# How to Tell If a Study Is Likely to Replicate

## Introduction: The Everyday Hook and the Direct Answer

Scientific research serves as the invisible architecture of modern human society. It dictates the nutritional guidelines that shape global diets, the pedagogical strategies deployed in public education, the multi-billion-dollar investments made by the pharmaceutical industry, and the daily health decisions made by the general public. When a prominent study announces that a specific physical posture guarantees interview success, that a daily glass of wine prevents cognitive decline, or that childhood willpower determines lifelong financial stability, the public and policymakers swiftly adapt their behaviors and allocate resources accordingly [cite: 1, 2]. However, a profound and systemic vulnerability lies at the heart of this knowledge-production system: a significant portion of published, peer-reviewed scientific findings cannot be replicated. When foundational studies collapse under subsequent independent scrutiny, the resulting ripple effect is devastating. It wastes institutional resources, misdirects public policy, diverts subsequent research down blind alleys, and steadily erodes public trust in the overarching scientific enterprise [cite: 3, 4]. For the non-scientist, the ability to discern durable scientific truth from fragile statistical noise is no longer an academic exercise; it is a critical life skill.

The direct answer to assessing whether a research finding is likely to replicate—before a costly and time-consuming independent replication study is formally conducted—lies in a synthesized, multi-layered approach of structural evaluation, collective human forecasting, and emerging algorithmic prediction. First, the structural and methodological integrity of the original study must be evaluated, looking specifically for "green flags" such as massive and diverse sample sizes, preregistration of the hypothesis, open data sharing, and robust statistical power [cite: 4, 5]. Second, collective human intelligence—specifically through prediction markets and expert forecasting panels—has proven remarkably accurate at identifying fragility in published literature by aggregating the tacit knowledge of the scientific community [cite: 6]. Finally, emerging metascience frameworks utilizing Large Language Models (LLMs) and advanced machine learning are currently being deployed to assign automated, highly scalable "confidence scores" to scientific claims by analyzing the underlying text and data structures [cite: 3, 7]. By combining methodological transparency with these sophisticated predictive algorithms, the scientific community is shifting its paradigm from reactive correction to proactive verification.

## FAQ 1: What is the "Replication Crisis" and Why Does it Matter?

To grasp the magnitude of the issue currently reshaping the scientific landscape, it is essential to first distinguish between two foundational concepts that are often erroneously conflated: reproducibility and replicability. Methods reproducibility refers strictly to the ability to take the original researchers' raw data and analytical code, re-run the computational analysis, and arrive at the exact same numerical results [cite: 8, 9]. It is a test of clerical and computational accuracy. Replicability, by contrast, represents a much higher epistemic hurdle. It is the ability to conduct an entirely new experiment, following the original methodology but collecting a completely independent sample of data, and observing evidence that supports the original claim [cite: 8, 10]. 

The "replication crisis" refers to the stark realization, gaining significant momentum since the mid-2010s, that a vast swath of peer-reviewed studies across multiple disciplines fails to replicate when subjected to this independent verification [cite: 11, 12]. In massive, coordinated replication efforts—particularly those evaluating the social, behavioral, and biomedical sciences—the success rate often hovers around 50% to 55%. Furthermore, even when studies successfully replicate, the effect sizes of the replicated studies typically shrink to roughly half the magnitude of the original published claims [cite: 13, 14]. 

### Famous Historical Examples of Failed Replications

The crisis is best understood through the lens of highly publicized concepts that captured the public imagination before failing rigorous scientific scrutiny. These examples highlight how intuitive appeal often masks methodological fragility.

The phenomenon of "Power Posing" serves as perhaps the most prominent example. In 2010, researchers published a highly influential paper claiming that holding expansive, dominant postures for a mere two minutes not only increased subjective feelings of confidence but also caused significant biological transformations—namely, a sharp rise in testosterone and a corresponding drop in the stress hormone cortisol [cite: 1, 15]. This finding spawned one of the most-watched TED talks in history, profoundly influencing corporate training and self-help literature [cite: 2]. However, subsequent large-scale replication attempts dismantled these claims. A pivotal study led by Eva Ranehill and colleagues, utilizing a sample nearly five times larger than the original, failed to confirm any physiological effect or behavioral change in risk-taking [cite: 16]. Further extensive statistical analyses revealed that the existing evidence was simply too weak to justify the claims, showing unequivocally that power poses have no reliable effects on any behavioral or cognitive measure [cite: 2, 15, 16]. 

For decades, the field of social psychology was also heavily influenced by "social priming," the theory that subtle, unconscious environmental cues could dramatically alter complex human behavior. The most famous example was a 1996 study claiming that participants who unscrambled sentences containing words related to elderly stereotypes subsequently walked more slowly down a hallway [cite: 1, 16]. When independent laboratories attempted to replicate this, they consistently found no such effect. A rigorous 2012 replication demonstrated that the original effect was likely a result of the experimenters' own biased expectations and flawed experimental setups, rather than a genuine psychological phenomenon [cite: 11, 16]. The widespread collapse of these and similar findings prompted Nobel laureate Daniel Kahneman to refer to the state of social priming research as a "train wreck" [cite: 1].

The classic "Marshmallow Test" of delayed gratification suggested that a young child's ability to resist eating a single marshmallow for 15 minutes to earn a second one was a powerful, intrinsic predictor of future academic achievement and lifelong success [cite: 1, 11]. However, a 2018 replication utilizing a much larger and significantly more socioeconomically diverse sample size demonstrated that the effect was statistically insignificant once socioeconomic background was adequately controlled for. The replication revealed that for children from unstable or impoverished environments, eating the marshmallow immediately was a rational adaptation to an unpredictable world, not a biological failure of willpower [cite: 1].

Similarly, the Facial Feedback Hypothesis, widely known as the "smile effect," posited that human facial expressions do not merely reflect emotions but actively dictate them. A highly cited 1988 study reported that participants forced to smile by holding a pen between their teeth found cartoons significantly funnier than those forced to pout by holding the pen with their lips [cite: 11, 16]. This "fake it till you make it" hypothesis collapsed in 2016 when a massive, pre-registered replication effort involving 17 independent laboratories and nearly 2,000 participants found absolutely no overall effect on subjective humor ratings [cite: 11, 16]. 

