Strategic Optimization of an IARPA World-Events Simulation Platform: A Comprehensive User Journey and Governance Analysis

Full report

On this page

The Epistemological Framework of Geopolitical Simulation and Hybrid Intelligence

The Intelligence Advanced Research Projects Activity (IARPA) occupies a unique and critical position within the Office of the Director of National Intelligence (ODNI). Tasked with anticipating technological surprises and delivering overwhelming advantages to the United States Intelligence Community (IC), IARPA operates without an operational field mandate, focusing entirely on high-risk, high-payoff research that facilitates the transition of revolutionary capabilities to IC partners1. Among its most profound structural impacts on the intelligence apparatus has been the pioneering of quantitative geopolitical forecasting. Beginning with the Aggregative Contingent Estimation (ACE) program, which launched the Superforecasting industry and constituted the world’s largest forecasting experiment, IARPA demonstrated that the mathematical aggregation of widely dispersed human judgments could consistently outperform individual expert analysis1. However, as the global information environment has grown exponentially more complex, human cognitive capacity alone has proven insufficient for ingesting the sheer volume of structural data, news feeds, and open-source indicators required for modern predictive analysis. To address this, subsequent initiatives such as the Hybrid Forecasting Competition (HFC) and the Crowdsourcing Evidence, Argumentation, Thinking and Evaluation (CREATE) program have iteratively refined the fusion of human cognitive reasoning with machine learning (ML) architectures4. The HFC program, for instance, successfully beat the state-of-the-art for human-only forecast systems by 10 percent, proving the efficacy of integrating human intuition with the rapid data-processing capabilities of AI4. Designing an interactive, public-facing world-events simulation for IARPA requires translating these highly specialized, multi-year scientific endeavors into an accessible, compliant, and highly engaging digital user funnel. A world-events simulation platform must systematically convert passive internet users into calibrated analysts capable of eliciting, weighting, and combining probabilistic judgments for contingent events3. This transformation must occur while strictly adhering to the U.S. Web Design System (USWDS) guidelines, the 21st Century Integrated Digital Experience Act (IDEA), and IARPA's internal Research and Technology Protection (RTP) protocols8. This report provides an exhaustive architectural and strategic breakdown of the user funnel required to maximize discovery, comprehension, trust, viral dissemination, and long-term retention within an IARPA world-events simulation. It integrates findings from computational linguistics, cognitive psychology, hybrid intelligence design, and federal compliance frameworks to map a continuous pathway that preserves deterministic facts, respects publication controls, and enforces rigorous editorial governance.

Phase 1: Eliciting Discovery and Engineering Acquisition

The foundational challenge of any crowdsourced intelligence initiative is the acquisition of a highly diverse, intellectually rigorous participant base. The IC relies on forecasting platforms to overcome the "hidden profile problem," a phenomenon where critical information is unevenly scattered across different members of a group, making it challenging to piece together a coherent predictive model11. If the simulation attracts a homogenous user base, the resulting forecasts will suffer from profound cognitive and demographic biases. Therefore, the top of the acquisition funnel—comprising search engine visibility and social media propagation—must be meticulously engineered to attract a cognitively diverse crowd while maintaining the epistemic authority of the federal government.

Search Snippets: Projecting Federal Authority and Epistemic Rigor

Search engine discovery is heavily mediated by the structural presentation of metadata. For a platform operating under the auspices of the ODNI1, search snippets must immediately telegraph institutional authority, scientific rigor, and public utility. When users input queries related to geopolitical risk, technological forecasting, S\&T milestones, or crisis recovery modeling (e.g., searches mirroring the COVID-19 recovery efforts tasked to the SWARM technology developed by the University of Melbourne under the CREATE program7), the search snippet must differentiate the IARPA platform from commercial news aggregators, speculative betting markets, or sensationalist media. This differentiation requires the implementation of precise schema markup and metadata strategies aligned with Google Search Central guidelines13. The search snippet must prioritize the communication of "Gold Standard Science," a core guiding principle of IARPA's research integrity1. Instead of utilizing click-driven, deterministic titles, the SEO strategy should utilize structured data to present current, active forecasting questions. For example, a search snippet might read: "What is the probability of a major science and technology milestone in quantum advantage by 2026? Contribute to the IARPA ForeST program simulation." The meta-description must explicitly frame the platform as a USG-backed scientific experiment aimed at harnessing collective intelligence to enhance national security1. To optimize discovery without running afoul of sensationalism, the platform must utilize semantic precision. By employing natural language processing and ontological tagging methodologies developed through the BETTER program—which aims to extract personalized, mission-relevant semantic information from massive volumes of multilingual text16—the platform can ensure its search terms perfectly match the precise queries utilized by analysts, academics, and highly engaged citizens. Crucially, search snippets should reflect dynamic uncertainty rather than deterministic outcomes. If the simulation tracks the likelihood of a viral outbreak, the snippet must clearly state "Aggregate Crowd Probability: 42%" rather than implying a definitive future state. This preemptively trains the user to think in probabilistic terms before they even arrive at the landing page12.

