ARPA-H's Autism Program Staggered Its Deadlines Across Three Months. The Order Tells You What It Wants First.

On September 17, 2026, ARPA-H launched SPECTRA — Systems for Phenotypic Evaluation, Clinical Trajectories, Response, and Agency — under solicitation ARPA-H-SOL-26-163. The program manager is Russell Shilling, Ph.D. The award instrument is Other Transaction agreements, which is standard for ARPA-H and consequential in ways covered below.

The program's premise is that autism is currently diagnosed and treated as a single category when the underlying biology, developmental trajectories, and support needs are heterogeneous — and that the tooling to distinguish those trajectories at the individual level does not exist. The framing statistics: 1 in 31 eight-year-olds in the US have been identified with autism spectrum disorder, and projected annual US autism-related costs reached roughly $461 billion.

What makes this solicitation structurally different from most federal health research calls is the deadline architecture. There is no single due date. There are four technical areas with three separate Solution Summary deadlines spread across three months, and the order is not alphabetical.

The deadline structure, and what the order signals

Full proposal deadlines are to be announced.

Note the first window: submissions open October 4 and close October 18. That is a two-week submission window opening just over two weeks after program launch. If TA 4 is your lane, the effective preparation period started the day the program was announced, not the day submissions open.

The sequencing is a dependency graph rendered as a calendar. TA 4 builds the instruments that generate real-world functional data. TA 1 and TA 2 analyze and model data. TA 3 uses the models to select and test interventions. ARPA-H is procuring the measurement layer first, the inference layer second, and the intervention layer last — which is the order in which a program of this type has to actually work.

It also means the three-month gap between the TA 4 deadline and the TA 3 deadline is not slack. It is the window in which a TA 3 proposer can watch what gets built upstream and write a proposal that connects to it.

What each technical area is actually asking for

TA 1 — Analysis. Tools to characterize the genetic, environmental, family, metabolic, immune, and developmental factors associated with distinct developmental, health, communication, and functional trajectories. The list of factor domains is unusually broad — it explicitly spans genetics, environmental exposures, epigenetics, microbiome biology, immune and metabolic measures. A proposal that covers one domain deeply is a legitimate submission, but the solicitation language is written for integration across domains, which favors teams that already have multi-modal data or credible access to it.

TA 2 — Modeling. Federated, privacy-preserving AI platforms that map trajectories and identify modifiable outcome drivers and actionable targets. The word "federated" is doing heavy lifting. This is a commitment to models that learn across institutional datasets without centralizing the data — which sidesteps the data-sharing agreements that have historically stalled autism cohort research, and simultaneously raises the engineering bar considerably. If your modeling approach requires pooling raw data in one place, it is not responsive to TA 2 as written.

Note also "modifiable outcome drivers." A model that predicts outcomes well but identifies nothing actionable is explicitly not the deliverable. The target is causal-adjacent and intervention-relevant.

TA 3 — Precision Intervention. Individualized selection and rigorous testing of interventions to improve outcomes across the lifespan. Two things in that sentence matter. "Rigorous testing" means this TA carries clinical trial obligations, cost, and timeline. "Across the lifespan" is a deliberate rejection of the field's heavy concentration on early childhood — adult autism outcomes are chronically under-researched, and a proposal weighted toward adolescence and adulthood is reading the solicitation correctly.

TA 4 — Enabling Technologies. Accessible tools for communication, quality of life, and embedded assessment that measure real-world function as a byproduct of ordinary use. That last clause is the whole technical requirement. Not a clinic visit. Not a structured assessment battery. Instrumentation that produces valid functional measurement while a person is simply living — using a communication device, navigating a day, doing a job. This is the hardest engineering problem in the solicitation and it has the earliest deadline.

The Solution Summary process is a filter, not a formality

ARPA-H does not run a single-submission competition. The process is:

1. Submit a Solution Summary — a short abstract giving a synopsis of the proposed approach from both a technical and budgetary perspective — through the ARPA-H Solutions site at solutions.arpa-h.gov.
2. Receive a feedback letter: Encourage or Discourage.
3. Submit a full Proposal if encouraged.

Three practical consequences.

First, the budget is in the Solution Summary. This is not an abstract that describes the science and defers cost to the full proposal. Proposers who submit a compelling technical concept with a hand-waved budget are telling ARPA-H they have not scoped the work.

Second, a Discourage is not a rejection you appeal — it is information you use. ARPA-H's stated intent for the Solution Summary stage is to provide guidance on strategies within program scope that are likely to achieve program goals. Teams that treat a Discourage as a data point about scope, and re-aim at a different TA or a different framing, are using the mechanism as designed.

Third, this is where the staggered deadlines become strategically useful. A team with breadth across TA 4 and TA 1/TA 2 can submit to TA 4 in October, receive feedback, and apply what that feedback reveals about how the program office is reading scope to a December TA 1/TA 2 submission. That is a genuine advantage of the sequencing, and it only exists for teams that move on the earliest deadline.