Other casualties of the replication crisis include the concept of "Ego Depletion," which incorrectly theorized that willpower operates like a muscle that gets tired with use, and the "Dunning-Kruger Effect," the famous claim that low-ability individuals wildly overestimate their competence. Recent high-powered replications of the Dunning-Kruger effect have shown mixed results, suggesting the effect is generally much weaker than originally reported and may largely be a statistical artifact rather than a profound cognitive bias [cite: 1, 11]. 

## FAQ 2: Doesn't Peer Review Catch Bad Science and Fraud?

A pervasive and dangerous misconception among the public, policymakers, and the media is that publication in a peer-reviewed academic journal serves as an ironclad guarantee of scientific truth. In reality, peer review is a basic sanity check designed to evaluate the theoretical plausibility of a study, the appropriateness of its methodology, and its contribution to the field; it is absolutely neither designed nor equipped to detect deliberate fraud, nor can it guarantee future replicability [cite: 17, 18]. 

### The Structural Limitations of Peer Review

When academic reviewers assess a manuscript, they operate under the fundamental assumption of collegial honesty [cite: 19]. They evaluate the summarized data exactly as it is presented to them by the authors. Because reviewers rarely have access to the underlying raw data—and even more rarely possess the vast amount of time required to conduct a full statistical re-analysis—they simply cannot identify fabricated numbers or covert manipulations [cite: 17, 18, 20]. In some instances, asking authors to submit their raw data during the review process caused half of them to retract their own work, strongly suggesting that the data never existed in the first place [cite: 20].

As historical cases demonstrate, fraudulent science routinely passes peer review with flying colors. The physics fabrications of Jan Hendrik Schön, the massive data manipulations by social psychologist Diederik Stapel, and the deeply flawed biomedical claims of Theranos all survived the peer review process. In all these cases, the fabricated data survived until independent replication attempts, post-publication whistleblowing, or formal institutional inquiries revealed the truth [cite: 17, 19]. Furthermore, the peer review system itself is vulnerable to exploitation. There have been numerous documented cases where researchers entered false email addresses for suggested reviewers, subsequently accepting the invitations to review their own papers and submitting highly favorable reports [cite: 19]. Consequently, peer review should be viewed as only one fragile layer in a much broader integrity system, not a lie detector [cite: 17, 20].

### Fraud vs. Replication Failure: A Critical Distinction

It is vital to distinguish between deliberate scientific fraud and standard replication failure. Conflating the two damages the public understanding of the scientific method and misdirects reform efforts [cite: 21]. 

Fraud involves the deliberate, intentional fabrication, falsification, or covert manipulation of data to achieve a desired, usually career-advancing, outcome [cite: 21, 22]. While devastating when it occurs, blatant fraud is statistically rare compared to other systemic issues [cite: 21]. Fraud requires malicious intent and maps closely onto the definitions of occupational fraud seen in corporate environments, where perpetrators use active deception to achieve an illicit gain at the expense of an organization [cite: 23, 24]. Just as in the corporate sector, where the Association of Certified Fraud Examiners notes that lack of internal controls contributes to over half of fraud cases, the lack of rigorous data auditing in academia allows rogue researchers to operate undetected [cite: 24, 25].

Replication failures, conversely, are primarily the result of statistical noise, honest methodological errors, or deeply ingrained systemic biases within the academic research culture. The pressure to publish novel, positive results in high-impact journals incentivizes Questionable Research Practices (QRPs) [cite: 4]. These include *p-hacking* (the practice of endlessly re-analyzing data in different ways until a statistically significant result magically emerges) and *HARKing* (Hypothesizing After the Results are Known), which drastically distorts the scientific method [cite: 4, 6, 26]. Furthermore, publication bias—the tendency of journals to only publish positive, ground-breaking results while ignoring negative findings or failed replications—creates a skewed literature full of false positives [cite: 4]. 

A failed replication does not automatically imply that the original researchers were unethical or malicious; it frequently indicates that the original finding was highly context-dependent, relied on a sample size too small to be generalizable, or was simply a statistical artifact [cite: 15, 21]. Correcting these errors through rigorous replication is not a sign of science breaking down. Rather, finding errors requires intense scientific work, and doing so is a demonstration of the scientific enterprise engaging in its most vital function: self-correction [cite: 15, 21]. 

Grassroots initiatives and individuals have stepped into the void left by peer review to enforce this self-correction. Platforms like PubPeer allow for anonymous, post-publication peer review, where the community can flag inconsistencies. Individuals like Elisabeth Bik have reviewed tens of thousands of scientific images to identify duplications and manipulations, exposing fraud and sparking institutional responses, while organizations like Retraction Watch meticulously document why studies are pulled from the literature, proving that correction is a feature of science, not a bug [cite: 21].

## FAQ 3: How Can We Predict Replicability Before Doing the Study?

Conducting high-quality replication studies is heavily resource-intensive, requiring immense funding, laboratory time, and researcher effort. It is practically impossible to replicate every published claim. To maximize the efficiency of the scientific enterprise, researchers need robust triage mechanisms to predict which studies are structurally sound and which require urgent, independent verification. Over the last decade, the methodology for doing this has evolved dramatically, shifting from crowd-sourced human prediction markets to advanced Artificial Intelligence frameworks.

### The Power of Human Forecasting: The DARPA SCORE Project

One of the most ambitious and comprehensive metascience initiatives to date is the Systematizing Confidence in Open Research and Evidence (SCORE) program, funded by the U.S. Defense Advanced Research Projects Agency (DARPA) [cite: 3, 13]. The Department of Defense heavily leverages social and behavioral science to design plans, guide investments, and build models of human systems. Recognizing the real-world implications of relying on flawed studies, the SCORE program aimed to assign quantitative, explainable "confidence scores" to social and behavioral science claims to help government consumers instantly understand the reliability of the research they utilize [cite: 3, 6].

To achieve this, researchers deployed large-scale prediction markets and expert surveys. In a prediction market, scientists act as traders, placing bets using real or virtual currency on whether a specific hypothesis will successfully replicate [cite: 6, 27]. The SCORE project sampled thousands of claims from 60 prominent journals spanning a ten-year publication period from 2009 to 2018 [cite: 6, 13]. The results of these markets were revelatory: human forecasters achieved a highly impressive classification accuracy of up to 73% (and 68% in parallel forecasting teams), correctly distinguishing between claims that would subsequently pass or fail independent replication tests [cite: 28, 29]. 

