Grant proposal review that cites every finding to the clause it comes from
Every proposal reviewed inside the five-day window, every value tied to its clause.
See one case, screen by screen ↓Software that checks a proposal's parts, budget, biosketches and forms against the sponsor's own rules before it is submitted. Grant Proposal Review adds seven agents that read the parts, rebuild the budget, find disagreements between parts and check each senior or key person's forms — a named specialist decides, and only the authorized organizational representative submits.
AI in education: a buyer's guide to agentic AI for colleges, universities and school districts →
Updated
Grant proposal review is the work of checking a dozen separate files against a sponsor's rules before a proposal can go out, and a research office gets five business days to do it. The files were written by different people in different weeks — the narrative, the budget, the budget justification, each senior and key person's biosketch and Current and Pending (Other) Support — and nothing forces them to agree with each other. A co-investigator's effort in the justification can read differently from the effort on her own support form. A person named in the narrative can be missing from the budget.
The budget is arithmetic nobody re-does by hand: facilities and administrative costs charged on a whole subaward instead of the first $50,000 the rule allows, or a salary budgeted above the federal cap. The forms move too — from January 25, 2026 the sponsor requires a Common Form with a signed certification and an ORCID iD for every senior and key person, and an old-format biosketch gets the application returned. None of this is usually found until a reviewer counts pages by hand on the day it is due — and a sponsor that finds it first withdraws the application before review, with no second chance until the next cycle.
Estimated hours for a twelve-part proposal like Halden State's P-2026-01184, by hand against with the agents doing the reading.
Seven agents read every part against the sponsor's rules and rebuild the budget; a named person decides, and only the representative submits.
Halden State University, proposal P-2026-01184: an NIH R01 from Dr. Kwame Adeyemi-Lund, College of Engineering, due February 5, 2026 at 5:00 PM. Here is what happens next, screen by screen, in the working solution.
Tomas Eklund routed the package on Thursday, January 29 at 4:12 PM — day 1 of the office's five-business-day window. Today is Tuesday, February 3: day 4 of 5, two business days to the sponsor deadline. Lena has it: 12 parts, 61 pages, already read by seven agents, with 11 findings pinned to the parts they belong to.
12 parts · 61 pages · 11 findings, 8 blocking
Finding R-02: the budget justification commits Co-Investigator Dr. Ines Halvorsen to 2.4 calendar months; her Current and Pending (Other) Support form shows 1.8 person months for the same application. The agent found it at 96% confidence and returns both passages, cited to their page and line — it never chooses between them.
“2.4 calendar months” against “1.8 person months,” both cited
Lena asks the department, confirms 2.4 is right, and accepts the agent's proposed correction to the Common Form with reason code COR-02, "Corrected at the office's request." Her name, the time, the reason and the evidence are written to the audit trail.
COR-02 · Corrected at the office's request
Facilities and administrative costs were charged on the whole $612,000 Brightwater University subaward. Under 2 CFR 200.1, only the first $50,000 of each subaward belongs in the base. The agent proposes a base of $1,808,000 instead of $2,370,000 for this line alone — $1,012,480 instead of $1,327,200 at the negotiated 56% rate — at 98% confidence, blocking.
Only the first $50,000 of a subaward is in the base — 2 CFR 200.1
Every budget line, in-base or out, with the reason: equipment $90,000 out, tuition remission $48,000 out, and only the first $50,000 of the subaward in. Lena accepts the recomputation; the request falls from $3,787,200 to $3,354,340 — $432,860 that would otherwise have been cut after award or questioned later.
$3,787,200 → $3,354,340
Dr. Adeyemi-Lund is budgeted on an institutional base of $264,000 at 3.0 calendar months, asking $66,000 a year. The federal salary cap is $228,000 from January 1, 2026, so the most that may be requested is $57,000 — over by $9,000 a year. The other three senior and key persons are under the cap.
$66,000 asked · $57,000 allowed
Accepting the recomputation closes three findings at once: a base of $1,701,500 instead of $2,370,000, a salary request of $57,000 a year instead of $66,000, and a total request of $3,354,340 — $432,860 lower. Nothing was removed from the science.
Base $2,370,000 → $1,701,500
Four people's effort agrees across the budget, the justification and the Common Form; Dr. Halvorsen's does not — that is finding R-02. Dr. Adeyemi-Lund's committed effort across all his awards totals 12.9 person months now, over the 12-month year, and 10.1 person months at the project's July 1, 2026 start, once two other awards end March 31.
12.9 person months now · 10.1 at the project start
Two of the four senior and key persons would be rejected at submission: Dr. Halvorsen's biosketch is still the five-page 2024 format with no certification, and Dr. Soo-Min Rhee's Common Form has no ORCID iD in the persistent identifier field. Common Forms are required for due dates on or after January 25, 2026.
2 of 4 forms would be rejected at submission
With blocking findings still open, the submit control stays locked and names the count — only Marcus Reyes, the authorized organizational representative, may press it once it is not. Dana waives one advisory finding with reason code WAV-04, "Sponsor guidance is a recommendation, not a limit," and signs the reason.
Locked while a blocking finding is open
Every check, fix, waiver and the submission itself land on one trail, person and agent alike — from the intake agent splitting the package at 08:15 to Lena's accepted recomputation. Marcus signs and submits on day 4 of 5, two business days before the deadline, and the evidence pack exports for a sponsor inquiry.
Submitted day 4 of 5 · two business days in hand
The same proposal, seen by the five people who own it — what their week looks like, and what it looks like now.
Seven agents read every part against the sponsor's rules and rebuild the budget; a named person decides, and only the representative submits.
Pulls the package from the research administration system, splits it into its named parts, counts pages, lines and characters, and picks the sponsor rule pack for the mechanism and the due date.
