What the Research Proposal Reviews
This rubric looks at whether your proposal presents a focused, answerable research question; explains why the answer matters; and gives a workable plan for data, methods, evaluation, risks, and sources. It also checks that you clearly separate your own work from code, guidance, or materials you start with, and that your claims stay within what your proposed data can support.
What We Check
Whether the proposal states one research question that names what will be studied and what will be measured or compared, and that a result could answer.
Whether the proposal gives a reason the field, a named group, or a practice needs this answer, beyond the writer's own interest.
Whether the proposal names its data, where the data came from, how much there is, what one record looks like, and what the data covers.
Whether the proposal names a method, matches it to the question, names the implementation it starts from, and separates the writer's own work from what the program, mentor, or existing code provides.
Whether the proposal names a measure that fits the task, a number to compare it against, a plot, and a way of looking at individual results.
Whether the challenges named belong to this project, each has a response, and the work is bounded to fit the program.
Whether the references are locatable, tied to sentences in the proposal, and include the sources the project actually builds on.
Mistakes We Flag
- A topic instead of a testable question
- Naming a field, tool, or dataset does not show what you will measure or compare. State what you will study and the quantity or comparison your project can answer.
- Explaining only your personal interest
- Your interest is welcome, but the proposal also needs to name a current gap, limitation, cost, error, or need in the field or for a specific group.
- Naming a data host but not the data’s origin
- A site where you found data is not necessarily who collected or published it. Explain who collected the data or how you will collect it, and provide the dataset’s own source.
- A vague method or unclear ownership
- Readers need to know what the model takes in and produces, what existing code or method you will start from, and what specific change or experiment you will do yourself.
- An evaluation plan without a real comparison
- Model confidence or a general statement that you will see how well it works does not test the result. Name a task-appropriate metric, compare it with a meaningful reference, and plan to inspect both plots and individual examples.
How this rubric reads your essay
- Built from published reader guidance — the sources are listed below.
- Scored per section against defined criteria, not general impressions.
- Calibrated for consistent scoring, so a better draft shows up in the score.
References
The course proposal questions the Inspirit template mirrors — the problem and why it is interesting, the challenges, the dataset and how it will be collected, the method, and how results will be evaluated, at 300–500 words — and the milestone's ask for a model description and a training strategy such as the loss function.
Asks for a scientific question, which parts will be implemented versus downloaded, baselines, and at least one well-defined numerical evaluation metric with the scores it will be compared against.
Explicit input and output, at least five related-work references, how many training, validation, and test examples and what preprocessing was done, a citation for the dataset's source, primary metrics explained before results, both quantitative and qualitative results, examples of failures, and whether the model overfit.
Evaluation criteria: technical quality, significance (a real problem or only a small toy problem), novelty, and clarity of the write-up.
The research plan components written before experimentation: rationale, research question or hypothesis and expected outcomes, procedures, risk and safety, data analysis, and bibliography; AI may be used as a project resource but must be cited.
Procedures must include the methods for data collection and, when applicable, the source of data used, and must delineate what the student will do and what will be done by the mentor.
Research Question scored first: a clear and focused purpose, a contribution to the field, and testability; Execution scored on reproducibility of results, appropriate statistical methods, and sufficient data to support conclusions; the interview on understanding the limitations of results and the degree of independence.
Milestone rubric: a precisely stated, well-scoped problem; whether risks are highlighted; preliminary results presented and interpreted, or a concrete plan with a realistic timeline.
Lists lack of focus, too much project, and lack of planning among the pitfalls reviewers see; asks for a design manageable within the timeframe.
A good research question is clear, focused, and not answerable with a simple yes or no.
Judging criteria include the statement of the research problem, logical conclusions relevant to it, whether students recognize their contribution to the field, skill in communicating results, and references stated; the abstract must include the hypothesis and conclusions; presenters should avoid jargon or explain specialized terms.
Recommends peer-reviewed articles, textbooks, and official sites; asks writers to avoid Wikipedia and blogs.
Meaningful and naive baselines, accuracy on imbalanced classes (a classifier that always outputs the larger class scores 90% while being useless), sequential overfitting when tuning on the test set, and not generalizing beyond the data.
Frequently Asked Questions
What should my research question include?
It should name what you will study and a relationship, quantity, or comparison you can measure. A question that only names a topic or asks a simple yes-or-no question is not enough.
What should I say about my dataset?
Name the data, its original collector or publisher or your collection method, its size, what one record looks like, and what the data covers. If the data involve people, address consent, licensing, de-identification, or ethics.
How do I explain what work is mine?
Name the codebase, library model, notebook, pretrained weights, or paper method you will start from, then state the specific change, addition, or experiment you will do. Make clear what a mentor or existing code provides.
What belongs in the evaluation plan?
Name a numerical measure that fits the task, a meaningful number or reference to compare against, a specific type of plot, and a plan to examine individual results. For classification, accuracy alone may not be enough when classes are imbalanced.
Get your draft scored across all 7 dimension with specific, actionable feedback. Two free reviews per day — no credit card required.