Interestingly, these human markets revealed deep-seated, collective perceptions regarding the methodological rigor of different academic disciplines. Forecasters predicted the highest replication rates for the field of economics (averaging an expected 58% success rate) and the lowest for psychology and educational sciences (42% success rate) [cite: 6, 14, 30]. Furthermore, the markets anticipated a distinct upward trend in replication success over time, predicting a rise from 43% in the 2009/2010 cohort to 55% in the 2017/2018 cohort. This expectation likely reflects the scientific community's awareness that recent methodological changes, such as the widespread adoption of open science practices, are having a tangible, positive impact on replication rates [cite: 6, 14]. 

### The Algorithmic Frontier: AI Predictors and the "AI Scientist" (2024-2026)

While human prediction markets are remarkably accurate, they are incredibly difficult to scale. Eliciting forecasts requires vast amounts of expert time, complex coordination, and significant financial overhead, often taking up to a year to run for a small batch of papers [cite: 28, 29]. To solve this critical bottleneck, the metascience community has increasingly turned to Large Language Models (LLMs) and sophisticated machine learning classifiers to assess research credibility instantaneously [cite: 10].

**The Predicting Replicability Challenge:** Motivated by the strain on scientific publishing—where scientific output has risen from 4 million publications in 2000 to over 10 million in 2024—the Center for Open Science launched the Predicting Replicability Challenge. This public competition tasks AI developers with evaluating a held-out set of social-behavioral claims and assigning a confidence score between 0 and 1 representing the likelihood of successful replication [cite: 7, 10]. In the initial round, early algorithms struggled to outperform a naive baseline (a Brier score of 0.25, which represents guessing 0.5 for every claim). However, by the second and third rounds in 2025 and 2026, the best-performing models began effectively discriminating between successful and unsuccessful outcomes [cite: 7]. The algorithms demonstrated an ability to assign higher confidence scores to claims that empirically replicated, proving that automated prediction represents a viable, scalable solution to the challenge of research evaluation [cite: 7]. 



**The SciPredict Benchmark and the Calibration Gap:** Despite these rapid advancements, the deployment of AI predictors is not without profound limitations. The highly rigorous 2025/2026 *SciPredict* benchmark evaluated state-of-the-art frontier LLMs (such as Claude Opus, GPT-5 variants, and Gemini Pro) across 405 expert-curated experimental prediction tasks directly derived from recently published empirical studies in physics, chemistry, and biology [cite: 31, 32]. The evaluations revealed that frontier models achieved prediction accuracies ranging from 14% to 26%, which only slightly exceeded or matched the human expert baseline of roughly 20% [cite: 31, 32]. 

More concerningly, the study identified a fatal "calibration gap" in artificial intelligence. When human scientists rate an experiment's outcome as highly predictable without needing physical experimentation, their actual forecasting accuracy scales rapidly up to 80%. In stark contrast, LLMs completely fail to distinguish between reliable and unreliable predictions, achieving roughly 20% accuracy regardless of how confident the algorithm claims to be [cite: 31, 32].

[image delta #1, 0 bytes]

 This lack of self-awareness means that while AI can offer scalable triage, deploying it as a sole arbiter in real-world scientific experimentation pipelines remains untrustworthy. Superhuman performance in experimental science requires not just better predictions, but a better awareness of prediction reliability [cite: 31, 32].

**The "AI Scientist" Paradigm:** Metascience is rapidly moving beyond mere prediction toward full generative automation. By 2026, highly advanced frameworks conceptualized as the "AI Scientist" (such as AutoSOTA, Lila, and AstroAgents) were engineered to automate the entire scientific lifecycle [cite: 33, 34, 35]. These autonomous, closed-loop agentic systems can conduct massive literature reviews, generate novel testable hypotheses, write complex experimental code, execute computational analyses, and autonomously draft full academic manuscripts ready for peer review [cite: 34, 36]. While this promises to dramatically accelerate scientific discovery—pushing fields toward a state of "frictionless reproducibility"—it simultaneously creates a terrifying structural vulnerability. If AI expands the production of scientific claims faster than the community expands its capacity to check them, it results in a massive "verification gap" [cite: 34, 37]. The supply of generated results will vastly outpace the systems that can verify whether those results are actually reproducible, necessitating that the very same LLM capabilities used to accelerate production be immediately redirected toward scalable verification [cite: 34].

## FAQ 4: How is the Global Scientific Community Fixing This?

To combat the reproducibility crisis and manage the influx of AI-generated research, a globally diverse coalition of scientists, funders, and policymakers is constructing a massive, interconnected infrastructure dedicated to Open Science. The goal is to move beyond post-hoc prediction and build systems that structurally prevent fragility.

**The European Open Science Cloud (EOSC):** The European Commission has heavily invested in the EOSC, a federated environment designed to provide a "Web of FAIR (Findable, Accessible, Interoperable, and Reusable) Data and Services" [cite: 38, 39]. By ensuring that researchers across the continent utilize standardized digital platforms and machine-actionable data, the EOSC directly tackles the isolated data silos that historically prevented independent verification and replication [cite: 38, 40]. Within this ecosystem, projects like TIER2 are developing practical reproducibility tools, such as automated checklists and reproducibility monitoring dashboards, that integrate directly into the EOSC, establishing reproducibility as a cornerstone of European research infrastructure [cite: 41, 42].

**The African Reproducibility Network (AREN):** In the Global South, grassroots movements like AREN are bridging crucial gaps in research culture and preventing systemic inequalities. Led by advocates like Emmanuel Boakye, AREN connects over 400 members across 32 African countries, training local researchers and establishing communities of practice to drive open science adoption [cite: 43, 44, 45]. A major challenge in global science is "data colonialism," where researchers in the Global South are relegated to analyzing datasets generated by the Global North rather than leading their own well-funded collections [cite: 5, 46]. By partnering with international platforms like protocols.io through the Research4Life initiative, AREN removes structural and financial barriers, allowing African scientists to create, manage, and share reproducible research methodologies on equal footing, drastically increasing regional scientific output and visibility [cite: 43, 46].

