AI writing tools all have similar taglines: smart, efficient, one click. But what really determines "how it works and what you do" is the set of terms inside it. If you don't understand the terms, you can't use the features fully; once you do, you may not even need the tutorial.
This article lays out the 8 keywords of MuchSci Paper Sailor in work order, giving each one a one-sentence definition plus a little usage advice. If you're a senior who has just received your thesis topic this season, you can get started after reading this.
Figure 1 | The 8 terms in work order: understand the terms, and you understand the human–machine division of labor.
01 | Idea Card
One-sentence definition: The smallest artifact that turns a vague idea into something discussable.
Your advisor gives you a broad direction, and writing the thesis directly from it will definitely go off track. The Idea Card is the first narrowing of that direction: one direction can generate multiple cards, and you pick the most defensible one and condense it into a research question. Before proposal season, generate several and compare them side by side before deciding.
02 | Authenticity Check
One-sentence definition: The security gate before literature enters the library.
Entries returned by literature search (arXiv / Semantic Scholar / MCP multiple sources) must pass the Authenticity Check before they are allowed into the library—only literature actually returned can become your references, and "literature" hallucinated by AI cannot get through this gate. Writing reviews and building citation entries (Bib/RIS) are both based on a library that has passed the security check.This is the precondition that lets you confidently put citations in your references.
03 | Writing Blueprint
One-sentence definition: A chapter-level construction plan that clearly marks what each chapter depends on.
Generating an outline is only the first step. The key to the blueprint is annotating dependency relationships for each chapter: which chapters do not depend on data and can be written now, and which must wait for real results. With this map, the order of writing the introduction and literature review first and the results and discussion later is not decided by guesswork.
04 | data_needed
One-sentence definition: A placeholder marker meaning "this chapter must wait for real data."
This is the most important of the 8 terms and the most distinctive design of MuchSci Paper Sailor: the Writing Blueprint marks chapters that depend on experiments/data as data_needed, leaves placeholders in the main text, waits for your real data, and never fills in numbers on its own. A chapter being blank is not a defect; it is the rule—it blocks the most painful kind of failure in a graduation thesis: nonexistent data appearing in the results chapter.
Figure 2 | data_needed: blank is the rule; it is waiting for real data.
05 | Data Profile
One-sentence definition: A health check for your data first.
Upload CSV / Excel / JSON, and it automatically uses pandas to produce fields, distributions, and missing-data conditions, then maps gaps against the outline—what is missing and what to fill first—turning it into an action checklist. Profile your data before you start writing, and the problem of "I can't write the results chapter" is eliminated before writing begins.
06 | Review State Machine
One-sentence definition: Revision is not an infinite loop; at most 3 rounds.
Each review round is automatically backed up and can be rolled back, so no matter which version you revise to, there is a record. It solves the old problem of AI writing tools—"the more you revise, the messier it gets, and you can't go back." When your advisor asks you to revert to the previous version, the backup is your regret pill.
Figure 3 | Review State Machine: 3-round cap, per-round backup, rollback supported.
07 | AI-Flavor Index
One-sentence definition: A quantitative report of "how much your draft sounds like AI."
The rule engine checks paragraph by paragraph for AI-speak such as boilerplate transitions, empty strong assertions, and translationese, and produces an index report; high-scoring items are rewritten one by one until they reach zero. Schools are now increasingly sensitive to AIGC traces, so finish this item before finalizing your draft; only then can the draft you submit withstand scrutiny.
Figure 4 | AI-Flavor Index: detected items are downgraded one by one until they reach zero.
08 | Language Version
One-sentence definition: Chinese draft or English draft—you choose once before delivery.
After you choose, the output directory and file names all change accordingly; there will be no mixed Chinese and English, and no one will decide on your behalf. What this design is really saying is one sentence: the tool does the work for you, but it does not make a single decision for you.
Read These 8 Terms Together
Idea Card sets the direction, Authenticity Check guards the literature, Writing Blueprint schedules the work, data_needed stays blank waiting for data, Data Profile clears mines in advance, Review State Machine manages revisions, AI-Flavor Index handles the finishing pass, and Language Version returns the decision to you—behind the terminology is a division of labor: it is responsible for doing the work and exposing gaps; you are responsible for providing truth and making decisions. Reverse that division, and you have an accident.