These two cases are constructed teaching composites. Their people, proposals, and events are fictional; neither case describes an actual applicant, reviewer, institution, or funding decision. They are designed for short decisions under uncertainty, not for guessing a hidden correct outcome. Students should use only the synthetic text supplied in class and must never paste confidential, proprietary, unpublished, or review-protected material into an AI tool.
Dr. Rivera is reviewing a confidential federal research proposal. With the critique deadline approaching, a colleague suggests pasting the proposal into a public generative-AI service to obtain a summary and draft review. The colleague says that removing names should be enough and adds, “Applicants use these tools too.” Dr. Rivera instead considers paraphrasing the proposal’s novel idea in the tool. For class, one group receives an NIH-reviewer version of the case and another receives an NSF-reviewer version. No actual proposal text is distributed: the instructor supplies only this fictional description and, if needed, an instructor-written synthetic paragraph.
Create a three-column allowed / prohibited / context-dependent decision and a five-step safe workflow. The workflow must identify the person’s role, the governing rule, the information classification, the permitted tool and task, and a stop-and-consult step for ambiguity. It must explicitly prohibit entering proposal content, original concepts, review records, or a critique into the proposed public tool. Students make the decision on paper or in the course workspace; they do not test it by uploading text anywhere.
One minute to identify the role; two minutes to classify the information and proposed actions; three minutes to build the decision and workflow; one minute to compare the NIH and NSF branches; one-minute whole-class check that applicant and reviewer rules were not conflated.
For NIH peer review, generative AI may not be used to analyze applications or formulate critiques, and application content or original concepts may not be uploaded or shared with online generative-AI tools. For NSF review, proposal content, review information, and related records may not be uploaded to non-approved generative-AI tools; NSF’s notice separately says reviewers may share publicly available information with current-generation tools. Neither rule becomes optional because names are removed or because an applicant may have used AI.
Applicant rules are a different branch. NSF encourages proposers to describe their generative-AI use and holds them responsible for proposal accuracy and authenticity. NIH’s applicant notice says AI may assist limited aspects in some circumstances but treats applications or sections substantially developed by AI as not representing the applicant’s original ideas. A classroom exercise is not federal peer review, but it still depends on course, institutional, tool, privacy, and intellectual-property rules. Public or instructor-created synthetic material can make a classroom exercise possible; it does not authorize using protected review material. When authorization is unclear, stop and consult the designated agency or institutional official rather than improvising an exception.
An early-career team at a less widely known institution proposes a bioinspired sensor. The proposal presents a strong literature-grounded rationale, a novel integration of two methods, prespecified controls and milestones, a risk-mitigation plan, relevant publications from the team, a collaborator covering one missing specialty, and documentation that the needed facilities are available. Its preliminary results are limited to calibration and feasibility checks; it does not yet contain outcome data for the central experiment. During preliminary discussion, one reviewer says, “I do not recognize this principal investigator or institution. Without results from a famous lab, no preliminary data means this is unfundable.” The instructor assigns each group either (a) the current NIH three-factor framework for a covered research project grant or (b) NSF’s Intellectual Merit and Broader Impacts criteria, together with the relevant fictional funding-opportunity excerpt. That excerpt explicitly states whether preliminary evidence is required, encouraged, or left to criterion-based judgment for this exercise.
Write a three-sentence constructive critique: (1) one criterion-linked strength, (2) one bounded evidence gap and why it matters, and (3) a conditional request, clarification, or risk-reduction step. For the NIH branch, distinguish Factor 1: Importance of the Research, Factor 2: Rigor and Feasibility, and Factor 3: Expertise and Resources, treating Factor 3 as an evidence-based sufficiency assessment rather than a reputation ranking. For the NSF branch, apply Intellectual Merit, Broader Impacts, and the relevant review elements without using prestige as a proxy for qualifications or resources.
One minute to mark evidence, inference, and missing information; two minutes to map the assigned criteria; three minutes to draft the critique; one minute to exchange it with the other agency group; one-minute revision to remove any unsupported prestige or categorical preliminary-data claim.
The correction is not that preliminary evidence, prior accomplishments, expertise, or facilities never matter. It is that reviewers should evaluate the proposal’s documented evidence under the applicable criteria and the specific funding opportunity, explain material gaps, and avoid treating fame as evidence. Under NIH’s simplified framework for covered research project grants, Importance and Rigor/Feasibility are scored factors, while Expertise/Resources is assessed for sufficiency and, if not sufficient, specific gaps. NSF uses Intellectual Merit and Broader Impacts, may add solicitation-specific criteria, and also considers whether the plan is well reasoned, the team is qualified, and resources are adequate. These frameworks support conditional, evidence-linked critique; they do not guarantee funding or make every proposal with limited preliminary data competitive.