Social Previews: Engineering the Discovery Loop through Cyberpsychology

While search engines capture users with high pre-existing intent, social media environments (such as X and LinkedIn, where IARPA maintains an active public presence18) are required to generate viral discovery and attract lateral thinkers. However, social media algorithms inherently prioritize emotional resonance, partisan division, and absolute certainty over the well-calibrated doubt required for effective geopolitical forecasting. Designing social previews (such as Open Graph tags and Twitter Cards) for a world-events simulation therefore requires sophisticated cyberpsychological engineering. The social preview acts as the primary hook in the acquisition funnel. Drawing on principles from IARPA's ReSCIND (Reimagining Security with Cyberpsychology-Informed Network Defenses) program—an interdisciplinary initiative that integrates human behavior, decision-making biases, and cognitive vulnerabilities into the cyber-domain19—the social preview should utilize "information gap theory" to provoke intellectual curiosity. If a user shares a simulation node regarding the ForeST (Forecasting Science & Technology) program's tracking of a specific patent milestone or technological breakthrough15, the social preview should visually display a stark contrast between disparate forecasting models. For example, indicating that "Machine Learning models predict a 20% chance of occurrence, while Top Human Superforecasters predict an 80% chance" creates an immediate cognitive dissonance. This visual disparity drives click-through rates by challenging the user to resolve the contradiction, leveraging the core premise of the HFC program, which found that highlighting the tension between human and machine integration increases both accuracy and user engagement17. Furthermore, to combat the proliferation of deepfakes and automated disinformation, social previews must distinctly feature the .gov domain indicator and official IARPA branding. This preemptively counters misinformation and signals to the prospective user that the link leads to a secure, federally vetted environment, adhering strictly to the USWDS principle of earning trust at every touchpoint8.

Phase 2: Onboarding, Cognitive Load Mitigation, and Trust Calibration

Once a user transitions from a search engine or social feed to the simulation platform, the system has a narrow temporal window to establish institutional credibility, explain a complex hybrid intelligence mechanism, and guide the user through their first probabilistic interaction. The design of this phase dictates the retention rate of the platform.

Landing-Page Promise: The USWDS Trust Architecture

The landing page is the primary interface for actualizing the "Earn Trust" and "Promote Continuity" design principles mandated by the U.S. Web Design System8. Under the stringent guidelines of the OMB M-23-22 memorandum for delivering a digital-first public experience, as well as the 21st Century Integrated Digital Experience Act (IDEA), the visual hierarchy and technical foundation of the landing page must be impeccable9. The immediate visual presentation must include the official USWDS banner component, assuring the user that the site is an official United States government digital service utilizing secure, up-to-date HTTPS certificates8. This is not merely an exercise in bureaucratic compliance; it is a fundamental psychological anchor. Users are being asked to contribute their intellectual labor, cognitive reasoning, and time to a national security experiment. Trust in the platform's data stewardship, reliability, and honesty is paramount for generating high-quality inputs8. The core "promise" articulated on the landing page must distill the highly technical mission of IARPA's Office of Analysis—to maximize insights from massive, disparate, unreliable, and dynamic data24—into a compelling, user-centric value proposition. The copy must explain that the user is participating in an advanced evolution of hybrid forecasting, utilizing systems akin to SAGE, a platform developed by the University of Southern California Information Sciences Institute (USC ISI) that anchors human judgments on objective machine benchmarks4. To maintain the integrity of deterministic facts, the landing page architecture must present a clear, visual, and ontological separation between historical data (verified, deterministic events) and forecasts (probabilistic, contingent futures). Historical facts should be presented with links to authoritative open-source indicators, while ongoing forecasts must be accompanied by confidence intervals, baseline rates, and explanations of Brier scoring mechanisms26.