**Targeted Replication Funding:** Historically, researchers struggled to secure funding to conduct replications, as grant agencies overwhelmingly favored novel discoveries. This paradigm is shifting. Initiatives like the Replicability Project: Health Behavior (RPHB), organized by the Center for Open Science, are systematically funding up to 60 replications of empirical health studies that impact public policy [cite: 47]. Similarly, Open Science NL, part of the Dutch Research Council (NWO), has dedicated millions of euros specifically to fund replication research, establishing the world's first funding instrument explicitly focused on reproduction [cite: 48, 49].

**The World Conference on Research Integrity (WCRI):** The evolving standards of global science are debated and formalized at forums like the 2026 WCRI in Vancouver, Canada. The modern metascience agenda extends far beyond simple statistical fixes. The 2026 conference explicitly addresses the tension between open science and national research security, exploring how to maintain transparency without compromising sensitive data [cite: 50, 51]. Furthermore, it highlights the integration of Indigenous Knowledge Systems into global research ethics, demanding genuine engagement with community-based research practices [cite: 50, 52]. Crucially, the conference dedicates significant attention to the dual nature of AI: utilizing it to detect manipulation (such as paper mills and fabricated bioimaging) while simultaneously managing the risks of AI-generated fraudulent outputs [cite: 46, 52].

## FAQ 5: How Should a Non-Scientist Read Science Journalism?

When the general public encounters a scientific finding in the daily news, they are almost exclusively subjected to a highly filtered, simplified, and sometimes heavily sensationalized version of the original literature. Science journalists often operate under incredibly tight deadlines, and the commercial pressure to highlight "breakthroughs," "game-changers," and "miracles" can result in hyperbolic reporting that distorts the actual underlying data [cite: 53]. Therefore, non-scientists must develop a practical mental framework to evaluate the credibility of reported science before altering their lifestyle, diet, or beliefs.

### The Mental Checklist for Reading Science News

To navigate the daily flood of science reporting, readers should intuitively apply the following evaluative steps:

First, locate the sample size of the study. A study claiming a revolutionary health benefit based on a sample of 12 individuals is highly unlikely to generalize to the broader population. Small sample sizes are breeding grounds for statistical flukes. Larger, more diverse sample sizes are an absolute prerequisite for reliability [cite: 53, 54].

Second, differentiate between observational and interventional research. Did the researchers simply observe a population (e.g., "people who drink coffee live longer"), which only proves correlation? Or did they conduct a randomized controlled trial altering a specific variable (e.g., "we gave half the group coffee and the other half a placebo"), which can imply actual causation? Confusing correlation with causation is the most common error in science journalism [cite: 53, 55].

Third, meticulously evaluate absolute versus relative risk. If a headline claims a new food additive "doubles your risk of disease," check the absolute baseline numbers. If the base risk is 1 in 10,000, doubling it to 2 in 10,000 represents a negligible absolute threat to an individual, despite the terrifying "100% increase" framing [cite: 54]. 

Fourth, identify the format of the source material. Is the news based on a peer-reviewed article in a reputable journal, or is it based on a "preprint" (a paper uploaded online but not yet vetted by peers) or, worse, a corporate press release? While preprints accelerate the speed of science, they are preliminary and require much heavier skepticism [cite: 56, 57]. 

Finally, look for the preponderance of evidence. Scientific truth is established through gradual consensus, not a single outlier study. Responsible journalism will contextualize a new finding against previous research. In science news, it is entirely appropriate if there is no typical journalistic "balance" where each side of a question is given equal weight. Giving equal credence to overwhelming scientific consensus and fringe, unsupported theories (such as anti-vaccination claims) represents a failure of science reporting [cite: 21, 54, 57]. Extraordinary claims require extraordinary evidence; if a study makes a massive claim based on a small sample, the article must acknowledge this caveat [cite: 54].

### The Replicability Radar: Green Flags and Red Flags

To streamline the evaluation of both primary literature and the journalism covering it, the following table contrasts the hallmarks of robust, replicable research against the structural warning signs of fragile, untrustworthy claims.

| Evaluative Dimension | 🟢 Green Flags (Indicators of High Replicability) | 🔴 Red Flags (Indicators of Low Replicability) |
| :--- | :--- | :--- |
| **Methodological Transparency** | The study is formally "preregistered," meaning the authors publicly committed to their hypothesis, methods, and analysis plan before collecting any data, preventing post-hoc manipulation. | Methods are vaguely described, or the authors rely on hidden analytical flexibility to find a result. |
| **Data Accessibility** | The raw datasets and analytical computational code are openly available in public repositories (e.g., OSF, GitHub) allowing anyone to verify the math. | Data is proprietary, hidden behind corporate NDAs, or the authors refuse to share it "upon reasonable request," shielding it from scrutiny. |
| **Language and Claims** | The reporting uses cautious, probabilistic language. It actively acknowledges study limitations, context-dependence, and inherent statistical uncertainty. | The reporting relies on sensationalized language: "Miracle," "Breakthrough," "Game-changer," or states that a single study "Proves" a massive theory. |
| **Sample Size & Power** | The study utilizes a large, highly diverse sample size relative to the effect being measured, providing high statistical power and generalizability. | The study relies on a tiny, homogenous sample (e.g., exclusively relying on Western college undergraduates) to make sweeping claims about universal human nature. |
| **Publication Venue** | Published in a reputable, highly-cited journal that mandates rigorous data-sharing and adherence to Open Science protocols. | Published in a known "pay-to-publish" predatory journal with no discernible peer review or quality control mechanism. |
| **Funding & Conflicts** | Discloses all funding sources clearly and explicitly indicates that funders had absolutely no role in study design or data analysis. | Funded by an entity with a direct financial stake in a positive outcome, with no transparent firewalls separating the sponsor from the scientists. |

## Bottom Line Conclusion

The ultimate strength of the scientific method is not its infallibility, but its relentless, built-in capacity for self-correction. The replication crisis, while initially a source of profound institutional anxiety, has successfully catalyzed a golden era of metascience and methodological reform. Today, determining whether a research finding is likely to replicate no longer requires blind faith in the authority of a journal's title. By critically evaluating methodological green flags—such as preregistration, open data, and adequate sample sizes—and by leveraging the combined predictive power of human forecasting markets and rapidly emerging AI assessment tools, the global scientific community is radically improving the reliability of its output. Simultaneously, global infrastructures are being built to ensure that these rigorous standards are accessible equitably across all regions. For the non-scientist, adopting a skeptical, checklist-driven approach to consuming science journalism ensures that personal health, financial, and policy decisions are grounded in robust, durable evidence rather than fleeting statistical noise.

