Vivek is a diagnostic neuroradiologist at Johns Hopkins and an NIH study-section reviewer. He came to Granted to test review-mode — feed an R01 draft to the AI committee and see whether the critique resembled a real study section. The early versions did not. Six reviewers scored equally with no concept of primary-vs-discussant weighting. The 1–9 scale was not tied to the five core criteria. A "Red Team" archetype dragged the panel structurally harsher than reality. Vivek named each of these in detail. What he shaped at Granted: the assigned-trio weighting (Reviewers 1–3 carry 2× weight in consensus vs. Reviewers 4–6 as discussants), the full 1–9 scale aligned to Significance / Investigators / Innovation / Approach / Environment, and the rename of Reviewer 6 from "Red Team" to "Independent Outside Reviewer" — so the panel reflects the center of mass of a real study section, not a global skeptic prior. His R01 in stroke-imaging biomarker development is what tested all of it. He also surfaced the file-handling bug that taught Granted to extract draft text server-side before the committee runs, and to hard-fail when reviewers signal the input is unreadable rather than retry until critique is fabricated. Review-mode looks like a real NIH study section because Vivek made it look like one.
The summary document seems to weight all six reviewers equally, which seems less realistic.
