Correctness for Scientific Computing Systems (NSF/DOE CS2): NSF 24-571 (2026 Deadline 11 August)

Correctness for Scientific Computing Systems (NSF/DOE CS2): NSF 24-571 (2026 Deadline 11 August)

The Correctness for Scientific Computing Systems (CS2) program is a joint National Science Foundation (NSF) and U.S. Department of Energy (DOE) initiative focused on one hard problem: scientific computing is increasingly central to research, yet critical results can be invalid because software and system behavior is wrong in subtle but expensive ways. NSF’s program page describes CS2 as a mechanism to make correctness a first-class goal alongside performance for scientific tools and workflows. This opportunity is particularly valuable for teams blending scientific computing with formal methods, and it is structured as a standard proposal route with a recurring full proposal deadline pattern.

The active opportunity page currently lists a full proposal deadline of 11 August 2026 and documents the program as an ongoing cycle that has supported awards across FY2025, FY2026, and FY2027. The annual profile is small and selective by design: about five awards per year are anticipated, with up to $800,000 per award and up to four years of duration.

Key details at a glance

What CS2 is (and is not)

CS2 is not a broad “software improvement” grant and not a general infrastructure fund. The page explicitly frames it as a program where correctness of scientific software and systems is the object of funding, from low-level libraries to complex multiphysics workflows. The problem statement is explicit: numerical errors, race conditions, memory faults, specification mismatches, and runtime instability can invalidate outcomes. CS2 exists to fund rigorous projects that increase confidence, not just speed.

This shapes the program in practical terms:

• Research is expected to be technical, with clear hypotheses and evaluation logic around correctness.
• Interdisciplinary pairing is expected, with at least one PI/co-PI on scientific computing and at least one on formal reasoning.
• It is not just code writing. Teams must show how their methods change confidence in correctness in production-like or research-relevant conditions.
• It is not a quick pilot grant; projects may run up to four years and are reviewed for scientific merit and implementation rigor.

The solicitation language also frames program-level behavior around mission alignment: this is a foundational reliability problem, not a grant for commercial launch strategy or non-research service delivery.

Who this is for and who to avoid applying now

Strong fit candidates

This program is strongest for teams that can demonstrate the following:

1. Cross-disciplinary core capacity: one side of the team can speak computational science deeply, while another side can deliver formal analysis and mechanized correctness reasoning.
2. A clearly articulated correctness gap: they can specify what can be wrong today and why current approaches fail.
3. A defensible project architecture: measurable claims with milestones, validation data, and clear integration points across software/hardware layers.
4. Institutional readiness: PI and co-PI appointment status, and support systems for federal submission.

Typical good-fit organizations include research universities, national labs, and mission-aligned non-profit research organizations in the U.S.

Candidates that usually fail

You should pause and reassess if your team is mostly:

• Only one disciplinary track (e.g., pure numerical methods without formal reasoning, or pure formal methods without embedded scientific workloads).
• Inexperienced with federal proposal workflows and compliance.
• In a position where PI is not institutionally eligible by NSF and the program’s stated PI criteria.

CS2 enforces PI constraints and participation limits across proposals and awards. Even high-quality science can be returned if rules are not met.

Eligibility in detail: not just who, but how many

From the solicitation, eligibility includes:

• Who may submit proposals:
  • Certain non-profit non-academic research organizations in the U.S.
  • U.S. institutions of higher education, including two- and four-year accredited institutions (including community colleges) acting on behalf of their faculty
  • DOE National Laboratories
• PI/co-PI eligibility:
  • Must be DOE national lab employee, or
  • hold a tenured/tenure-track position, or
  • hold a primary full-time paid appointment in research or teaching at an eligible U.S. organization

That means this is a fundamentally U.S.-submission program; cross-border institutions or overseas branch campuses do not fit the default PI path unless explicitly permitted by the solicitation’s exceptions.

Participation limits that materially affect strategy

The program makes one of its strictest constraints explicit:

• max two proposals per PI/co-PI per deadline,
• max two awards across the life of NSF 24-571 (FY2025–FY2027).