Other Transactions change who can compete

ARPA-H awards are usually Other Transactions — not procurement contracts, not grants, not cooperative agreements. OTs permit flexible contracting and negotiated terms, including on intellectual property, milestone structure, and consortium composition.

This matters for eligibility in a way that filters differently than NIH:

• Companies are first-class participants, not subawardees on a university prime. A device company, an AI platform company, or an assistive-communication technology firm can lead.
• Consortia are the norm. SPECTRA's four TAs are written for teams that combine clinical cohorts, computational capability, and device engineering. Very few single institutions hold all three.
• The negotiation is real. Terms are negotiated rather than accepted off a standard form. Organizations without OT experience should be talking to their contracts office before the Solution Summary, not after an Encourage letter arrives.
• NIH-style habits do not transfer cleanly. Specific Aims pages, biosketch-driven credibility, and study section framing are the wrong instincts here. ARPA-H buys milestones and go/no-go decision points.

ARPA-H has run this playbook across IGOR, ADVOCATE, and PROSPR, among others — the mechanics are consistent even as the science changes.

How SPECTRA relates to what NIH is already funding

SPECTRA is not landing on empty ground. NIH launched the Autism Data Science Initiative (ADSI) and awarded approximately $51 million through 13 Other Transaction agreements — an unusual instrument for NIH, and itself the subject of an HHS Office of Inspector General review that found the awards were made in accordance with federal requirements. ADSI's strategic aims are creating integrated data resources from existing data, filling gaps through targeted data generation, and supporting analysis linking genetic and non-genetic factors.

The overlap with SPECTRA's TA 1 is obvious, and that is the thing to address directly rather than avoid. ADSI is largely about assembling and analyzing data that already exists. SPECTRA's TA 4 is about generating a class of data that does not currently exist — continuous real-world functional measurement — and TA 2 is about modeling it in a federated architecture. The complementarity is real, and a proposal that names an ADSI-funded resource it will build on, or a specific analytic gap ADSI leaves open, demonstrates field awareness that a generic proposal cannot.

Where this also connects is the broader HHS posture on autism research, which has been politically visible and well-funded relative to its historical baseline. A program office operating in that environment has budget and attention, and also scrutiny.

The community-input requirement is not decorative

ARPA-H's description emphasizes incorporating input from the autism community and moving beyond standardized, one-size-fits-all care models. The word "Agency" in the program acronym — Systems for Phenotypic Evaluation, Clinical Trajectories, Response, and Agency — is a deliberate choice, and "agency" in this context means autistic people's capacity to direct their own care and lives.

TA 4's focus on communication tools and quality of life reinforces this. A program that wanted only biomarkers and diagnostics would not have a technical area dedicated to accessible communication and real-world quality-of-life measurement.

For proposers, this is a scoring reality, not a values statement. Autism research has a long and documented history of conflict between researcher priorities and community priorities, particularly around interventions framed as normalization. A proposal that describes autistic people exclusively as subjects, that defines success entirely as reduction of autistic traits, or that includes no autistic people in its design and governance structure is misaligned with the way this program is written. Teams should have named community partners and a described role for them in the Solution Summary, in October, not added at full-proposal stage.

What to do in the next two weeks

If TA 4 is your area, you have until October 18. Priorities, in order:

1. Pull the full solicitation from SAM.gov. The program page carries the technical area descriptions and deadlines; the solicitation carries the evaluation criteria, submission requirements, and any funding guidance. Do not write from the program page alone.
2. Answer the byproduct question in one sentence. TA 4's requirement is measurement of real-world function as a byproduct of ordinary use. If you cannot state what ordinary activity generates your measurement and why the measurement is valid, you do not yet have a TA 4 concept.
3. Scope the budget now. The Solution Summary requires a budgetary synopsis. A defensible number with a milestone structure behind it beats a range.
4. Name your community partners. Organizations, advisory structure, and their decision rights — briefly, but specifically.
5. Start the OT conversation internally. If your institution has not executed an Other Transaction, the contracts office needs to know now.

If TA 1 or TA 2 is your area, you have until December 2, with submissions opening November 18. Use the interval to lock down data access agreements — TA 2's federated requirement means your architecture story depends on which institutional datasets you can credibly reach. And watch what gets encouraged in TA 4; the measurement modality that survives that round is the data your model will eventually consume.

If TA 3 is your area, you have until January 18, 2027, and the longest runway. Spend it on the two things that sink intervention proposals: a credible plan for rigorous testing at a realistic timeline, and genuine lifespan coverage rather than an early-childhood study with an adult paragraph appended.

The staggered structure is a gift to teams that read it as a sequence and a trap for teams that see a program launch in September and assume there is a deadline in the spring. The first one is five weeks out.