ARPA-H's IGoR Is Trying to Rebuild Biomedical Research From Scratch. Here Is Why That Is Either Brilliant Or Catastrophic — And Why You Have Six Weeks To Decide Which.

The Advanced Research Projects Agency for Health quietly launched what may be the most architecturally ambitious biomedical research program of the decade on May 5, 2026. The Intelligent Generator of Research — IGoR for short — is a five-year, three-phase program that ARPA-H expects to fund through approximately three Other Transaction agreements. Solution Summaries are due June 25, 2026 at 12:00 PM Eastern. Full Proposals follow on August 6, 2026 at 12:00 PM Eastern. Proposers' Day is June 9, 2026 in the Washington, D.C. metro area. That is the entire critical timeline, and it compresses the largest strategic and team-formation decision of the year into roughly six weeks.

The headline is the technical vision: IGoR aims to deliver gold-standard biomedical science at least ten times faster than the field's current state of the art by combining mechanistic disease models, an AI orchestration layer that designs experiments to close knowledge gaps, a layered protocol architecture that any qualified laboratory can replicate, and a distributed marketplace of validated laboratories that execute those protocols and return data back into the model. The deeper story, however, is that ARPA-H is treating IGoR as an institutional experiment about how biomedical science itself is produced — and is testing the proposition that the reproducibility crisis is not a sociological problem about incentives but an engineering problem about infrastructure.

Whether that proposition is correct will determine whether IGoR is the most consequential biomedical funding decision of the decade or an expensive demonstration that the field's real problems do not yield to technological solutions. For the small number of teams that have a credible shot at the three awards, the strategic question is which framing to bet on.

The Architecture of the Bet

IGoR organizes its work into four Technical Areas, and the solicitation is unusually strict about insisting that every proposing team address all four. This is not a programmatic preference. It is a structural requirement. A proposal that addresses three of the four Technical Areas is functionally non-responsive.

Technical Area 1 — Mechanistic disease models. Teams must build comprehensive computational representations of complex disease processes that encode causal biological relationships rather than purely statistical correlations. The reference language inside the solicitation describes these as "digital twins" of disease, with the implication that they should support intervention modeling, not just descriptive simulation. This is the area in which the program leans most heavily on the past decade of systems biology, multi-scale modeling, and pharmacometric infrastructure work.

Technical Area 2 — The AI orchestration layer. This is the new science engine. The AI system must identify what the disease model does not yet know, prioritize among possible experiments by expected information gain, design the next experiment, hand the design off to the laboratory layer, and integrate the resulting data back into the model. Several existing autonomous-discovery platforms — including some commercial active-learning tools used in materials and chemistry — provide partial precedents, but no system has yet been built that closes the loop end to end in biomedical science.

Technical Area 3 — Layered protocol architecture. Teams must build a system for representing experimental protocols in a way that they can be executed faithfully at any qualified laboratory. This is the area in which IGoR most directly confronts the reproducibility crisis. The implicit claim is that much of biomedical irreproducibility is not fraud or bias but uncoded tacit knowledge — the gestures and judgments that experienced scientists use when running a protocol — and that those gestures can be formalized.

Technical Area 4 — The laboratory marketplace. Finally, teams must build a network of validated wet-lab facilities that will run the protocols on demand and return high-quality, standardized data. The performance bar named in the solicitation is striking: at least 90 percent inter-laboratory concordance across the standardized protocols, and automated updates to the underlying disease model within four hours of new data arriving. Both numbers represent at least an order-of-magnitude improvement over current cross-lab replication rates.

The Phases

IGoR's three-phase structure makes the timing of major commitments unusually legible.

Phase I — 18 months. Concept development and closed-loop demonstrations within each performer team. Teams must show that their disease model, AI orchestrator, protocol layer, and lab network function together end to end on a focused disease scope.

Phase II — 18 months. Cross-team integration and interoperability testing. This is the phase that separates IGoR from a conventional research consortium. Successful Phase I teams must demonstrate that their components can interoperate with components from other awardees — that protocols written by Team A can be executed by Team B's laboratory network, and that data flowing into Team B's model can be ingested by Team A's orchestrator.

Phase III — 24 months. Scaling, commercialization planning, and expansion into a second disease area. By the end of Phase III, the program expects teams to have proven that the architecture generalizes beyond its initial disease focus and that the resulting infrastructure is durable enough to be operated outside the ARPA-H investment cycle.

The phased gates create real risk for performers: each transition is a competitive merit review against the other awardees and against the milestones the team itself proposed. Teams that hit Phase I but cannot integrate during Phase II are vulnerable.

