Every thesis submission season, the library and the degree office see the same loop: a student receives rejection comments — wrong line spacing, a chapter that doesn’t start on a new page, references that don’t follow the citation standard; back to the lab to fix the format, reprint, re-bind, and queue again. A thesis bound by thread or hot glue cannot be “patched” with a stapler; every rejection is a full redo.
Most people blame staffing. We want to offer another perspective: the problem is not a lack of hands; it is that format checks happen too late. Every stage a check is delayed, the cost of fixing grows.
One rejection, three parties repeating the same work
Break the account down: for the student, one rejection means re-checking the whole document, reprinting, and re-binding; for the department secretary, it means going through the same thesis again; for the library and the degree office, it means sustained piles at peak season. All three are doing the same thing — checking the same specification, page by page.
Having the same specification manually executed three times is waste by design — and it still misses things, because humans checking against rules are precisely the most fatigue-prone and least consistent link in the chain.
A thesis specification is, in fact, a list of rules
Take a doctoral and master’s thesis specification revised in 2026 at a leading university and read it line by line: it is naturally machine-executable. By the difficulty of machine execution, it falls into four layers:
- Purely mechanical rules: body text in a specified Chinese typeface, English and numbers in Times New Roman, 20-point line spacing, each chapter starting on a new page, double-sided printing, thread or hot-glue binding — these are typesetting parameters a machine judges item by item;
- Structural rules: whether all twelve cover elements are present, the order of abstract, table of contents, and list of figures and tables, the “required if present” status of the figure and table list, the position of the classification marking on restricted theses — this is document structure a machine can check in full;
- External standards: whether references follow GB/T 7714, whether the Chinese Library Classification number is filled in — the standards are public and the judgments deterministic;
- Content-linked rules: the Chinese abstract generally 300–1,000 words, 3–8 keywords, the English abstract “should correspond to the Chinese abstract,” a title within 30 Chinese characters — here the check slides toward content.
The first three layers, machines can fully execute today. The fourth already touches the threshold of content review — a deeper topic this article deliberately stops short of, with one sentence: what it takes is not faster machines, but AI that understands a school’s own rules.
Collapse three gates into one pre-check
Today’s three gates: the student self-checks by feel — for most people, the first serious reading of the specification comes after the first rejection; the department’s formal review; the library’s and degree office’s acceptance check.
The pre-check division of labor: before submitting, the student lets the machine execute the specification in full and receives an item-by-item check report; the department reviews the report and the exceptions instead of the thesis; the acceptance stage handles appeals and sampling only. A pre-check does not add a gate to the process; it consolidates the repeated labor of three gates so the machine does it once.
This is also what distinguishes it from the student “going through the specification line by line”: a human reads a specification with comprehension, by sampling, and with fatigue; a machine reads it in full, consistently, and repeatably.
The prerequisite: translating the specification into a maintainable rule base
The foundation of a pre-check is translating specification clauses into structured rules: each rule carrying its clause number, its scope of application (academic degrees differ from professional degrees; restricted theses have dedicated marking rules), and an executable judgment. This translation is the first cost — and the first asset: it turns the specification from a document for printing into a list for execution.
More important is maintainability. Specifications get revised — the one mentioned above is explicitly marked “June 2026 revision.” If any cover element, abstract length, or citation standard changes, the rule base must be updated with it, or the pre-check tool will not survive one semester. The rule base’s version should follow the specification’s version.
Every judgment in the rule base comes from the school’s own specification and rubric — exactly the context problem Education Intelligence solves (see Education Intelligence): AI can run these checks for a school not because it is general, but because it has read that school’s rules.
A pre-check opinion is not an audit decision
The boundary must be clear: the pre-check report is an advisory self-check input, not an audit decision. Formal review authority stays with the department and the degree office; a student who disputes a machine judgment keeps a channel of appeal — the rule base can have gaps, and appeals are precisely how they surface.
Academic judgment is even further outside: originality, scholarly quality, plagiarism checks, and research integrity belong to review and dedicated processes. AI checks “compliance,” not “quality”; holding this boundary is what keeps the pre-check from becoming a new source of dispute.
The metric must follow: whether a pre-check works is not measured by how many students use it, but by how much the rejection rate and round-trips fall — the same logic as “measure the right thing” in The Software Is Live. Why Aren’t Teachers Using It?
A specification’s value lies not in being printed but in being executed; the season’s staff time should go to exceptions, not repetition. The best format review is the one where the student knows the result before submitting — move the check to before submission, and the thesis-season queue will finally get shorter.
If your school is looking for a more efficient path for thesis format review, we would love to hear from you — contact us.