**Sources:**
1. [teaguehopkins.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFNFLO-RE1zpBWU9FyRCJvTe_Yk766jZiW2FxrHvj_g2yc1CF2f7TfXTn3BQ8n-iTAlRFMoXBqwOySi79V6uvddmo5JCEfQPGaauM8GsXcHwROr-fcNocPPoXzga0xVLetTvSSNm2KkMY38BHmNYpS2brxy0v9Psn8TpEPzSDBQynwH_zmffCSyNx5zJJa_IJ2gDTTlxUwvMGnCpfqJWw==)
2. [sciencedaily.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQETI9iI0gzYGqe0XIxdYNQDPZ1tRwo8REEpxLGlyqZ7_NCjb7ANzMMEgcD3yx9v9FzJvj_FJpWGCxBWq0s3EWf6ZFkVzz1jowmRM51GvbExvr5cJ6en-pO-Yy4zIUdqXbV3DxwvrjaIlRrq34u10jnKRlnyBw==)
3. [darpa.mil](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEgF8hv_Zbp2sRgEZXqbiZbXPBk1mlfP4G1gERnjwTX2VQqqNh_dto26gXy8tAiJGTBYDuz57139bSuc3myz9dy_6ozxBlgg7zrK1IzO1mGJLBrgT2MuWs22bH3fIz0hY6x0LSaOtJ8NkoXndckyVCUy4YcynTwbfpA7myc7wBXtDGuPH2MOIVocCRmv-X6FLlhnkk2)
4. [opusproject.eu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEjf_Lst7W6pMC_rNZ0mtRXG7q5OD3krnh7hLEcxu__pcVA9M5jomM1h5M-FFUCmqM0CFNqA09vyy0NSE8N_voEeeiaHwmpO2GNJT_LBqH76Si8Cmhy6p5SMh8xli-MWct1M6ozOeSNDsT2qzpJLY_-Aff34smDGCT_HpTZZQjD-j05TQ63W6mACiC2rzK33HfMBXSmjjTbpD19Xw==)
5. [nih.gov](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQG462fOtLMWzc3RvrkF2Tkm_UWZ4w6cg4Xb0vA3F7R9auGVs7r5x4A1D5bIfFsb7EizsMDDbHwHqHqmVmug_wSeQ664PwjDlSBKfwxJaCEbqNX9Rb24yT1OWfpCRaSHxc2aaCcS7c7Y5g==)
6. [royalsocietypublishing.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEPSKnlIA9Ct7-5ja9wtlwMhF9RGb1EW4pfigxREYJ7_hYN6UsUwrYhqXdKztNO-bpcrwoY-wYD_lNJN7L95TFnqLeU2v317IXSwK06h51IsvJ4cYU4ugmMaGDR2ftGmEO_I3E0p2ySulEwUWHM2-j03QqqLkiB_wXqn3vAgtjmqoOtF3SlTpj6IfkAzoZI38lzOIUB0B2QwcE0svSNaqQb_F8V48o=)
7. [cos.io](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGH0wN-PgJMqAWZQO4DFwlusNL9prz-ArsbOPN9U9wLIbHddLNJa09pvp2iT5WT1nK_tv31p2VpQ26Vw-rikFZTy_TKpfufyyO3UTes6MlWmAuehPlwpwN6AgM5py3lz098azJy2pqc8R-S2JrkB1cAb7c_4CXeWQVDbhY8DCOK9A==)
8. [cos.io](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHbGYkgp4ZVNN-q8drk2jH23pyz1_u5wtOmY8Aanh5gB1Gw_KZwO4w6BdbXAmHSDDUpCKotYpUT-cUOJ88lav7smJ6udbfM-ApVEbgCdYnv_F4cPUSl2UVq2guHn7drRhuD)
9. [open-science-conference.eu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGRvGFO1vlwT538QJNNz-Oke4YMV2mjPgtENxf1J7mCPXfSm1II8bp1Z9UhqdAMEVdFAyjE5mjByiT1Oh63S48U8DgAf9NodPve_qgVby39C4KOnd0Vc-I9QvyLFawtzU80yVQZtdYJq9WZvw==)
10. [cos.io](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGGHFNNnOyBFKGpA21jBuYdyz-SCiD995zV7YCjADqpxliDfe6O1Innob9I57uT9Y-D_k-ljpV9Ow_KXQekDCkxtswAFDLBv_VG31ISFz5omVEFmFd9JODNNpXqKQY115txoONQVcjkDLiOIQ==)
11. [aethermug.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHy1_e93bIJ9f7ycDw-WKEyLbROTeGheb0P8h01UPqoclYu701ewg788NJz2K4MXD52vZmlGNB4ytqpF3llVIiKe1XxzlsNWPj1Wng43Z5y7DIXzxTAASXIwQk6HD7anebQsRwP2YkewuUaRS1ZI75vHtSA41fpkQ0Z7d15ov9Kowh-hG2sts0niAxjQ-T2Vh80rw==)
12. [uscourts.gov](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEirGL8FleA9hok2F67Q0ELq-5mIQQsTKhp2yFgd0ObsPW3oAoY1rV7U7LcH4ygyOXaBdgKBdp6vQXEq7v3ecneL8CFejm-ZBarU0S6mlf0DRVR518TCwghAuptoagK3nMzM_RaWk48JntOg1IUAQIbEn4DK_IBAobyY_JdErxafiUe5pnxhs6oirOVxl-f5rXXOPa9Ptw_QFkUtk7Bl8yv_WQTMlpBQp1o4A==)
13. [forbes.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGTVPAdx9nNChszJgwfOGUrgvxyXo05_qQvuorhbFh4r5oZ8ICWEyxUrRWpM4A5-kctLQV4A1Ewgs33hvLAjD7oMws-FV0Ck5MPiU57jNQbC1IZCwLKVqCsTsjbYE8t0PBrj9yYtJ-0VrQpfA97UpuSPEDoio2N9FiL3hE9em-K2sOpL-IX5gR8phDa_e24H887x-o4JUj4PmYm7jG0Dyc91gC6KG_oMGGgCzZnsfpBXTMbL4BGzja5lgxU)