Who Should Actually Bid

The solicitation language ("Teams may include academic institutions, non-profit organizations, companies, or a combination of highly skilled performers across sectors") is technically open, but the program's structural demands narrow the field meaningfully.

A serious bid needs five capabilities in-house or under firm subcontract: deep domain expertise in a target disease area, multi-scale mechanistic modeling, frontier AI systems engineering, automation-grade laboratory operations, and the software engineering capacity to build the protocol layer as a deployable platform. Almost no single institution has all five. Realistic teams are consortia anchored by a major academic medical center, a national laboratory or large industry partner, an AI lab, and a contract research organization with proven automation.

Small businesses are eligible, but the cost structure of building all four Technical Areas simultaneously almost certainly forces a teaming arrangement. FFRDCs are excluded unless ARPA-H affirmatively determines that a unique capability is necessary. Government entities and government employees are excluded as participants. Current ARPA-H support contractors are excluded due to organizational conflict-of-interest rules. Foreign-entity restrictions tied to the CHIPS and Science Act apply.

The LLM Review Pilot — A First in Federal Biomedical Funding

Buried in the solicitation is a single sentence with outsized implications for federal grantmaking. ARPA-H states that IGoR will pilot the use of secure large language model tools to assist in the initial review of submitted Solution Summaries. The agency has separately published a statement on its use of AI tools in IGoR proposal reviews framing the practice as a way for AI to organize, summarize, and surface key information for human reviewers with subject-matter expertise.

This is the first major federal biomedical solicitation in which AI is explicitly used in the proposal-review pipeline. For applicants, three implications follow. First, Solution Summaries must be written for two audiences: the human expert reviewer and the LLM summarizer that will sit upstream of the human. Plain language, explicit linkage between technical claims and citations, and structured arguments will land better than dense prose. Second, the LLM tier will exaggerate the cost of buried-lede writing. Information not surfaced in the first two pages may not be surfaced at all. Third, ARPA-H's commitment to LLM-assisted review will become a precedent. The decisions other agencies make in 2026 and 2027 about whether to follow suit will be shaped in part by how cleanly the IGoR pilot runs. Applicants are, in effect, also pilot subjects.

What the Solicitation Tells Us About ARPA-H's Funding Philosophy

IGoR sits inside the ARPA-H Proactive Health Office and reflects a broader institutional bet that biomedical science needs infrastructure investment, not just discovery investment. The contrast with the National Institutes of Health is instructive. NIH's R01 mechanism funds individual investigator-driven hypotheses on a 3-to-5 year cycle, with no expectation that any single award reshapes the field's underlying production function. ARPA-H's IGoR is doing something categorically different: it is attempting to change the substrate on which biomedical knowledge is produced. Compare this with the NICHD FY2026 funding strategy, where the same fiscal-year environment that produced a 14 percent R01 cut at NICHD coexists with an ARPA-H willing to put nine figures of OT money behind three teams. The two agencies are operating with categorically different theories of how to spend marginal federal biomedical dollars.

The OT mechanism itself reinforces the bet. Other Transaction agreements give ARPA-H meaningfully more flexibility than traditional grants or contracts: tighter milestone control, easier off-ramps for underperforming teams, and faster pivoting on intellectual property terms. The flip side is that OT mechanisms place real management burden on performers and require sophistication that not every academic principal investigator has built.

What To Do Between Now and August 6

For teams with credible bids, the six weeks ahead are structurally pre-allocated. Solution Summaries must be five pages, English language, sans-serif font at minimum 11-point. They are the only path to a full proposal review. Teaming decisions must be locked in by Proposers' Day on June 9, and ARPA-H's dedicated teaming page is the right place to surface complementary capabilities. SAM.gov registration takes seven to ten business days and is non-waivable. Teams not already registered should treat that as a critical-path item this week.

For teams not bidding IGoR, the program is still worth tracking. The four Technical Areas — mechanistic models, AI orchestration, protocol architecture, lab marketplaces — are the same architecture that will shape biomedical funding at NIH, BARDA, and the foundations over the next five years. The vocabulary IGoR establishes will become the vocabulary the field uses, regardless of whether IGoR itself succeeds.

The bet IGoR represents is large. Three OT agreements, four Technical Areas, ten-times-faster science, ninety percent inter-laboratory concordance, a five-year horizon. The downside is that it is an expensive and visible failure if the reproducibility problem turns out to be sociological. The upside is that ARPA-H has done something the NIH has not been positioned to do in twenty years: bet a meaningful fraction of a generation's biomedical funding on a hypothesis about how the field itself works. We will know whether the bet paid off sometime around 2031. The bidding closes August 6.