If an individual exceeds limits, NSF accepts proposals based on earliest submission time, and extra submissions are returned without review. This can create a strategic race effect near deadline, so early readiness matters as much as final quality.

Why this matters for planning

Most applicants underestimate this: even if you have excellent project ideas, a strong proposal can still be blocked by PI-level participation limits. Teams often solve this too late by trying to split ideas into multiple proposals or asking another senior person to “front” a duplicate effort. In this program, duplicate/conflicting applications can raise process risk and weaken perceived governance.

What funding you can count on and when

The program page and linked solicitation provide the most defensible numbers:

• Estimated number of awards: ~5 per year
• Anticipated award amount: up to $800,000 per award
• Duration: up to 4 years
• Program annual funding: $3M NSF + $3M DOE

A key line in both the page and the solicitation confirms the recurring cadence and annual budget framing, including “approximately 5 awards in FY2025, FY2026, and FY2027.” For teams with 2026 timing, this is valuable because it makes near-term planning realistic: submissions are competitive and recurring, but not unlimited.

Practical interpretation of the cap structure

Even with a five-award cycle, competition is high. Your practical target is not just “fit the topic” but “fit the delivery model”:

• Show why correctness is core to your question, not peripheral.
• Show that your interdisciplinary approach is structurally necessary.
• Show how outcomes are verifiable through objective evidence.
• Show why this project belongs to this program specifically (not to adjacent NSF channels).

Submission workflow and application mechanics

The solicitation gives a plain but strict route:

1. Prepare full proposal materials according to NSF PAPPG requirements (or the NSF Grants.gov application guide depending on route).
2. Check for collaborative-proposal routing rules: collaborative proposals are required through Research.gov according to the solicitation’s instructions.
3. Respect the 2026 full proposal due date:
   • 11 August 2026, 5:00 pm local submitting organization time
4. Avoid last-minute submission because of required pre-check and post-check validation.

From the solicitation and NSF page:

• Submissions may be made through the standard NSF systems.
• NSF recommends submitting several business days early.
• There are no Letters of Intent or preliminary proposal submission requirements in this cycle.
• No voluntary cost sharing is required.
• Indirect cost is listed as not applicable in the solicitation’s program budget block.

Application artifacts you should prepare first (time saver)

You should build these in parallel before writing the full proposal:

• PI stack map: which PI/co-PI covers scientific computing and which covers formal methods.
• Problem decomposition: where correctness failures occur in your target workflow.
• Test and validation strategy: inputs, conditions, expected failures, evaluation signatures.
• Data management and reproducibility plan: clear enough for review scrutiny.
• A collaboration plan: if co-PI expertise is split across institutions, make transfer workflow explicit.

What not to do in late weeks

• Don’t wait until final week to secure compliance with PI role and eligibility constraints.
• Don’t submit a technically strong but institutionally weak application with unclear appointment status.
• Don’t assume one person can represent a portfolio of unrelated proposals without hitting participation limits.

Review logic: how to maximize odds under this solicitation

CS2 proposals are evaluated under standard NSF merit review criteria (Intellectual Merit and Broader Impacts), plus program-specific criteria. The program-specific criteria include:

• contributions at the intersection of scientific computing and formal methods,
• fit to the program’s core problem statement,
• strength of the collaboration plan.

Build your review narrative around these three layers

Layer 1: Intellectual Merit

Explicitly state the correctness gap and why your technical strategy changes it. This includes:

• failure modes you target,
• why existing tools or methods are insufficient,
• proof or evidence strategy (theoretical and experimental).

Layer 2: Broader Impacts

NSF’s broader impacts question is not optional and does not require commercialization claims. For CS2 it is often strongest when you explain:

• improved confidence in downstream scientific conclusions,
• reduction of costly false leads,
• stronger reliability for research communities.

Layer 3: Collaboration quality

The solicitation explicitly evaluates collaboration. Show this clearly by defining division of labor, communication cadence, and integration gates.

Timing and review realism

The solicitation says NSF expects review and recommendation action within roughly six months after submission and note that complex proposals may need more time. For teams planning 2026 publication/award cycles, this means post-deadline planning should include: if awarded, do you have immediate kickoff readiness.