14. [zenodo.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEr0xxdJnwHyxxND_ArC7SuW3zCV4pOqmyG_WWGNmtPEjK0rMyCAD7DenwfuHDgYwkv0AES44Z-wqKvyiHJJIQqR93spkQzJKqqNoXp8O8cWdejKOBTuKzXLfSp5LLPtWlAB1Xe2p-yGy7iyPF--LbV2EkvzoL4E48-Bp5hhf_uhMdjDM1784s=)
15. [columbia.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFYzN0EaWl7tKtNxrLN2Rq_pAGN_l1z9ce1UfZJAwFUK4JhI5K-lFvJCtszaCZQR2nnFTf7DZJWFU7mYxU4lZy6-RBQmYrEgw7v5xJIeDCK1lAkRU7i1WpmxmAMppbZphWDq-vGf7zLy8fymZYnoIGbgjKxdTgK8QBTQ_nBqEI-DPJFGpZlTWSLTVlY7qubE3M8XCpv-SPJwTM-63l-nSlRItaB4ZianA3GJTNdPHqM5U2OceUg3v-2A2JWNPkk0yrv2kctBtU4tciMRNdCOxra8Iq4)
16. [bps.org.uk](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEw6qr3LYEtz8cvYe7JZomoRjOaY1_o4elAC2vEmOZRsbOKG0m8tcIEBpaN78afd1BoBrZwkQKP72yYwUc2GUoMVCaYU2s5r0Tok0Tpf_yOo2_XSo2F-jNpQN0oIC3XdGNlKgdmaXYuiilpkgMWPfLZKzgtFcaHeE6LbgtLg2z6_mHfyepmSjL6PQjQQfMzYcnDK8V4g7QRAsw=)
17. [enago.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEUs7p4FL9_Sp9jIwD8ld6vREQGTZRwTu1VFb0m0BW3DLfxZnB9dbREno2hziSVg5nD0tjyUd_xkQyb105J7ob3m9VyboaUBanvkbl_st7fZKIgdnyBuL_s6-9jfC6UD1KRhPGkutkZtnIB8j4YZv1glRa14tpnT-BV63y4mp8xlljmug7tP95SP66bWpPk9MYXQIWnBPdpdtC70y5KoPo-c0qaZXK7eZW-W8p2aw==)
18. [stackexchange.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGMVJi-T9SSXKLioFD1EO6MwIdY17EorEN83klJBMtdXU0X0GvkpKkfzzi8-g6jl8ZyJznC_6SWCOuVSyhHcHqZ8kpVhMoau67J75fZd1ZAH-BlqdHI3Efo5GFOq9khPqTGrC6Ot_9j6GjMcVnQsOUPOKUiPDUuT596Hjk7gZHYr4q3loVOeRGuaH0n4GbQm4HX4CHVKyoWmIgYZsmXn4kWaAH_YvYLHAJKiAM5dZx_DEsOBdVgEzg=)
19. [sspnet.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHkcA70cZVIquTT_ooaKAdB2kwwL9HBqxePJaoFsRGL0kR_4GPhHV45e1JyTrGKQ2-ZBaCXziYGMMUIHrZtOKu5qVMgQKOGAYvqW716PwQF6NkLmDILeCCv3dRTg-7TBD8UcqWPzSuu5a8lxo5jFvC5tkW3kt1XAHYrN710PhwLxpXPh3fX83VMFgyIyM_iYvirub2a9ys3dPyib4hBJnKRZ2yxLb60VnTlvc-IgC2iBS-U0E-UJhirGBxaA-PRzi-opvXSidfj2Z_4K01C859LW3tpz8BWdody)
20. [thequantasticjournal.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQF2KFI0DxLNZTmQSkIO4QYMEO0_7FzXhWBml07YF7Zg6kKVkxo9BBl6YYmCky-z2I1GK7S5iUeG6D-2LcUXpvVdVTa39mVnyxFOXpm2_7uSSZjol3Mz5UcB8L1xoSb9qjPMetWiDDesHjh4HCA6dAqppaB3wK_tSHesLOcP_8w4RpI_RbZOZxGSpsAbNI2dDtsOUYHY1jUyMaNzBw==)
21. [scicomm.be](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEW8QY1CQBPEn-4_1KNB9ooiCZP6FrW0n4uradfCVCYMRhvHAldAk39YPalQAafUO_ST6CJNdE6XdkTfiuGqrd5vOrbL_H0LnkUH2btErn8ILeiKzEU5Y2HRxW1Rlq9Wvi9GHytKq8JI8q9ITeUgi4qJfDP6bcf1LvvvARuhLMZpRSu9LmSETGmFZAWY8lgUHV9BUE=)
22. [frontiersin.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEqcg01lj_qG-m7Os_TrwZMcEFPVctUHF0dxl7YSYo7sg7JFKcWSEXpZImsQEEKYnOIAjUg-0PNiUmfVzD2SQJ44oGplrpWq_iNi0dBkFuCXCkStrCEdkvlqFo1txgNxXVbuD9qH9Nvp6kc-etg5HIP5XLCivqoYRNGb6RkATsqdquBNMPX1TIR-DMD0_U-fyB5LypWKha-rAD044okdYNUgA==)
23. [acfe.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHLUDhRNUBlm4ZIxKORWJkZQl1WBA80ilfXX0ls2SNPvKKbesTHR2fgSsk-4UYt9SrK91L28Gduk31e0BGtPMsSXF7K1Xoe9CooajhT3h3E23sch10fha7IwpWDatJ9f1G87l38tocnXm8=)
24. [seldenfox.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFdEwV-6scBkbEbbmlfxu3PCf5vLdz8a7NrLOSmeWi3iQbqLkmBjiOPpc-kwUZyp3zZaK6PPbd3Kmyl3qusPFj_iOAbZusNLkMUyTIUHulfvzYYUM9Ge_fYpLaTRBaCeJ0M6YxtnewgyM3M_Rxh4NLRZ-Q9xa67WjYyowNLNuYdUj6HVA4rqwW7Afw=)
25. [pkfod.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEubKuz23mChW-LZlTcuBAkKDh7BhZwpyy_6kzLt8O2LLgDD_7bzm3rn4dMubz3hnTZAha6PSFbbFVAxZdsThQB_pNLplsOU5ohu9cY7N9Cw0JlCZ51wGK8Mt2qXIdHHXIkQ93n0is4E61Hvwa4CRuWW1yvIaybHZ88UrD2jaLn)