First Interaction: Elicitation, Anchoring, and the SAGE Framework

The first active interaction within the simulation—the moment the user submits their first forecast—is the point of highest UX friction. The original ACE program recognized that the efficient elicitation of probabilistic judgments, especially concerning complex conditional probabilities for contingent global events, is cognitively overwhelming for untrained individuals3. To lower this barrier to entry without sacrificing the rigorous data collection required by the IC, the platform should embed structured analytic techniques (SATs) directly into the user interface. The REASON (Rapid Explanation, Analysis and Sourcing ONline) program, managed by Dr. Steven Rieber, focuses precisely on enhancing logic, critical thinking, and rational judgment27. The simulation UI should guide the first-time user through a proprietary mechanism modeled on the NEUER bias-sensitive structured analytic system28. This system assists users in assessing the quality of their information, acknowledging their own cognitive dispositions, and subsequently assigning highly calibrated probabilistic values to their outputs28. Instead of presenting a blank text box asking "Will Event X occur by December?", the interface should present a guided, multi-step cognitive breakdown:

1. Base Rate Presentation: What is the historical frequency of similar events occurring within this timeframe?

2. Machine Benchmark: What does the integrated Artificial Intelligence model currently predict? Research from the HFC program demonstrates that human forecasters achieve higher accuracy when they can anchor their judgments against an objective machine baseline4.

3. Human Adjustment: The user inputs their probability using a visual slider, deliberately adjusting the machine's forecast based on their localized knowledge, unique information diet, or human intuition regarding geopolitical nuances that the machine may lack.

Crucially, this entire interaction sequence must be built on the USWDS progressively enhanced HTML foundation, ensuring core experiences function across all browsers29. Furthermore, the components must meet or exceed WCAG 2.1 Level AA conformance, surpassing baseline Section 508 requirements, to guarantee that users relying on assistive technologies can fully and seamlessly participate in the forecasting ecosystem21.

Phase 3: Value Realization and the Mechanics of Dissemination

To successfully convert a transient, first-time visitor into an active, long-term forecaster, the platform must deliver immediate cognitive value to the user and provide a seamless, policy-compliant mechanism for sharing that value with their external professional and social networks.

First Meaningful Insight: Calibration, Brier Scores, and the Hidden Profile

The psychological "aha" moment in a geopolitical forecasting simulation occurs when a user receives definitive feedback on their cognitive accuracy and forecasting skill. Because geopolitical events, military conflicts, and disease outbreaks often take months or even years to resolve (for example, the extended timelines required to predict the availability of a COVID-19 vaccine12), the system cannot always provide immediate outcome-based feedback. In the absence of immediate event resolution, the first meaningful insight delivered to the user must be highly personalized calibration feedback. Drawing from the architectural successes of the HFC and ACE programs, the platform should immediately compute and display how the new user's forecast interacts with the aggregate wisdom of the crowd and the baseline machine models4. The SAGE system's advanced weighting protocols can visually demonstrate to the user how their specific input shifted the global consensus, adjusting in real-time for overconfidence and weighting for propinquity25. Furthermore, the simulation can utilize Variational Autoencoders (VAEs) to analyze the user's information sources and demonstrate how their judgment correlates with others in the system. Research has shown that while demographic diversity is vital, search engines often turn demographic similarity into information similarity, causing groups to make correlated errors11. If a user relies heavily on standard algorithmic search results, the system can reveal how that specific information diet aligns them with a redundant cohort, thereby teaching them about the "hidden profile problem" and group modularity11. This delivery of meta-knowledge—helping the user understand how they think, rather than simply scoring what they think—is the defining feature of the reasoning improvements pioneered by the CREATE program5.

