Thesis architecture & defense-prep KNOWLEDGE — BSc, MSc, PhD structure; classic 5-chapter + cumulative (article-based) variants; RQ↔chapter mapping with word_budget.py --rq-trace; Turkish thesis (YÖK) requirements (front-matter order, Özet↔Abstract equivalence, ethics declaration, acknowledgments, approval page); institute guideline ingestion; the concepts behind defense prep (viva questions, answer skeletons, presentation outline). Use when structuring or auditing a thesis, or preparing Özet/Abstract. To actually RUN guided defense preparation (jury question bank + presentation scaffold) use the /defense-prep command — this skill is its underlying knowledge, not the command.
Thesis architecture & defense-prep KNOWLEDGE — BSc, MSc, PhD structure; classic 5-chapter + cumulative (article-based) variants; RQ↔chapter mapping with word_budget.py --rq-trace; Turkish thesis (YÖK) requirements (front-matter order, Özet↔Abstract equivalence, ethics declaration, acknowledgments, approval page); institute guideline ingestion; the concepts behind defense prep (viva questions, answer skeletons, presentation outline). Use when structuring or auditing a thesis, or preparing Özet/Abstract. To actually RUN guided defense preparation (jury question bank + presentation scaffold) use the /defense-prep command — this skill is its underlying knowledge, not the command.
1. Thesis Architecture Types
1.1 Classic 5-Chapter Structure
Standard for all thesis levels (BSc, MSc, PhD). Each chapter has a defined purpose and typical word proportion:
Chapter
Purpose
Typical %
Content core
1. Introduction
Frame problem, state RQs, claim novelty
15%
background, gap, research questions, thesis outline
study design, participants/data, procedures, analysis strategy, validity measures
4. Results/Findings
Present outcomes, data analysis results
20%
organized per RQ (headings must mirror RQ statements), tables, figures, no interpretation
5. Discussion & Conclusion
Interpret, compare, claim contributions
20%
answer each RQ, situate vs literature, limitations, future work, practical implications
RQ-Mapping Rule (enforced by word_budget.py --rq-trace):
Every RQ stated in Ch.1 must have a corresponding Results section heading in Ch.4 and a claim in Ch.5. Execute:
python scripts/word_budget.py --rq-trace
Returns mapping: RQ1 → Section 4.1.1 heading → Discussion answer, or FLAG if missing.
Typical word counts (adjust per institution, field, degree):
BSc: 8,000–15,000 words
MSc: 25,000–50,000 words
PhD: 80,000–150,000 words
1.2 Cumulative (Article-Based) Thesis
Structure when your thesis comprises ≥2 peer-reviewed / submitted journal papers:
Ch.1: General Introduction (framing, overall RQ, synthesis of papers' motivation)
Ch.2: Literature Review (cross-cutting themes across the papers)
Paper 1 (standalone Methods + Results, verbatim or lightly adapted)
Paper 2 (standalone Methods + Results)
Ch.5: General Discussion (synthesize across papers, implications, limitations, future work)
Front-matter (same as classic) + binding narrative between papers. Each paper retains its own abstract; the dissertation gains an Özet + Abstract at the start. Common in STEM fields where students have multiple publications by completion.
2. Cross-Chapter Logic Chain
Integrity checks (audit before finalization)
Novelty cascade:
Ch.1 Literature Review identifies a gap.
Ch.3 Methodology describes your unique approach to close it.
Ch.4 Results present the outcome.
Ch.5 claims this outcome answers the gap.
RQ traceability:
Ch.1: "RQ1: Does method X outperform baseline Y under constraint Z?"
Ch.4: Heading "4.2.1 Performance Comparison: Method X vs Baseline Y" (no rewording).
Ch.5: "RQ1 was answered via [comparison table 5]: X achieved ±% better accuracy…"
Ch.5 explicitly names what you couldn't do and why (sample size, generalizability, confounds).
Consistency across languages (if bilingual):
Özet (Turkish) Ch.1–Ch.5 structure mirrors English Abstract word for word. See reference/yok-tr.md Section 2.
3. Chapter-by-Chapter Guidance
Chapter 1: Introduction (target: 12–18% of total)
Moves (in order):
Hook: Context — why the problem matters (broader field, practical need, or theoretical gap).
Problem statement: Specific gap or unresolved question.
Literature positioning: "Prior work has shown X and Y, but no one has studied [your angle]."
Explicit RQs: Numbered (RQ1, RQ2, etc.). Testable, not rhetorical.
Thesis outline: One sentence per chapter ("Ch.2 reviews…").
Discipline: Active voice, present tense for established fact, future for your work ("This thesis explores…").
Chapter 2: Literature Review (18–22%)
Synthesis, not summary. Group sources by theme, NOT by paper.
Structure option A (thematic):
2.1 Foundational Concept (3–4 sources in conversation)
2.2 Prior Approaches (contrast 2–3 methods)
2.3 Limitations and Gaps (what the literature misses)
Structure option B (chronological with themes):
2.1 Early Work (1990–2005)
2.1.1 Methodological approaches in that era
2.1.2 Key findings and limitations
2.2 Recent Developments (2006–present)
2.2.1 Emerging methods
2.2.2 Unresolved questions
Anti-pattern: "Author A says X. Author B says Y. Author C says Z." (annotated bibliography) → FWD: Replace with synthesis. One paragraph should reference ≥2 sources in substantive conversation.
Paragraph formula:
Topic sentence + Evidence (cite EV-cards or [@surname2024]) + Interpretation + Gap statement + Bridge to next paragraph.