Common mistakes in CS2 applications and how to avoid them

This section reflects recurring reviewer traps in correctness-style programs.

1) Under-specifying the correctness target

Many projects say “improve reliability” without formalizing what is being proved or measured. If there is no precise target, reviewers cannot assess risk and novelty. Use failure classes and metrics.

2) Cross-disciplinary imbalance

Teams often over-rely on one side of the expertise pair. The solicitation itself requires breadth: at least one co-PI in scientific computing and at least one in formal reasoning. Do not make the second area a token mention.

3) Ignoring PI cap mechanics

Submission timing can make the difference between review and non-review. This is especially critical for teams where senior people sit on many related grants.

4) Overcrowded proposal design

This program rewards conceptual clarity. “Everything for everyone” proposals underperform because they do not optimize for proof obligations and evidence path.

5) No distinct collaboration operations

If collaboration is your only advantage, it must be operational, not descriptive. Define decision points, data ownership, milestone sequencing, and integration points.

Preparation checklist for a CS2-ready application

Use this as a practical pre-submission list:

1. Validate PI and co-PI eligibility against program rules.
2. Map each work package to a correctness objective.
3. Ensure at least one co-PI has formal method depth and one has scientific computing depth.
4. Choose whether this route is Research.gov or Grants.gov based on collaboration structure.
5. Confirm your institution’s AOR and internal approvals before the final week.
6. Draft a detailed review strategy section that explicitly addresses:
   • Intellectual Merit,
   • Broader Impacts,
   • Collaboration quality.
7. Include a data sharing and validation section that shows how evidence will be produced and interpreted.
8. Leave the final 5 business days for validation rather than writing.

Common questions (directly mapped to official requirements)

Is this only for very large national institutions?

No. The submission base includes universities, non-profit research organizations, and DOE labs. What matters is programmatic fit and qualified PI structure, not institution size.

Is there one-off funding or can this support a long project?

The anticipated duration is up to four years, with awards up to $800,000. It is not a one-off workshop grant.

Is this open now?

As of the checked snapshot on 2026-06-01, the opportunity is listed with an upcoming 2026 deadline and recurring program cadence. Confirm your version on the official page before you submit.

Does this have strict PI limit rules?

Yes. The solicitation lists clear limits and even says proposals exceeding limits can be returned based on submission order.

Can I submit a proposal that is mostly theoretical?

You can, if correctness is central and the proposal remains tied to scientific computing systems with clear measurable outcomes. Purely abstract work without system-level relevance is a weak fit.

2027 planning angle

If your team misses 2026, there is no reason to discard work. The program documents a recurring cycle with FY2027 activity expected in its model. Teams should preserve core materials, then revise based on feedback or peer review from the 2026 submission and resubmit with a stronger collaboration and evidence narrative.

For 2027 preparation specifically:

• keep your test infrastructure portable,
• retain all reproducibility artifacts,
• include reviewer feedback logs,
• pre-document any partnership updates.

Why this matters in the larger research funding landscape

For software-heavy science teams, CS2 is unusually explicit about combining formal proofs with computational relevance. Many similar calls are too broad or too broad-spectrum for rigorous correctness work. This one does not ask for polished software for the sake of itself; it asks for fundamental risk reduction in the systems that produce scientific conclusions.

If your project solves this exact seam—where modern compute is powerful but not trusted enough to be policy, engineering, or medical decisions—you likely have a better match here than in broader infrastructure or purely AI-oriented calls.

Official links and next steps

• Program page: https://www.nsf.gov/funding/opportunities/cs2-correctness-scientific-computing-systems
• Active solicitation: https://www.nsf.gov/funding/opportunities/cs2-correctness-scientific-computing-systems/nsf24-571/solicitation
• For NSF Proposal and Award policies: https://www.nsf.gov/publications/pub_summ.jsp?ods_key=pappg

Before you click submit, re-check the call page for any updates to program contacts and due-date notices. For 2026/2027 planning, the value in CS2 is less about a single application and more about disciplined interdisciplinary project architecture that the review process can verify.