one being tested. **Scripted answers.** Recognise it when an answer is a memorised paragraph. The candidate will deliver it to the wrong question. Rehearse the three-part shape, not the words. **Defending before conceding.** Recognise it when an answer's first sentence contains "however" or "but". Move the concession first and see how much shorter the answer becomes. **Long answers.** Recognise it by timing. Over sixty seconds, the answer is doing something other than answering. **Preparing evenly.** Recognise it when every chapter got equal effort. The committee profile and the weakest-points map say where the time goes. **Skipping the committee profile.** Recognise it when the question bank could have been written for any thesis in the field. An hour on the examiners' own work is the highest-yield hour available. **New analysis in the final week.** Recognise it when the candidate is running code three days out. Stop; rehearse instead. **No record of the required revisions.** Recognise it when the candidate leaves the room without a written list. Write them down as they are said and read them back. **Improvising a result.** Recognise it in rehearsal, when an answer contains a number the candidate cannot locate. Replace with "I do not know, and here is how it could be checked". **Answering the question you prepared.** Recognise it in rehearsal when the questioner deviates and the answer does not. Practise with questions asked out of order and in unexpected phrasing. ## Edge cases **A known weakness that cannot be fixed before the examination.** This is the normal case, not an exception. Prepare the concession, prepare the bound on how much it could matter, and prepare what would be required to resolve it. Examiners assess whether the candidate understands the limitation, not whether the thesis is flawless. **A hostile examiner, or one with a competing paper.** Prepare the specific comparison as an exhibit, concede what is genuinely better about their approach, and state precisely what this thesis does that theirs does not, in under a minute. Do not argue past one exchange, and do not adopt their framing of the field. **Examiners disagree with each other during the examination.** Do not adjudicate. Answer the question that was asked, acknowledge that there are two views, and state which the thesis takes and why. Candidates who take a side in an examiners' disagreement lose regardless of which side is right. **An error is found during the viva.** Concede it immediately and precisely, say what it affects, and say what it does not affect. A conceded error is a correction; a defended error is a credibility problem that spreads to everything else. **The thesis is by publication and a chapter is coauthored.** Prepare the contribution statement for each coauthored chapter and be able to say specifically what the candidate did. This is asked routinely and vagueness here is damaging out of proportion to the question. **A closed viva with no presentation.** The talk disappears and the opening two sentences become more important, because there is no other opportunity to frame the thesis. Prepare a ninety-second answer to "tell us about your thesis", which will be the first question. **The examination is in the candidate's second language.** Rehearse aloud considerably more, keep answers shorter, and prepare the explicit phrase for asking an examiner to repeat or rephrase a question. Using it once early is fine and normalises it. **Very little time, a week or less.** Compress to the chapter map, the committee profile, the top twelve questions weighted to the weakest points, and the concessions, rehearsed aloud. Skip the full bank. The concessions are the part that must not be skipped. ## Quality bar - Every question in the bank is written in the words an examiner would use, not in analytical shorthand. - Every answer leads with the direct answer or the concession, and none exceeds sixty seconds when spoken. - Every answer names the exhibit and page that supports it. - The three or four concessions are written out verbatim and have been rehearsed aloud until they can be said calmly. - The question bank covers every weak point the chapter map identified, weighted toward the ones the committee profile predicts. - Rehearsal happened aloud, timed, with questions asked out of order by another person. - The talk, where required, states the contribution in the first ninety seconds and the limitations before any examiner raises them. - The candidate has a written plan for recording required revisions during the examination and reading them back. ## Adapting this to your context Built around a closed viva on a quantitative economics thesis, with a bank weighted towards identification. The three-part answer holds anywhere; the format and the bank are local. - **The format.** A closed examination by two examiners is one model. A United States committee defence opens with a public talk then closes the room; a Brazilian banca is public with fixed time per member; Nordic and Dutch defences give a formal opponent a prescribed role. Establish who asks, for how long, in what order. - **The method fundamentals bank.** Ten questions on estimators, clustering and estimands is the economics version. Substitute yours: how missing data was handled, the factor retention decision, what the intraclass correlation implies for the nesting, or in qualitative work how the coding frame was built and what would have counted as a disconfirming case. - **The concessions.** Here, identification weaknesses. Elsewhere, a single site, an underpowered sample, a self-report measure standing in for behaviour, or a coding frame applied by one person. Name yours and rehearse saying them calmly. - **The talk.** Twenty to forty minutes assumes a presentation is required. Where none is, preparation shifts entirely to the bank and the chapter map. - **What not to change.** Lead with the direct answer or the concession, know the page and exhibit behind every answer, and write each required revision down in the examiner's own words on the day. ## Related skills `thesis-chapter-review` runs first and supplies the weak points this skill turns into questions. `thesis-advisor` decides whether the thesis is ready to submit and what the post-viva correction plan can realistically contain. `identification-defense` supplies the per-design attack list and the bounding exercises that make a concession credible rather than merely honest. `conference-presentation-deck` builds the defence talk and the indexed appendix. `response-to-reviewers` supplies the ledger discipline for the required revisions after the examination. `analysis-audit` prevents the worst viva moment, which is an examiner finding a number that cannot be reproduced. `replication-package` is a deposit requirement at many institutions and belongs in the post-defence plan.