Chapter 3: Methodology (22–28%)
Replicability standard: Another researcher must be able to reproduce your study using this chapter alone.
Environment: Python version, libraries (versions!), hardware (GPU model if relevant).
Hyperparameters: table of search range and final selection.
Baselines: justify each baseline choice; why is it fair?
Seeds: "N=5 independent runs with random seeds [list]."
Ablations: what's being tested in each ablation, why?
Dispatch methodology-advisor before drafting this chapter (DESIGN mode) and
again after drafting it (CRITIQUE mode) — it is the kit's read-only methodology
expert and this chapter is exactly its INPUT CONTRACT's domain (classic empirical
design + ML experiment design):
What: "Propose a methodology for these RQs" (DESIGN, before Section 3.1 is
written) or "Audit this Methods draft — run the replicability test" (CRITIQUE,
after /draft methodology produces drafts/methodology.md)
Context: RQs/hypotheses from PROJECT.md, field, data availability/constraints;
for ML theses also task type, datasets under consideration, candidate baselines
Draft path (CRITIQUE only):drafts/methodology.md
The agent returns (read-only, no edits): verdict (sound | fixable | redesign-needed)
or the recommended design, numbered issues/decisions, a validity-threat table, and
the matching reporting-guideline pointer. Apply its fixes before /critique —
reviewer-critic will independently re-score the same weaknesses if they reach a
full peer review unaddressed.
Chapter 4: Results / Findings (18–25%)
No interpretation; facts only. All interpretation → Discussion.
Organization per RQ:
4.1 Results for RQ1
4.1.1 [Specific outcome heading]
[Table / Figure]
4.2 Results for RQ2
[analogous structure]
Discipline:
Heading text must match the RQ phrasing (word_budget.py --rq-trace checks this).
Describe figures/tables in text (never: "see Figure 3").
Report with precision: means ± std (over N runs), confidence intervals, p-values.
For qualitative: theme labels + representative quotes with page numbers.
ML theses:
Comparison table: method (rows) × criterion (columns); bold best per column.
Include baseline results (necessary for fair comparison).
Report N seeds, mean±std, statistical test of significance (if applicable).
Ablation results (what happens when you remove component X?).
Chapter 5: Discussion & Conclusion (18–25%)
Structure:
5.1 Interpretation of Findings (per RQ)
5.2 Comparison with Prior Work
5.3 Limitations
5.4 Practical Implications
5.5 Future Work
5.6 Conclusion (summary statement)
5.1: Interpretation
Restate each RQ, answer it with your result.
Explain why the result occurred.
Use hedging language appropriate to evidence strength (demonstrates > indicates > suggests).
5.2: Comparison
How do your results align with / diverge from the literature?
If conflict exists, propose an explanation (context differences, methodological, sample).
Cite both supporting and contradicting prior work; don't hide conflicts.
5.3: Limitations
Name constraints: sample size, generalizability scope, design trade-offs, confounds.
For each, describe its impact: "This limits our ability to [X], relevant for [scenario]."
NOT: "We had a small sample" (obvious). INSTEAD: "N=20 limits power to detect medium effects (β>0.20)."
5.4: Practical Implications
Who benefits from this work? How?
What should practitioners do differently based on these results?
What should policymakers or institutions consider?
5.5: Future Work
Logical next steps emerging from your findings or unresolved limitations.
Avoid vague ("Further research is needed").
INSTEAD: "Future work should test whether [mechanism] generalizes to [new population], which would require [design change]."
5.6: Conclusion
3–5 sentences. Restate the thesis claim, main finding, and significance.
One-sentence vision: "This work opens the possibility of [X]."
4. Turkish Theses (YÖK Compliance)
Full Turkish-specific requirements, guidelines for Özet↔Abstract equivalence, and how to ingest your institution's guideline are in reference/yok-tr.md. Key points:
Front-matter order (non-negotiable in Türk univ.):
Inner cover (iç kapak)
Approval page (onay sayfası)
Ethics declaration (etik beyan)
Acknowledgments (teşekkür)
Turkish summary (Özet) — if Turkish is the thesis language
English summary (Abstract)
Table of contents (İçindekiler)
List of tables (Tablo listesi)
List of figures (Şekil listesi)
Özet↔Abstract equivalence: Same structure, same paragraph breaks, same keyword list (in respective languages). Not a free translation; a parallel rendering with identical claims and evidence markers.
Institute guideline: Drop your PDF/DOCX guideline into templates/institution/GUIDELINE-NOTES.md (user manually extracts rules); this skill then references it as authoritative.
5. Defense Preparation
Full committee-question templates, answer frameworks, and presentation outlines are in reference/defense.md. Quick summary:
Likely questions by chapter:
Ch.1 (gap & novelty): "Why is this problem important? What makes your approach different?"
Ch.2 (prior work): "Which prior work is most relevant? How does it compare to yours?"
Ch.3 (methods): "Why did you choose this design? What are the limitations?"
Ch.4 (results): "What surprised you? How reliable are these findings?"
Ch.5 (claims): "What's the take-home message? Who benefits from this?"
Presentation skeleton: 12–15 slides — one per chapter + outcomes slide. See reference/defense.md for detailed outline.
6. Integrated Commands
Check thesis structure:
# Verify RQ ↔ Chapter 4 heading ↔ Chapter 5 answer mapping
python scripts/word_budget.py --rq-trace
# Summary of structure and word budget
/status
Audit Turkish Özet↔Abstract:
# Run style-auditor in TR mode
/critique # or manual check against reference/yok-tr.md checklist
