ETA_Program_Announcement_DOI-ETA-FY26-30_-_Amendment_0002_-_Q_A_20-73_0002.pdf

EMERGING TECHNOLOGIES ACCELERATOR (ETA) PROGRAM

AMENDMENT 0002

Q&A

Note: Please read the ETA Program Announcement in its entirety prior to submitting Questions.

20. Does the government anticipate publication restrictions on any of the IARPA ETA topics?

Answer: Reference ETA Program Announcement Section 6.B.5

21. MOVES: When will the program datasets be made available?

Answer: We anticipate early in the Program.

22. MOVES: What is the anticipated length per volunteer of total video from that volunteer?

Answer: To be determined; however, we anticipate it to be in the minutes range.

23. MOVES: Will the data be completely anonymized?

Answer: Videos will be unaltered; no biographical information will be provided.

24. COSMIC: Would an edge-orchestrated or distributed software architecture be considered responsive if the delivered prototype is fully containerized and executable in the Government’s T&E environment?

Answer: Either an edge-orchestrated or distributed software solution is acceptable.

However, the Government’s goal is to obtain methodology prototypes.

25. COSMIC: The program notes that development of novel change detection algorithms is out of scope. Would a proposal be considered responsive if it focuses on pseudo-persistent data construction, multi-source harmonization, temporal modeling, confidence scoring, and agent-queryable interfaces, while using existing Government-identified or open change detection algorithms for evaluation?

Answer: Currently, the Government is not expecting proposals. A white paper submission addressing an approach as described above is acceptable.

26. COSMIC: Can onboard, near-source, or bandwidth-aware preprocessing be included as part of a COSMIC methodology if no new sensor system, hardware platform, or collection platform is being developed?

Answer: No.

27. COSMIC: For commercial remote sensing data used during development, are performers expected to deliver the underlying commercial imagery to the Government, or is it acceptable to deliver derived datasets, embeddings, metadata, model artifacts, and documentation where commercial licensing restricts redistribution of raw imagery?

Answer: The Government will consider deliverables described above on a case-by-case basis.

28. COSMIC: For non-image GEOINT inputs such as weather, traffic, or building permit data, are publicly available and commercially licensed sources acceptable if the performer provides documentation of lawful acquisition and usage rights?

Answer: Yes, provided that data from these sources can be distributed to the Government for Government purposes.

29. Our organization develops methods for quantifying and controlling uncertainty in neural network models. Though predictive accuracy is mentioned in many of the program descriptions, none of them specify methods to assess fragility or predict model uncertainty. This could apply to any trained neural network.

A. Is this work in scope for any/some of the programs? Our current work is on CV backbones so we should prioritize the programs containing CV elements: MOVES, COSMIC, ARCADE.

Answer: We encourage you to carefully review the ETA Program Announcement, submit a White Paper and explain your approach/solution.

B. What is the best way to apply? Should we submit a proposal and be placed as part of a team, or should we attempt to find a team first? Should we forgo applying if we do not find a team?

Answer: Read the ETA Program Announcement in its entirety:

https://sam.gov/workspace/contract/opp/d79c75d19ea649d3ba118f3a716c9113/view

C. Are there additional IARPA programs you are aware of that may be a better fit for the type of work we do?

Answer: Reference SAM.Gov for additional opportunities.

https://sam.gov/workspace/contract/opp/d79c75d19ea649d3ba118f3a716c9113/view

30. ARCADE: We believe the two 6-month long Sprints mentioned in Arcade are not needed for us - we believe going through a 12-month long program like this would be unproductive for us and the government.

Answer: We encourage you to submit a White Paper and explain your approach/solution.

31. For the “Cost and Schedule” section of the White Paper, the instructions state, “Provide a high-level cost estimate/Rough Order of Magnitude, inclusive of Other Team Members, over the proposed timeline of the project by Federal Government FY.” However, there is no start date for writers to use to determine the fiscal years. Can you advise?

Answer: Assume FY27 (1 October 2026) performance start date. If Proposers have already submitted a White Paper, do not cancel it or pull it back. The goal is to find out approximately how much your technical solution/approach is going to cost.

