ARPA-H's SURPASS Will Not Let You Bid One Piece of It. That Requirement Is the Program.

The headline number in SURPASS is a factor-of-three claim: take clinical development of a drug or biologic from "more than a decade to less than four years." ARPA-H programs make claims like that routinely, and the right response is usually to read the milestones rather than the press release.

In this case the milestones are more interesting than the claim, and one structural rule in the solicitation does more to determine who can win than any technical requirement in it.

Simulation-augmented, Real-time Platform Adaptive Seamless Trials (SURPASS), solicitation ARPA-H-SOL-26-164, is managed by program manager Daria Fedyukina, Ph.D. The schedule runs: informational webinar October 15, 2026 (registration closed October 12), questions due November 2, Proposers' Day November 6 with registration by October 28, Solution Summary due November 30, 2026, and a pitch package due January 22, 2027 with virtual pitches expected in early February. Awards will be Other Transaction agreements, multiple anticipated, over a five-year period of performance split into a 24-month Stage I and a 36-month Stage II.

The no-unbundling rule

The solicitation defines three technical areas:

Technical Area 1 — Phaseless Design Engine. A simulation-based environment for designing adaptive, seamless platform trials, with integrated planning across first-in-patient, efficacy, and where proposed confirmatory development, instead of optimizing conventional Phase I/II/III separately. It must incorporate digital twins, synthetic control arms, disease progression models, PK/PD models, and predictive safety models, with reproducible workflows and tooling usable by trialists who are not software developers.

Technical Area 2 — Continuous Inference. An inference and adaptation validity engine supporting always-valid, real-time or on-demand analysis as data accumulate, through "authorized, reproducible, and auditable procedures." It must support adding and removing treatment arms, seamless transitions between development stages, and integration of prognostic scores and digital-twin outputs — all without compromising inferential validity.

Technical Area 3 — Agentic Operations. An AI-enabled operations layer automating master protocol documentation, site activation, regulatory workflows, participant prescreening, data management, safety reporting, and the mechanical implementation of trial adaptations.

Then it closes the door: teams must address all three technical areas as an integrated platform. Individual technical area bids are not permitted.

That single rule is the program's real design decision, and it is worth understanding why ARPA-H made it.

Each of the three technical areas corresponds to an existing, mature commercial sector. TA1 is the world of trial-simulation and model-informed drug development vendors. TA2 is the biostatistics and adaptive-design methodology community. TA3 is clinical trial operations software — eCOA, eTMF, CTMS, site activation platforms. Each of those sectors has credible players who could write a strong proposal for their slice and almost nobody who could write a strong one for the other two.

By forbidding unbundling, ARPA-H has declared that the bottleneck is not any of the three capabilities. It is the integration between them. A simulation engine whose outputs the inference layer cannot legally use for a decision is worthless. An inference engine that signals a trial adaptation the operations layer takes eleven weeks to implement has not shortened anything. The agency is funding the seams.

The practical consequence for bidders: this is a consortium solicitation that does not say the word consortium. No single company in the clinical research ecosystem spans all three areas at the required depth. Expect winning teams to look like a trial-simulation firm, an academic or industrial biostatistics group, a clinical operations technology provider, an academic health system network, and at least two pharmaceutical or biotech sponsors — assembled, with teaming agreements and data-rights terms negotiated, in the eight weeks between the solicitation's publication and the November 30 solution summary.

Eight weeks is not enough time to form that consortium from a cold start. It is enough time to formalize relationships that already exist. If you are reading this and do not already know your TA2 and TA3 partners, the realistic move is to attend Proposers' Day on November 6 to join someone else's team rather than to lead one.

The milestones are the hard part

ARPA-H programs are judged on milestone commitments, and SURPASS's are unusually concrete and unusually demanding.

