Hi Court, Regarding MOAT: For the March deadline, there is only one track (A), hence either remove (Track A/B/C) or change to (Track A). On the other hand, for the June deadline, as communicated in the MOAT 1-slide summary here <https://urldefense.us/v3/__https://www.dropbox.com/scl/fi/gptvjw7ae69qrbrcc4... > and in the detailed “punch list” here <https://urldefense.us/v3/__https://www.dropbox.com/scl/fi/29xe8ye2ef2ed663ym... >, we have three tracks: Track A: Expand agentic workflows (connect logbooks, optimizers, facility-specific reasoning) Track B: Surrogate models of particle accelerator facilities Track C: (if resources allow): Conceptualizing self-improving accelerator lattice design reasoning model Note that as indicated, Track C is conditional on funding and bandwidth So we propose to reformulate the June milestone to something like, e.g., “succeeded in 2 of 3 tracks (Tracks A/B/C)” Dependencies: As indicated at the all-hands meeting in our extended demo documents (March demo <https://urldefense.us/v3/__https://www.dropbox.com/scl/fi/1gzj67hl7kyzgbma6j... >, June demo <https://urldefense.us/v3/__https://www.dropbox.com/scl/fi/qiha4jcz3pmcyurz37... >), the demos depend (for collaborators at some of the participating facilities) on access to frontier commercial models (e.g., Anthropic) at each facility, and administrative changes at laboratories to allow for the use of commercial models and full support for the deployment of self-hosted models. Additional context: DOE (via Genesis) is pushing for transformative adoption of AI, but approvals to use DOE and lab-owned data in essential tasks (such as API calls to non-locally-hosted foundation models) are moving slowly through the laboratory site offices and the standard Authorization to Test/Use processes. Accelerating this process is essential for achieving high-impact outcomes for Genesis. It is essential that all labs have broad and robust access to the latest foundation models. This is an immediate need for the seed effort but extends to future MOAT activities as well. There is significant variation in foundation model access among the various labs. Without access or data permissions, some groups must use sub-par local models (e.g. gpt-oss-120b), which are inadequate for a high-performance agentic system. This represents a serious bottleneck and directly impacts our ability to meet our stated deliverables. Genesis could aid this process by communicating the importance of this access to DOE and national laboratory management. Best, The MOAT POCs

From: "Corley, Courtney D via Modcon-mt" <[email protected]> Subject: [Modcon-mt] [Request for Model Teams] Review MT milestones set at Dec Hackathon Date: January 7, 2026 at 7:29:57 PM PST To: "[email protected]" <[email protected]> Cc: "[email protected]" <[email protected]> Reply-To: "Corley, Courtney D" <[email protected]>

This Message Is From an External Sender This message came from outside your organization. Hi Model Teams and Model Team POCs, We have compiled the current set of ModCon and seedling model team milestones identified to date, including the March and June milestones and activities communicated during the December hackathon closeout. These milestones are used to track progress with ModCon and ASCR leadership, and we will be requesting updates on progress toward these outcomes on a regular cadence. Please review the attached milestones for your model team and coordinate with your Model Team POC to provide any updates to the March and June milestones. Updates may include, but are not limited to: Dependencies on cross-cutting team capabilities American Science Cloud, infrastructure, or other external dependencies Revisions or refinements to milestone scope Additional milestones, deliverables, or changes in planned outcomes If no changes are required, please confirm that your milestones remain accurate as written with Emily, Tom, and I. Please provide updates (or confirmation of no change) by Friday at 5:00 PM Pacific (8:00 PM Eastern). Thank you for your attention to this, and for helping ensure we maintain alignment and visibility across ModCon and seedling model team progress. Best regards, Court _________________________________ <> Court Corley, PhD Deputy Director, ModCon Director, Center for AI @PNNL Chief Scientist for AI Pacific Northwest National Laboratory (509) 375-2326 | [email protected] <mailto:[email protected]>



-- Modcon-mt mailing list [email protected] <mailto:[email protected]> https://lists.cels.anl.gov/mailman/listinfo/modcon-mt