32. LOCUS: I have a clarification question regarding: “Methods that utilize biometric identification methods (e.g. facial recognition, voice recognition).”

Does “Voice recognition” mean speaker identification or voice biometrics?

Answer: Both.

33. LOCUS: I have other questions regarding the technologies that can be used:

A. Are Automatic Speech Recognition, spoken language ID, Dialect ID allowed for LocUS?

Answer: Yes.

B. Are metadata from the video and audio clips allowed to be used?

Answer: Yes, but metadata may not be available or reliable in many cases.

34. LOCUS: In Section 6 (“Government Furnished Equipment/Information”), it is mentioned that example reference datasets will be provided to ensure that performer systems can process expected file types. However, I was unable to locate any link or access details for these sample datasets.

Could you please share the location or access instructions for the example datasets, if they are currently available? Also, is there a deadline to submit the proposal?

Answer: Example datasets are not currently available. Additional information will be provided at program kick-off. The window to submit White Papers is guaranteed to stay open until 27 May 2026.

35. LOCUS: Test queries, will systems receive a defined search area or geographic prior, such as neighborhood, city, country, region, or continent, or must systems infer the search space entirely from the multimedia content?

Answer: The most performant systems would infer the search entirely from the multimedia content; however, priors may be provided but are not guaranteed. Additional information will be provided at program kick-off.

36. LOCUS: What type, amount, and timing of LocUS reference data should performers expect before each test event? Will the reference data include images, videos, audio, maps, indoor, outdoor, or venue-specific material?

Answer: Additional information will be provided at program kick-off.

37. LOCUS states that existing data sources may be used if they can be made available to the Government for evaluation purposes. For licensed geotagged multimedia datasets, what level of Government access is required: raw data delivery, API access, model index access, or another mechanism?

Answer: Access that will allow the Government to achieve its goals including transition.

Also, reference ETA Program Announcement.

38. LOCUS: Are single-modality modules acceptable as subcomponents of a single integrated multimodal LocUS system, provided the final submitted system uses all available modalities? Does it work exclusively as an audio-only or image-only solution?

Answer: Yes. LOCUS systems may be applied to audio-only or image-only problems.

39. LOCUS: Will test events include queries with missing, degraded, or unusable modalities—for example, video with poor audio, audio with no useful speech, blurred imagery, or very short clips—and will performance be scored separately by modality availability?

Answer: Yes. Specific scoring details will be discussed after kick-off.

40. LOCUS: Will indoor LocUS queries include private or commercial spaces such as hotels, malls, offices, transit facilities, or residential interiors? If so, what assumptions should proposers make about reference-data availability for such spaces?

Answer: Possibly. Additional information will be provided at program kick-off.

41. DECIPHER: Is the evaluation limited to multilingual text collections, or will speech/audio transcription be included in the evaluation?

Answer: Transcription may be included.

42. DECIPHER: Should proposers focus primarily on high- and medium-resource languages, as described in Attachment 1, or should systems also be designed for low-resource languages and non-Latin scripts?

Answer: Some high- and medium-resource languages use non-Latin scripts. Proposers should focus primarily on high- and medium-resource languages; however, applicability to low-resource languages is desirable.

43. DECIPHER: At test time, will performer systems receive any domain descriptions, communication-channel metadata, temporal metadata, or other context to guide specialized-language detection and definition generation?

Answer: Additional information will be provided at program kick-off.

44. DECIPHER: May performers use public web, forum, social-media, technical-literature, or messaging-style corpora for development if the proposer documents compliance with U.S. law, copyright, terms of service, EULAs, and privacy requirements?

Answer: Yes.

45. DECIPHER: For definition generation, will the Government provide a preferred format or schema for English-language definitions, glosses, confidence scores, contextual annotations, and glossary outputs?

Answer: Additional information will be provided at program kick-off.

46. DECIPHER: Will T&E assess semantic drift over time as a separate metric, or will drift detection be evaluated as part of specialized-term detection and definition quality?

Answer: T&E approach for semantic drift measurement will be discussed during the early stages of the program.

47. COSMIC: What modalities, approximate volume, geographic diversity, and timing should proposers expect for Government-furnished COSMIC example data?

Answer: It will vary throughout the program. This will be discussed at program kick-off.