Sharing Trigger: Engineering Viral Loops Under Publication Controls

Once a user experiences this cognitive insight and realizes their predictive model outperforms the baseline, the natural psychological impulse is to share their status publicly (e.g., "My hybrid forecast beat the machine learning baseline by 15% on Middle East stability metrics"). However, operating a USG-backed intelligence platform means that social sharing triggers are heavily constrained by Research and Technology Protection (RTP) protocols10 and strict publication controls. The platform must operate continuously under the guidance of National Security Decision Directive (NSDD) 189, which establishes the cornerstone policy distinguishing unclassified, fundamental research from classified operational intelligence10. While the simulation deals exclusively in unclassified, open-source intelligence, the aggregated crowdsourced data, if overly specific, could potentially present a security vulnerability by inadvertently forecasting sensitive USG equities or revealing strategic blind spots10. IARPA's RTP team specifically researches, identifies, and evaluates vulnerabilities to the IC posed by proposed research goals and open scientific collaboration10. To enable safe sharing that drives viral acquisition without compromising national security, the platform must implement stringent data sanitization workflows:

1. Automated Sanitization: The sharing mechanism must automatically strip out any underlying sensitive rationales, highly specific geospatial coordinates, or unvetted free-text arguments generated by the user.

2. Abstracted Metrics: Users should be encouraged to share their relative Brier score, their percentile rank within the broader "Human Forest"28, or broad, abstracted trend lines. They must be technologically prevented from exporting raw underlying datasets or API feeds to unauthorized third-party platforms.

3. Editorial Review Governance: Any user-generated structured narratives or detailed argumentation mapping (a key feature of the CREATE program's SWARM platform used for policy analysis)7 must remain securely within the simulation's authenticated walled garden. This prevents the viral spread of user-generated speculation or misinformation bearing the visual hallmark and implicit endorsement of a .gov domain.

By restricting external sharing to visually engaging, USWDS-styled infographics of personal accuracy metrics and abstracted rankings, the platform creates a secure viral loop. This loop naturally attracts highly competitive, analytically minded professionals and academics—the precise demographic required to cultivate a new generation of Superforecasters2.

Phase 4: Retention, Privacy Architecture, and Persistent Engagement

A predictive market or simulation is only as valuable as the longitudinal persistence of its participants. Geopolitical forecasting requires continuous Bayesian updating; users must be drawn back to the platform over months and years to revise their forecasts as new evidence emerges. Sustaining this engagement requires a flawless privacy architecture and intelligent, asynchronous return triggers.

Saved State: Cookie-less Architecture and NSPM-33 Privacy Guidelines

Retaining user state, tracking Brier scores over time, and maintaining longitudinal forecasting data presents a significant technical challenge in the era of strict federal privacy guidelines. The OMB M-23-22 directive, alongside IRS IRM 10.5.1 and broader federal privacy acts, mandates the careful stewardship of user data, heavily restricting the use of invasive third-party tracking mechanisms and cross-site cookies22. Furthermore, the implementation of National Security Presidential Memorandum-33 (NSPM-33) requires rigorous assessment of foreign conflicts of interest for funded researchers and institutions1, establishing a baseline culture of extreme data security and vetting at IARPA. To maintain a persistent "Saved State" for the user without relying on invasive tracking, the platform must utilize first-party, secure authentication protocols. More importantly, the system must protect the privacy of the machine learning models themselves. As identified in IARPA's Secure, Assured, Intelligent Learning Systems (SAILS) program, ML models can inadvertently "memorize" their training data. This leads to critical vulnerabilities where malicious actors can perform membership inference attacks to identify the individuals who trained the model, or execute training data reconstruction attacks to extract sensitive inputs34. The simulation must employ robust statistics and cryptography to ensure "apples-to-apples" defensive measures against these attacks, accommodating both white-box and black-box adversary access34. The user's saved state—their comprehensive history of forecasts, their identified cognitive biases, and their historical accuracy—must be cryptographically decoupled from their personally identifiable information (PII) using advanced differential privacy techniques. This ensures that even if the aggregate human-machine data is queried or compromised, individual user judgments and identities cannot be reverse-engineered or exploited.