Checks a single part against its own format and content rules — page and line limits, required headings, required elements, font and margin properties — and pins what fails, citing the clause and the page.
Rebuilds the modified total direct cost base line by line with the required exclusions, applies the negotiated rate, tests every salary against the sponsor's cap, and checks escalation, fringe rates and cost sharing against policy.
Compares the parts with each other: effort in the justification against Current and Pending, people in the narrative against the budget, aims referenced against aims written, subawardee in the letters against the budget, dates against the project period.
Reads each senior or key person's Biographical Sketch and Current and Pending (Other) Support against the Common Form structure, confirms the ORCID persistent identifier and the signed, dated certification.
Watches the sponsors' published guides and notices, drafts the change to a rule pack with its effective date and source, and puts it to a named approver.
Assembles the evidence pack, re-runs every check against the final files, confirms no blocking finding is open, and hands the package to the authorized organizational representative.
Accept, waive, sign. The agents propose; a named person decides.
Ask about any part, finding or number on a proposal — or say what to change. Every answer points to its evidence.
Why do the effort numbers disagree?
The budget justification commits Dr. Ines Halvorsen to 2.4 calendar months; her Current and Pending (Other) Support form shows 1.8 person months for the same application. On the Common Form, effort is in person months only, so 2.4 calendar months of a 12-month appointment should read 2.4. The fix is to regenerate her Common Form at 2.4.
Is anyone over the salary cap?
One. Dr. Kwame Adeyemi-Lund is budgeted on an institutional base of $264,000 at 3.0 calendar months, which asks for $66,000 a year. The cap is $228,000 for awards issued on or after January 1, 2026, so the most that may be requested is $57,000. The other three senior and key persons are under the cap. Across five budget periods that is $58,500 in salary and fringe, and $32,760 in facilities and administrative costs.
Which forms would be rejected?
Two of the four senior and key persons. Dr. Halvorsen's biosketch is the 2024 five-page format, not the Common Form, and it carries no certification. Dr. Rhee's Common Form has an empty persistent identifier field — the verified identifier is on the researcher profile and can be carried across.
What is still blocking submission?
While any blocking finding is open, the package stays locked and the assistant names exactly which ones. Once every blocking finding is closed and each advisory one is decided, the package unlocks — and only the authorized organizational representative may submit.
13 screens from the working solution, on its sample data. Pick one to see it large.
The window, the approvers and how the agents work are settings, chosen by the research office.
| Setting | Default | Choose from |
|---|---|---|
| Internal review window | 5 business days before the sponsor deadline | set by the Director of Sponsored Programs |
| Blocking severities | Format limit, budget arithmetic, salary cap, Common Form, missing required part | set by the director |
| Waiver reason codes | WAV-01 to WAV-04 | WAV-01 Effort resolves before the project start · WAV-02 Sponsor confirmed in writing · WAV-03 Institutional policy exception approved · WAV-04 Sponsor guidance is a recommendation, not a limit |
| Negotiated rate | 56% of modified total direct costs, on-campus research | set by the administrator, for the rate agreement's term |
| Salary cap table | $228,000 from January 1, 2026 | set by the administrator, per sponsor and effective date |
| Agent confidence threshold | Findings below 85% go to a specialist rather than closing | set by the director |
| Student names in budgets and letters | Masked for anyone outside the office | set by the administrator |
| Retention | 3 years after the final financial report | set by the administrator; redaction applied on every export |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample proposal data · “target” = the design goal, measured in the live solution · “typical” = published figures (NSF NCSES Higher Education Research and Development Survey, FY2024; Federal Demonstration Partnership Faculty Workload Survey; current federal salary cap notices). Regulatory context: the federal cost-principles rule on subaward costs in the facilities and administrative base; sponsor Common Forms requirements effective January 25, 2026. People, the university, proposals, budgets and persons named on this page are fictional — characters and sample data in the working solution.
Software that checks a proposal's parts, budget, biosketches and forms against the sponsor's own rules before it is submitted. Grant Proposal Review adds seven agents that read the parts, rebuild the budget, find disagreements between parts and check each senior or key person's forms — a named specialist decides, and only the authorized organizational representative submits.
The Budget agent rebuilds the facilities and administrative base line by line — equipment, tuition remission and the portion of each subaward above the first $50,000 out — applies the negotiated rate, and tests every salary against the sponsor's current cap. It shows its working as a worksheet a person can accept or argue with.
The Cross-Document agent compares effort in the budget justification against Current and Pending (Other) Support, people named in the narrative against the budget, aims referenced against aims written, and the subawardee in the letters against the budget. It returns both passages, cited — it never chooses between them.
The Common Forms agent reads each senior or key person's Biographical Sketch and Current and Pending (Other) Support against the Common Form structure, and confirms the ORCID identifier and the signed, dated certification are present before the package can unlock.
Yes. Agents propose findings and a corrected worksheet; they never edit the proposal, sign a certification or submit it. A named specialist accepts or rejects each finding, only the director may waive one, and only the authorized organizational representative may submit.
Yes. The internal review window, which findings block submission, the waiver and correction reason codes, the negotiated rate and salary cap table, the confidence threshold, record masking and retention, and the approvers for waivers and submission are all settings.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your sponsor rule packs, rate agreement and a sample of past proposals — and it goes live once every quality gate has passed. We will walk you through it on your own proposals first.
Five high-fit calls in minutes, each limit and cost share cited to the sponsor's own text.
1-min demoSubrecipient Risk Assessment & MonitoringRisk scored before signing and every invoice line checked before the university pays.
1-min demoIRB Protocol Pre-ReviewEvery protocol back to its investigator on the first day, each fix tied to its rule.
1-min demoWe'll run Grant Proposal Review on a sample of your own recent submissions.