48. COSMIC states that Government-furnished example data is not expected to be sufficient for training. Are commercially licensed EO, SAR, multispectral, UAV, traffic, weather, or other geospatial datasets acceptable if the license permits Government and T&E access but does not allow public redistribution?

Answer: Yes.

49. COSMIC: Must all data used to build or train the PPD be deliverable to the Government, or is it acceptable to deliver trained models, embeddings, documentation, provenance records, and access mechanisms when raw commercial data redistribution is restricted?

Answer: Data used to build or train the PPD is expected to be delivered to the Government.

50. COSMIC states that development of novel change-detection algorithms and novel agentic AI systems is out of scope. Is it acceptable for a performer to provide APIs, data formats, embeddings, uncertainty estimates, and visualization outputs that are compatible with Government-designated change-detection and agentic systems, without developing those systems?

Answer: Yes.

51. COSMIC: Will T&E include modalities, sensors, geographic regions, or collection conditions not represented in the performer’s development data? If so, will proposers receive any characterization of these expected distribution shifts at kickoff?

Answer: Yes. No, complete information about distribution shifts will not be provided.

52. COSMIC: Will the Government identify the specific change-detection algorithms to be used in COSMIC evaluation at kickoff, before each evaluation event, or only at T&E?

Answer: The Government may identify some change detection algorithms at kick-off;

however, these may evolve and change throughout the program.

53. LOCUS: I understand you require our proposed system to work with images, sound, and video for geolocation. But at test time, is our system supposed to work with any number (1, 2 or 3) of modalities? i.e. Is it supposed to work for all these scenarios:

A. One modality, e.g. audio B. Two modalities, e.g. audio + image C. Three modalities, e.g. audio + image + video

Answer: Yes to all.

54. For purposes of satisfying the OT eligibility conditions under 50 U.S.C. § 3024(m)(6), may a U.S. academic institution serving as the lead proposer and performing significant technical work qualify as the “non-profit research institution participating to a significant extent in the prototype project”?

Answer: A U.S. academic institution can be awarded an OT if it qualifies as a “non-profit research institution” and participates to a significant extent in the prototype project.

55. If a U.S. academic institution may qualify as the participating non-profit research institution, should the white paper identify the academic institution itself under Condition 1, or does the Government interpret “non-profit research institution” as limited to non-university research nonprofits?

Answer: Yes, Condition 1.

56. How will the Government assess whether an NDC or non-profit research institution is participating “to a significant extent”? For example, will this be evaluated qualitatively based on technical role and key personnel, or is there an expected minimum percentage of cost, labor, or task ownership?

Answer: The Government Program Manager will decide qualitatively based on the technical approach proposed.

57. Section 4.C.1 states that white papers are continuous after the posting of Q&As, preferred. Will white papers submitted before the Q&A amendment be accepted and evaluated, or should proposers wait until after the Q&A amendment is posted?

Answers: All submitted White Papers will be evaluated. It is up to each Proposer to determine time of submission.

58. If a program is amended from “Active” to “Not Active” after a white paper has been submitted through IDEAS but before the Government has completed its assessment, will that already-submitted white paper still be evaluated?

Answer: Yes.

59. May one lead organization submit separate white papers to more than one ETA program, provided each white paper addresses only one program? Separately, may a single lead organization submit multiple white papers to the same ETA program if the technical approaches and teams are materially distinct?

Answer: Submit one White Paper per ETA. Choose the best possible technical approach/solution.

60. Are the Academic Institution Acknowledgment Letter, NSPM-33 covered-individual disclosures, Research Security Program certifications, and current/pending support documents required only with invited Volume I proposals, or are any of these required at the white-paper stage?

Answers: Items listed in your question do not need to be submitted with the White Paper.

The requirements for the White Paper are listed ETA Program Announcement Section 4.A.1.

61. For programs that involve LLMs, diffusion models, foundation models, or other external AI services during development, should proposers assume that final T&E and transition environments will be self-hosted, closed-network, and unable to call external commercial APIs?

Answer: This will very per program, however some will be implemented on self-contained networks. Further details will be available at individual program kick-offs.