26. [nih.gov](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGLwxlv8jrkA10PhwuJvlJ-urWtcp-PFIunDg-DH5I4i-AS5JMh3ZVnVRp9QKhTeZWzgb-FMEetR4sxJngXI8h4KPy1ths03Xrm-Kq1Hdx6p5GFE8rlbht4SDvPAtEHCBgkEJCKLvHI)
27. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGVlhH6KuuYQ5eGDbAbPflhD8Fk_ITp4K04Se2jCq2yfI2Itn_e5ALQLTnF_knBTQXTXOJ7_3n8pOOIZBpKkDZ3_rVjyO29Y1ZLsLKSAmZhT--Wfxna9237MVCs_FKPusg3cngV3cUx1stmtPq_k0Nvj2UL1o5tdhkaifrdASG6-jGY5EvoTNZmYtGP1HnjLsleRHxNYbwJx8yqkSFrewQ1tG2-d84G1kBSNUK5PuRJwxbqUS0M6bNXXqtUQ53aUP-2tQNRvWdg6HgdTQFEm7Tupn8PV1rb_nMVgFd3USUv-eJNr9y7cRCe858moRIj8el9gvFRiZ8=)
28. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQH-bWII9MLrQE_7uw4ls87vU4lk1jqFMCI5VpydHj-roIXeMCQw9bWY96nAZhLBR7oIyZz5DQXtrN7_4On0Jo_L9dXRcXZMxYvegl1xeYemXrahsIRTRwBu9N9QOvwKw4nopXF33Fkk01mGcYmwDFEgR7hc7arvdi-amT3vHJViPQaMkD6NcApncwhHTXHgcz_s6M5wZITYoWrfG-2vIdgXm-IlgAxkpVVFtTIx0KY9TFaQLlqsRM4uMf1anqTBfFhSgg1LYNCkGNDHRs7zPjIGn9w78nlRJq_z1KK9sc2JBf_yMr50693gZg==)
29. [pnas.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHi28LMIIApyeW8uU2-7QO-qwffzQR5Iv3EuUV5X06jJsCuYJ9G7tECdmxQzcgm7638bSkJ5YPCp6s0mjkjxsiwhr2rXGrHRH1DJf-dMaXLc0XLZiuOypN3ZYrxglRGE6R3xwZlQAQ=)
30. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHWqJLdCYhswoMkmrScimm1QZa5xWn5ZTyWAdlNO9SyQ9pAQZ_SsKfLSlKgqOi3Ew0F7nWEpnYkYtgHnSIBcJAczPtwXhSdQFy92kYLlN9QIrI98wFh32t_JzD06IXkAHAnLNey2NOSjKgeqKzXoVOlI_BmyoRuH2TeYeNjWMgdZ7LCqjASQ82qAetU6K9z13qd7gzY0SfspVEDJ-BgzEYLL-_evm1NNVWg3tpsvBtny1rK6ubDHtp92JcftWNA7aY3mae1_d-9BZUGoyQEpoEa1ml0ShpJ_RX3pQ==)
31. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFQNCZIa4-bMjQKf-HntTYVot-xLijpIAato533NVvENOQiR3pNMIUEFPYVX9bSF0p74oS5ucjKuXVYYtXzQWtZJZL27Vc4Jb65-NQtTYBfE0BYeQsrQVpk2Q==)
32. [openreview.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHt46Z42UKmPEf6Ff3mvD6rPNcOR3_VC5ymw7r7QV96qw_q9z6MGmwSBL6MvUzzKbFr1PCgedrm1pf3d3dN_QJ1mn6XOt1XLI3j4P5jRs3PnNZNvcPDhWBsN9aAN_di)
33. [preprints.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEO0QKBJ2w0krtixuIm-D1iMJiZckOCjr3TqkmOQmT0ZxFsHtS04MubHa9CP2Mi2cJE-Lied-veSezarwt0DYTWDVFhzsPUJr92qZa5B_5mUNXTGcxPREr4p2FDMoiCoQWw8_DcIVYDsxAIE9P2wbsCvDjHAbTzmrvTN9qBmd-ZEWmztu9Dj2BDYXZD3xyoFCHd)
34. [osf.io](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFpVPVlAnG0_hZCNf41DIT2OKrJDAAVeK2NQl9Tv6RQUw8ohFGZx03sdMr6oxaq6bUQssN1-t1rT9suQBR5xCDk-uFLhkK1lCcpLi8nTP3914u8s0W8ya7lu-YdAo7lVHSE5Hwzcahp)
35. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFcc7tOahdfO5hIAroDBOO54H3zvRfQTXQuMeCnDjcpJn-u5X1J2-EOdp5J8yjui37wVjBD6wzqop-Lb3DWrVH4a484QTQWJs1tQh6_Z-_XctS-OuMf4q-U7w==)
36. [aimultiple.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQG8x1RbWhVNldKEkkUhV6n8TjrLYrzjtP3iiVHEJp1CR2PmBY0FjcxN1dfSWxxaiEB8ySbh345cPqVJKqiKDszj84I5-ZjofGhyy5-6ACG1XqtSa9-ZjL15HQ==)
37. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHlO93Yr_LPYcHG_WyqSjVxDFlObg57tuU0zZoouffn9__b9yXMB2gcIDZdMYJ-tPxISMEj6xR0VqkXcYX6saHDxoHxYTDp78FThePdB40M9N4jFdg3XQ==)
38. [europa.eu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHzkgECzCZyOnYe2PN7rzccaVh4aEmCNKNdBe65uzGla4PT5XX9Pjcq9isiIW2j0LK8-n11_hBkjSm8qFLQSNjST0cb-RVJXEu0IoWTDFOF52GpHS5zjfNZsozyXYJh4_64sMlJ1TRzExrUU93WCtW3HVIifA5OVXnWvrIzcNcDiBbLy49udB_ei7KtCL8GnkzeO3PFIO9e7Wem9tlKzOZAcwmyMWZ2-gu1qQUPuqyqOrpZ7_XHaIa_06jY2SiX7kAEi6i_nSUy44thZU7LUNyTxg==)
39. [oecd.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGPNDrYr6PojWIul0MIKJWxUO8bpeGPHVLfG-550Luuj13yiqxYw289VWk8yd8grXVz2VIUst9cvk18ivplSjIHTSLt2P89T3OZAC8tjK_KNNOSzGhIk8qaEN3M8hhyQO6ZEdEeIBDR8iz5-LsBmor9Jg159uR-X2lDuvGXh3vHp3uT41o3OXQawxau5zCGnRO9DoMXW5T2-zhirSGIgLaZxfjwT0uZcD89dDm3HvjkT5oZ3sbuHC5GvcSUGdLUE0_DG9UJ2kOx6CW08_wnoGO1ZgWzSj7G2pnuGqY68_V-0shDhDN8R41SZwIr8w9g)