Stage I, 24 months:

• Develop all three technical areas
• Onboard a minimum of three intervention arms
• Open the first clinical site by the end of Year 2
• Establish a multi-site network, with a minimum of six sites by program end
• Secure FDA authorization for a Master Protocol IND and at least one investigational arm

Stage II, 36 months:

• Prospectively validate the integrated capabilities
• Implement and validate live trial adaptations
• Progress, terminate, or graduate arms under an approved decision framework

Design requirements that bind the proposal itself:

• A minimum of three drug or biologic arms proposed
• At least two must be unapproved interventions, and one must have no prior human data
• A minimum of two for-profit sponsors, at least one of which is a startup or small company
• A minimum of six clinical sites by program conclusion

Read the third bullet in that last group carefully. SURPASS is not a methods-development program with a simulated validation. It requires teams to put a molecule with zero human data into a first-in-patient arm inside a master protocol whose adaptation rules are driven by a simulation engine being built concurrently. The regulatory posture required to do that is the single highest-risk element of the whole program, and it has a hard gate on it: an FDA-authorized Master Protocol IND within 24 months.

The sponsor requirements are a second, subtler screen. Two for-profit sponsors, one of them small, means a team must have secured actual pharmaceutical assets — real molecules, with real sponsors who have agreed to let them be tested inside someone else's adaptive platform. Pharmaceutical companies do not casually assign unapproved assets to externally controlled master protocols. That negotiation is harder and slower than any software deliverable in the solicitation, and it has to be substantially complete before a credible solution summary can be written.

The small-company requirement is the most interesting inclusion. A startup with a single asset and no internal trial infrastructure has far more to gain from a shared platform with a common control group than a large pharma does — and far less bargaining leverage to refuse. ARPA-H appears to have written the sponsor rule to guarantee at least one participant whose incentives align with the platform rather than against it.

Where SURPASS sits in ARPA-H's stack

SURPASS is not a standalone bet. It arrives on top of a set of ARPA-H investments in trial infrastructure that it depends on: STACK, on AI-driven clinical site activation; COMMONS, a privacy-preserving national consent architecture; and CINCH, on real-world data collection and care navigation.

That matters for proposal writing. A team that treats site activation and consent as problems it must solve from scratch is proposing to rebuild capabilities ARPA-H is already funding elsewhere. A team that proposes to consume those capabilities — with explicit interfaces, assumptions, and fallbacks if those programs slip — is reading the agency's portfolio correctly. The agentic operations layer in TA3 should be designed as an orchestration layer over existing infrastructure, not as a replacement for it.

The regulatory environment moved in the same direction in September. The FDA launched an Expedited Investigational New Drug Pilot on September 15, 2026, aimed at accelerating first-in-human development. SURPASS's Stage I gate — a Master Protocol IND with at least one authorized investigational arm inside 24 months — is substantially more plausible against that backdrop than it would have been a year ago. A proposal that explicitly maps its regulatory strategy onto the current pilot landscape is answering a question reviewers will otherwise ask.

Eligibility and mechanics

Standard ARPA-H other-transaction terms apply. Respondents must be registered in SAM with a Unique Entity Identifier. All responsible sources may submit. FFRDCs and government entities are generally prohibited from acting as prime — which means a national laboratory or a federal institute with relevant modeling expertise participates as a team member, not as the lead, and the prime must be a company, university, or nonprofit capable of holding an OT agreement.

The Solution Summary due November 30 is required, not optional. This is a different posture than many federal programs where a pre-proposal is encouraged. Under ARPA-H's process, the solution summary is the screening gate: teams that are not invited forward do not submit a pitch package. The eight weeks between publication and November 30 are the entire competition for most bidders.

The honest assessment

SURPASS's premise is correct. Drug development takes more than a decade, can cost up to $2 billion, and fails more than 90% of the time — and a large share of that cost is structural rather than scientific: sequential phases that re-recruit from scratch, control arms rebuilt for every trial, site activation measured in months, and statistical designs that must be frozen before the first patient enrolls because the inferential machinery cannot handle mid-course change.

Platform trials with master protocols have already demonstrated that much of this waste is avoidable. What they have not demonstrated is that the design, inference, and operations layers can be fused into something reusable across sponsors and therapeutic areas, under a regulatory framework that will accept decisions made partly on simulated evidence.

That is the actual bet. The factor-of-three timeline claim is downstream of it, and it is the claim most likely to be revised. The milestone structure — IND authorization by month 24, six sites, three arms, one with no human data — is where the program will be judged, and it is where a proposal should concentrate its risk disclosure rather than its optimism. Teams that write to the gates will read as serious. Teams that write to the headline will not.