Return Trigger: Asynchronous Collaboration and Event Resolution

Because world events do not adhere to a predictable schedule, users must be proactively drawn back to the platform to update their beliefs when structural data shifts. The architecture for these return triggers should be modeled on the FOCUS program, which enhanced analysis tradecraft by drastically improving analysts' ability to learn lessons from the past27. When a contingent event officially resolves (e.g., a foreign political election concludes, or a specific S\&T patent is filed), the system must push an automated, secure notification to the user detailing the final, deterministic outcome juxtaposed against their predicted probability. This closed-loop feedback is the primary mechanism for improving long-term human reasoning. Additionally, the system should trigger returns when the machine model detects a significant shift. If the AI component of the hybrid system detects a massive anomaly in streaming structural data, news feeds, or global web search queries (akin to the automated continuous analysis methodologies pioneered in the Mercury and Open Source Indicators (OSI) programs)20, it should immediately alert the human forecasters. The alert would prompt them to review the new data parameters and update their subjective probabilities accordingly. This continuous, asynchronous interplay solidifies the human-machine trust dynamic that was critical to the HFC program's success4.

Phase 5: Deeper Research Path and the Evolution of the Analyst

As users mature on the platform and demonstrate sustained accuracy, they must be offered deeper, more cognitively complex research pathways. The platform should gracefully transition from a simple, slider-based forecasting interface to a sophisticated, professional-grade environment for intelligence analysis. Advanced users who rank highly in the system should be granted access to specialized, structured collaboration tools developed under the CREATE program. This would allow geographically dispersed teams to collaborate seamlessly to assess complex evidence, identify key information gaps, and challenge deeply held assumptions using tools similar to the Smartly-assembled Wiki-style Argument Marshalling (SWARM) platform7. They can also be invited to participate in combinatorial prediction markets, similar to the SciCast tournament managed by ForeST-funded researchers, generating public forecasts for highly technical, interrelated S\&T events15. To facilitate this deeper research, the platform must integrate the profound Natural Language Processing (NLP) advancements achieved by IARPA's Human Language Technology (HLT) initiatives. By integrating capabilities from the Babel, MATERIAL, and BETTER programs, the platform can empower users to rapidly extract semantic information, entities, and events from diverse, multilingual, unstructured text2. This allows the crowdsourced forecasters to analyze foreign-language news feeds or open-source datasets without needing specialized linguistic expertise, dramatically compressing the information discovery cycle and minimizing the amount of English training data needed to develop functional predictive models1.

Governance, Editorial Review, and Data Integrity

An IARPA-backed simulation cannot function as an unmoderated public square. The aggregation of crowd wisdom is highly susceptible to coordinated disinformation campaigns, groupthink, and adversarial data poisoning. Strict governance architectures must be baked into the platform's codebase and standard operating procedures.

Preserving Deterministic Facts

In a digital platform built entirely around subjective probability and uncertain futures, deterministic facts must be visually and structurally sacrosanct. The UI must aggressively separate verified baseline data from user and machine forecasts. Utilizing USWDS design tokens, customized CSS architectures, and typography utilities23, the system should visually demarcate established facts (e.g., "The GDP of Country X in 2025 was $Y") using distinct color systems and spacing units. This visual language ensures users never mistake a highly confident 90% forecast for a historical certainty.

Editorial Review Governance and Quality Control

The open internet is prone to catastrophic noise. To ensure that the analytical output is not contaminated by bad actors or automated botnets, the platform must implement automated analytic triage6. Drawing on the logic of the DIVA (Deep Intermodal Video Analytics) and KDD (Knowledge Discovery and Dissemination) programs, which focus on rapidly extracting actionable intelligence and highlighting critical activities from massive data streams35, the platform can deploy AI to continuously monitor user inputs for semantic anomalies or coordinated inauthentic behavior. If a swarm of newly registered accounts suddenly inputs highly correlated, heavily biased forecasts on a specific geopolitical crisis, the system's variational autoencoders11 should instantly flag these nodes. Their data must be temporarily sequestered from the primary aggregation algorithm until a human editorial review board can verify the integrity of the accounts and their underlying rationale. Furthermore, editorial governance requires that the actual forecasting questions are generated, scoped, and phrased exclusively by domain experts, not the general crowd. The precise phrasing of contingent events dictates the mathematical validity of the resulting Brier scores. IARPA Program Managers, IC methodologists, and authorized researchers must retain absolute, centralized control over the platform's overarching ontology and question framing27.