62. Are open-source models, model weights, software libraries, and datasets permissible if their licenses allow Government use, modification, reproduction, evaluation, and transition consistent with the rights offered in the proposal? Should proposers identify all such licenses in the white paper or only in the invited Volume I proposal?

Answers: In Volume I, if invited to submit.

63. For performer-developed software, model weights, technical data, and datasets, will Government Purpose Rights generally be sufficient for transition, or will offering Unlimited Rights be viewed more favorably during evaluation?

Answer: Restrictions are ultimately a decision for the Proposer. Both, Unlimited Rights or Government Purpose Rights, work for a smooth transition.

64. The announcement states that proposals using proprietary data that restricts Government transfer or transition are out of scope for several programs. For commercial datasets, what minimum rights must the Government receive for T&E, transition, and future use?

For example, is a license sufficient if it allows Government and T&E use but not public redistribution?

Answer: Yes, your example is sufficient.

65. Will white papers be reviewed by the same SETA and support organizations listed for proposal evaluation? If so, is there a mechanism at the white-paper stage for proposers to object to review by one or more listed support organizations?

Answer: Let us know on the Cover Page of the White Paper. If invited to Submit Volume I, Section 1, “Cover Sheet.”

66. At the white-paper stage, is a high-level ROM budget using institutional cost assumptions sufficient, or should proposers include estimated indirect costs, subcontractor costs, software/IP costs, and travel costs in a structured table?

Answer: The ROM should include high level cost estimates for the Prime and Team Members. Do not submit a highly detailed cost breakdown. The goal is to find out approximately how much your technical solution/approach is going to cost.

67. In the Q&A published on May 5th, it was noted that the Principal Investigator may also serve as Lead System Integrator for the ARCADE program. Can you kindly confirm that the same applies to the required LSI role under the Decipher program?

Answer: Yes.

68. ARCADE: When does the Government expect to provide the initial ARCADE component library and technical documentation—before white-paper submission, after white-paper selection, at kickoff, or after award?

Answer: At kick-off.

69. ARCADE: What component classes are expected in the initial ARCADE library, and should proposers assume the test set will include passive, active, electromechanical, RF, power-management, and integrated-circuit components?

Answer: The library will include a wide variety of commercially available components, including passive, active, electromechanical, RF, power-management, sensors, connectors, microcontroller families, and some FPGAs.

70. ARCADE: May performers augment the Government-provided ARCADE library with publicly available manufacturer datasheets, application notes, reference designs, and other public technical documentation, provided the proposer can document lawful acquisition and use?

Answer: Yes, performers are encouraged to contribute additional data to enhance the collective knowledge base throughout the program.

71. ARCADE T&E, should proposers assume that final systems must operate without access to external commercial LLM APIs? If a commercial or open foundation model is used during development, must the final system provide deliverable model weights or an equivalent self-hosted substitute?

Answer: ARCADE performers are encouraged to utilize LLMs during development and provide complete codebase and documentation for the ARCADE T&E to evaluate using the same LLMs. Final architectures should remain adaptable for customers requiring self-hosted, closed network deployments, but T&E will only evaluate solutions on the models chosen by performers.

72. ARCADE: Numerical reasoning and interpolation metrics, will ground truth be determined from manufacturer data, Government-provided source-of-truth calculations, human SME assessment, or a combination of these?

Answer: ARCADE interpolation errors will be evaluated on a combination of manufacturer data and human SME assessment

73. Section 4.B.2 states, “The Government anticipates cost reimbursement proposals;

however, Proposers requesting other than cost reimbursement (i.e., FFP) may be directed by the AO to provide “other than certified cost or pricing data” and/or cost supporting information in a different format than described below. The resultant OTs will reflect Fixed-Cost Milestones/Deliverables.” Assuming Proposers bid cost-reimbursement:

A. Can you please confirm proposed fee/profit is allowed?

Answer: Yes, a reasonable fee/profit is allowed.

B. Can you confirm that the resultant OT will essentially act as a fixed price award with milestone payments so that the awardee will be paid the fixed amount for each completed milestone regardless of actual incurred costs?

Answer: Milestones/deliverables will have a reasonable, agreed upon fixed cost associated with them. The expectation is that the fixed cost per milestone/deliverable will be fully supported by the cost volume.