17.11 A grant proposal as a capstone artifact

There is a familiar institutional form for the transition from a scientific gap to an experimental program: the research-grant proposal. Scientists ask an agency to provide the equipment, personnel, data, travel, and time needed to test a theory that cannot yet be admitted. A competitive proposal must digest the literature, explain why the state of the art is incomplete, formulate a novel and consequential hypothesis, organize several years of specific aims, and make a credible case that the proposed experiments can decide something important. In that sense, grant writing is already a socially embedded form of theory-and-experiment compilation.

This analogy gives the GLP-1 challenge a tangible capstone. Its product need not be an encyclopedic model of incretin biology. It can instead be a grant-ready research strategy generated from a temporally bounded corpus and an incomplete theory:

Here \(\Omega \) is the localized gap, \(\mathsf S^{+}\) the proposed extension, \(\mathcal A\) the live alternatives, \(\mathcal E\) the linked experimental aims, and \(\mathcal R\) the expertise and resources required to execute them. The proposal is therefore not merely prose about a topic. Every aim should be traceable to the obstruction, every experiment should discriminate among declared alternatives, and every important premise should retain source and epistemic status.

The analogy is more than rhetorical. The current simplified NIH framework for most research-project grants asks reviewers to assess the importance of the research, the rigor and feasibility of the approach, and the adequacy of the expertise and resources [ National Institutes of Health , 2026 ] . These factors align respectively with the value of the obstruction and extension, the quality of the compiled tests, and the operational realizability of the program. They also expose a distinction that a theory constructor must respect: inventing a promising experiment does not imply that a particular team or institution can execute it.

The resulting capstone challenge, GLP1–GRANT, is intentionally ambitious. Given a frozen literature cutoff, a funding opportunity, and an explicit investigator-and-resource profile, the system produces a one-page Specific Aims document and a source-bearing research strategy. Its theory map, claim ledger, experimental dependencies, controls, milestones, stopping rules, and failure contingencies remain machine-auditable even when the narrative is written for human reviewers. A blinded panel can compare this artifact with a generic language-model proposal, retrieval-augmented synthesis, flat evidence-to-grant generation, and—where available—a human reference.

Actual funding is an aspirational external outcome, not a clean benchmark label. Awards depend on program priorities, budgets, investigator credentials, institutional resources, and review variability as well as scientific merit. The defensible initial endpoint is therefore narrower: can independent domain experts identify a real gap, follow every specific aim back to the proposed theory extension, and judge that the experiments would distinguish it from credible alternatives? Nor should the constructor fabricate administrative elements such as institutional commitments, biosketches, facilities, or human- subjects approvals; these must be supplied or completed by responsible human participants.

Experiment: GLP1–GRANT capstone: theory extension to research program. Epistemic status: capstone evaluation specification; the subsequent GRANT rungs are controlled precursors, not a completed capstone result.
Input: temporally bounded literature, maintained GLP-1 sketch, funding opportunity, and explicit team-and-resource profile.
Artifact: Specific Aims, source-bearing research strategy, theory extension, linked experiments, controls, milestones, alternatives, and resource requirements.
Evaluation: blinded scientific review against generic, retrieval- augmented, flat-synthesis, and human-reference proposals where available.
Admission: importance, rigor, feasibility, provenance, falsifiability, and coherence of the multi-aim program.
Boundary: grant-ready scientific strategy, not a claim of funding, clinical validity, institutional authorization, or autonomous submission.

The controlled studies below are organized by semantic obligation rather than by an attempt to perfect grant-writing style. GRANT–0 tests structural compilation; GRANT–1 adds epistemic types; GRANT–2 separates a typed intermediate program from its prose realization; GRANT–3 adds cross-field scientific invariants; and the GRANT–4 series asks how an admitted program can be rendered without omission, accretion, or loss of auditability. Each negative result freezes the failed artifact and changes one declared layer. The scientific object throughout is the source-bearing research program, not the fluency of its surface prose. In particular, GRANT–4 stress-tests one interface—the passage from an admitted typed program to human-readable prose—without upgrading the biomedical evidence of the underlying program.

. A research proposal is not the theory extension itself. It is an externally reviewable realization of that extension. Rhetorical fluency is admissible only when it preserves the source ledger, epistemic types, open decisions, resource boundaries, and discriminating structure of the underlying program.

The experiment identifiers preserve protocol provenance; they do not define the exposition. The following subsections are therefore layers of one constructor, with each frozen rejection exposing an interface that the next layer must represent explicitly.