Managing Human Trust in AI Coworkers

A critical finding of the HFC program was the absolute necessity of foundational research into human trust of machine models4. If the machine baseline provided to the user is consistently wrong or inexplicable, human forecasters will quickly abandon the platform. Conversely, if the machine is perceived as an omniscient black box, humans will suffer from severe automation bias, simply agreeing with the AI and defeating the entire purpose of a hybrid intelligence system4. To mitigate this, the platform must maintain radical transparency regarding the machine's capabilities and limitations. When presenting an algorithmic forecast, the system must expose the "reasoning for results"24—explicitly highlighting the open-source indicators, historical structural data, or specific news feeds that heavily influenced the model's current weightings20. By demystifying the AI, the platform empowers human analysts to identify the machine's contextual blind spots and contribute their unique, human cognitive insights to cover those gaps, maximizing the accuracy of the resulting hybrid forecast20.

Proposed Measurable Experiments

To continuously optimize this complex user journey and ensure the platform scales effectively while maintaining rigorous scientific standards, IARPA must employ systematic A/B testing methodologies. These experiments must adhere strictly to federal privacy rules, ensuring no cross-site tracking or unauthorized PII collection occurs during the evaluation process9. The following table outlines highly specific, measurable experiments designed to evaluate and optimize critical friction points within the simulation funnel:

By rigorously testing these variables, the platform administrators can systematically refine the user experience, ensuring that the simulation operates at the absolute frontier of both user engagement and scientific validity.

Strategic Synthesis

The realization of a public-facing IARPA world-events simulation represents a monumental convergence of human-computer interaction, cognitive science, and national security policy. To successfully leverage the aggregated wisdom of the crowd, the platform cannot simply mirror commercial web applications or speculative prediction markets. It must be architected from the ground up as a highly controlled, deeply secure epistemic environment. By strictly adhering to the components and design principles of the U.S. Web Design System, the simulation establishes an immediate and unassailable perimeter of public trust8. By integrating the hard-won lessons of the ACE, HFC, and CREATE programs3, the user interface mitigates innate cognitive biases and effectively combines the nuanced reasoning of human analysts with the immense, real-time data-processing capabilities of machine models4. Finally, by enforcing the stringent guidelines of Research and Technology Protection, OMB digital mandates, and the SAILS privacy framework10, the platform ensures that the rapid discovery and dissemination of open-source intelligence does not inadvertently compromise the very national security it seeks to bolster. Through the continuous execution of the measurable experiments proposed in this analysis, IARPA can optimize this digital funnel, identifying the next generation of Superforecasters and drastically compressing the information discovery cycle2. Ultimately, the strategic deployment and rigorous governance of this simulation will fulfill IARPA's foundational mission: pushing the boundaries of science to empower the Intelligence Community with an overwhelming, anticipatory advantage in a profoundly complex geopolitical landscape1.

Works cited

1. About IARPA, https://www.iarpa.gov/who-we-are/about-us

2. Our Programs \- IARPA, https://www.iarpa.gov/who-we-are/history/our-programs

3. ACE \- IARPA, https://www.iarpa.gov/research-programs/ace

4. HFC \- IARPA, https://www.iarpa.gov/research-programs/hfc

5. IARPA Launches "CREATE" Program To Improve Reasoning Through Crowdsourcing, https://www.iarpa.gov/newsroom/article/iarpa-launches-create-program-to-improve-reasoning-through-crowdsourcing

6. HFC \- IARPA, https://www.iarpa.gov/images/OA-Slicksheets/HFC\_SlickSheet\_FINAL\_05072021\_Prepubapproved.pdf

7. CREATE \- IARPA, https://www.iarpa.gov/research-programs/create

8. Design principles | U.S. Web Design System (USWDS) \- Digital.gov, https://designsystem.digital.gov/design-principles/

9. GAO-24-106764, DIGITAL EXPERIENCE: Agency Compliance with Statutory Requirements, https://www.gao.gov/assets/gao-24-106764.pdf

