To judge the progress of a graduation thesis, don't ask "How far along are you?"—the answer to that is usually "Pretty good, I'm writing." A more reliable question is: How many files do you have in hand right now?
A thesis isn't written in one breath; it's accumulated file by file. The output of the previous step is the input for the next. Once the files are all there, the thesis naturally comes together. This post lists 14 files in chronological order so you can check which ones you're missing. The tool used as an example is MuchSci's Paper Sailor, whose feature happens to fit this explanation perfectly: every step produces a real file; no step ends with "we talked and nothing was left behind."
I. Proposal season: first, build two files
File 1 | Idea Card / Research Question Card. Turn "the broad direction your advisor gave you" into "a question you can write about." What it does is generate an idea card from an idea and distill a research question—don't underestimate this step. More than half of the topics rejected at proposal defenses fail because the question is "big and empty."
File 2 | Proposal first draft. Turn the research question into a Proposal. This is the first formal deliverable you can use to align with your advisor. Starting to write before the direction is aligned is the biggest source of rework in a graduation thesis.
Figure 1 | Idea card and Proposal: the confidence for the proposal meeting comes from these two files.
II. Literature season: three files, none can be skipped
File 3 | Literature library. Multi-source search (arXiv / Semantic Scholar / MCP service for deduplication), automatic PDF download, and metadata storage. MuchSci Paper Sailor specifically builds in an authenticity check on this path—the library contains only entries actually returned by search and ingestion; hallucinated references have no entry channel.
File 4 | Literature review first draft. It grows out of the library; it is not written out of thin air. Later, writing the introduction and the research status section will continue from this file.
File 5 | Citation entries (Bib / RIS). Citation management + three-way verification. Build them now, cite easily while drafting the main text, and avoid last-minute format panic before finalization.
Figure 2 | Literature library, review first draft, citation entries: only real entries are allowed into the library.
III. Blueprint and data: four files set the whole picture
File 6 | Thesis outline + chapter skeleton. While generating the outline, annotate the dependencies of each chapter: which ones do not depend on data and can be written first, and which ones must have real results (data_needed, waiting for experiments). This "map" determines what you do first and what you do later over the coming months.
File 7 | Data profiling report. Upload CSV / Excel / JSON and automatically use pandas to produce fields, distributions, and missingness.
File 8 | Gap action list. The output of gap mapping between the profile and the outline: where data is sufficient, where it is missing, and what to fill first. Missing data is not scary; discovering missing data only when you write the results chapter is scary—this list is for clearing mines in advance.
Figure 3 | The outline skeleton annotates chapter dependencies, and the data profile connects to the gap list.
IV. Execution season: drafts, results, and figures
File 9 | Main-text drafts for each chapter. Proceed according to the blueprint: write chapters that can be written first, and leave data_needed chapters blank while waiting for real data—leaving them blank is the rule, not a flaw.
File 10 | Data analysis and subset filtering results. The premise is that you provide real experiments/data—it does the work; authenticity is on you.
File 11 | SVG vector figures. 8 chart types + multi-panel assembly, with journal column widths such as Nature 89 mm. Undergraduates may not need journal column widths, but vector figures don't blur when enlarged and stay crisp when pulled directly into defense PPTs.
Figure 4 | Final self-check: when the six items on the left are all in place, the thesis holds together.
V. Finalization season: the last three things
File 12 | AI-sounding index report. The rule engine detects boilerplate transitions, empty strong assertions, and translationese; rewrite high-severity items one by one until they reach zero. Do this before submission; the reason needs little explanation.
File 13 | Review backups (v1 / v2 / v3). A review state machine with up to 3 rounds, with automatic backup each round. When your advisor asks you to revert to the previous version, you'll thank them.
File 14 | Final manuscript + export. Choose the language version once before delivery (Chinese manuscript/English manuscript; the table of contents and filenames change accordingly); if you want PDF, you need the local version (the server version has no LaTeX environment); if the school only accepts Word, confirm in advance.