10. Research and Technology Protection \- IARPA, https://www.iarpa.gov/research-programs/research-and-technology-protection

11. Experiments in Collective Intelligence \- Niccolo Pescetelli, https://niccolopescetelli.com/starting-points/hybridIntelligence.html

12. Search \- IARPA, https://www.iarpa.gov/search?q=sign%20as%20day%20slides%20seem%20\&start=240

13. Help Us Pick the Next Stop in Europe for Search Central Live Deep Dive 2026\!, https://developers.google.com/search/blog/2026/06/scl-deep-dive-europe-2026

14. Search \- IARPA, https://www.iarpa.gov/search

15. ForeST \- IARPA, https://www.iarpa.gov/research-programs/forest

16. BETTER \- IARPA, https://www.iarpa.gov/research-programs/better

17. IARPA Launches Hybrid Forecasting Competition to Improve Predictions Through Human-Machine Integration, https://www.iarpa.gov/newsroom/article/iarpa-launches-hybrid-forecasting-competition-to-improve-predictions-through-human-machine-integration

18. IARPA \- Intelligence Advanced Research projects Activity \- Office of the Director of National Intelligence, https://www.iarpa.gov/

19. Using Psychology to ReSCIND Cyberattacks \- IARPA, https://www.iarpa.gov/newsroom/article/using-psychology-to-rescind-cyberattacks

20. IARPA launches Hybrid Forecasting Competition \- Intelligence Community News, https://intelligencecommunitynews.com/iarpa-launches-hybrid-forecasting-competition/

21. U.S. Web Design System — Grokipedia, https://grokipedia.com/page/us\_web\_design\_system

22. M-23-22: Delivering a Digital-First Public Experience \- The White House, https://www.whitehouse.gov/wp-content/uploads/2023/09/M-23-22-Delivering-a-Digital-First-Public-Experience.pdf

23. USWDS: The United States Web Design System | U.S. Web Design System (USWDS) \- Digital.gov, https://designsystem.digital.gov/

24. Research Programs \- IARPA, https://www.iarpa.gov/research-programs

25. Hybrid forecasting of geopolitical events \- Information Sciences Institute, https://www.isi.edu/results/publications/12975/hybrid-forecasting-of-geopolitical-events

26. CREATE Frequently Asked Questions (FAQs) Q) What is the difference between CREATE and ACE? A) ACE focused on improving forec \- IARPA, https://www.iarpa.gov/images/PropsersDayPDFs/CREATE/CREATE\_FAQs.pdf

27. Forecasting \- IARPA \- Program Managers, https://www.iarpa.gov/who-we-are/program-managers?areas\_of\_interest%5B%5D=Forecasting

28. IARPA REASON, https://www.iarpa.gov/images/PropsersDayPDFs/REASON/30\_R\_JOSEPH\_REASON\_Proposers\_Day2023.pdf

29. Documentation | U.S. Web Design System (USWDS), https://designsystem.digital.gov/documentation/developers/

30. Research | U.S. Web Design System (USWDS) \- Digital.gov, https://designsystem.digital.gov/about/research/

31. Accessible Design Using the U.S. Web Design System (USWDS) | Section508.gov, https://www.section508.gov/develop/accessible-design-using-uswds/

32. Search \- IARPA, https://www.iarpa.gov/search?q=but%20some%20in\&start=240

33. 10.5.1 Privacy Policy | Internal Revenue Service, https://www.irs.gov/irm/part10/irm\_10-005-001

34. Secure, Assured, Intelligent Learning Systems (SAILS) Trojans in Artificial Intelligence (TrojAI) \- IARPA, https://www.iarpa.gov/images/PropsersDayPDFs/TrojAI/SAILS-TrojAI\_proposersday\_briefing.pdf

35. Research Programs \- IARPA, https://www.iarpa.gov/research-programs?keyword=\&office\_name=analysis\&program\_managers=\&program\_managers\_hidden=\&scroll\_position=661\&show\_current\_past=past\&show\_office=2\&sortby=asc

36. Mercury \- IARPA, https://www.iarpa.gov/index.php/research-programs/mercury