40. [wideradvance.eu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFJvzJVvVMae-0PsdwWdSeH1DETQwfot_qxRywGJVaSzTM24ILC1W9UpvZWBbJO1zTIAC4x9AxHO8vvS-vVG6P83XdHlLAE7R1gszqlVm9z7TuBKX6jcogaxm4axyRz7JbF)
41. [openaire.eu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGnKueoVKczzgIuKadJuAvAwTAS2pxL6QNWH88lpJzYup4JwAmknEzTosxaUPJnHRipTYO1_p5CU7TiqK65uUdbf1Ey2K8ltLMJ8MCrlpeFpAdna3WUxDus76ZF4UOWlj_FJA0N83yq0j-vFFBjOaHcylcn4B18qdF2IdOwZeCp0ufxJoaZAW4LnHyo9jURICDiGio=)
42. [tier2-project.eu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHssintrWrJ819UjnnIZq0yuTe7jtb4DwrStx49yo0U9J-GsxSe4CxJaIyGx4Uo_vbrRF2v31_Yg_7ruz1GLtKNZQo4nmAzfDu9OFt08exG2S8fbLOveI1RxDCpejgqUz6zNQFK8q5cJKp8rI7r4V3f6CGMhlC446bETyNTbpwDeE04bLx3l2X58lvxoqxzULY=)
43. [springernature.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEEgqWTWlLN9Wf1Df_nDus6Cf40cBoZ37uF2EcxkGJ0rUIFQxVP5izsfqL5XG_QzGfN0xK8lssaF4BGNf2G0A_DnQ0luA_aXHn9DAkd54Q-Qt6hBI0Ixqhw4vxnl-9o8ABm9ymaB9qT5dA-IOX92ENpGvsXWfKuZFbh-qT19pyP7AhJwaZdbmbLkzO6BcuuvRLYsFpnbmkAUPgqSYOVIKgmlvlBtx1awtGdz0biICk7gsGUE30c)
44. [tcc-africa.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGlhjG1P7ULsjBpeES3kQ_fdpxWsnFlLuSl7iof7NeyQ9tFxCsWHtiheaqGOvkUs3Awfmf2CXDUEUZESXXFCTxeHeo4JQjMI6dQ8QMnVu80acBjDveifpSE9zCUdQqghOy0E0Ts-zMcr_Fl9MFIDgQfgrdqIou9-ZkdoqVaKanf9ecwRTfGmF-S6V0Tr_EDrjBudItBDsjK5M5K_9vj9mHY8f7CB6XcHqOhztKG0MOzeJPYWc0GENWs)
45. [africanrn.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHrf-BG7sFKJZvxTStEjda4N082oQcnk0f5SSp1ZFrpcrT5vvVVZWJGYmnUtTmGvYpG2Cb_tK_yFHja8ULE7MaBwBNNUTfTEavWx8eTFShGW6ACw5R2VXhojDWz1vKghoSMiKgxPBifDmN8CBiL5JDyoQ==)
46. [scienceintegritydigest.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEeo6tXFROyDxMM94HV9nYR6mePaN-3PNH8icAksTo9auq39dkWOGVLpUUoMLNnQA_x0DWuqlz5n_5egip3z_hGyVQfxDjqvsut_2_NJzMLnslYN3TNDupKniAt29L_VKGgRjrY_dy17qMFZl8UjTXGce_AvFU4cT-bAs4nlC_alBk-JMsT9GR6rY_csJR_mzpwlk-OQuF4OjhtCBj_F8-S488E)
47. [cos.io](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFyH34jAvJ474cwbdDiZmztqVH8Vpk73-XuNbEBcLZ2ZLPFU4v8IKMXJDbbA29OtTrQnrvkU8ueGiPskj-KddC12iAar88Td5EhHQ6xRM4vBYH9fqmWt5ITO9EkmyZO0MmTPBAwjj4ETO0vfDhcL_uTbpp15u9v_Vb4-IbY1mzTXBLlCQ==)
48. [grios.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQG7P50pAo4Ff9eFY0yGczSKzzb-NlaNK57QO9-u8HoLHJ7xAeSulbFXrxJUBXi5SPIqi3Vk52MvGDSz9xK_IV1uRZTzkLb_2VSFZYUV)
49. [openscience.nl](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFYt9t4gXy2HMb3Ns6G2aIEfeLpeAOiiUalZ2eS6oLq2Esf0v697upJvESqlmrvZCG--d6h6OS33Jv0FGmQ6JnBOMVchKDQqvW1DAzmm_IlIwzYe2FscyGg7i1nxB7kCOU-AmbMN6pT_rUKZB2uMbZFia5SkPv-Gsvmyw==)
50. [wcri2026.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHak2JO8F7KxYVTxZNe2A1Zq01nYkfWDEW2btg36S1riVUamHyPyZtal_bUQwY__F5TSAOZacjP2ufnhH1_FE_yxwXqW4FYNq_INP0=)
51. [wcri2026.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGAD0hjZW9RpvAYe8YB3vrKBQ7tVEoUgnhx5UunKh2fhGr5LYUHEJ-Amyi4vTEDSPGDw8zoQuU5xuqkdqVe7xVNUi57yca02YHxMlpfJRHoXMkWlXGAC9nEzzGX5A==)
52. [imagetwin.ai](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEIrPfSZ6Xi2X60WxKd9fKfumd6xzWVEcQcnznXxaLMObdLO3JuXRKE3rm5JojCsRb7YZCOEf2P_DUNnEpJdMLCZeR0Kj7pHHuFWNx_EtEZMm_wkzBtHiZq_yc4D-i0UbDgNe8cizjYe8ICkxrCyVh_rg2kr14iNyitn702o5bsHltD3I__sDokuveHBbC0Npfhjw==)
53. [niemanlab.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHAUEv-hMHX5-9F1GdG2fmCY9TciqE6mMgEwNeKoshvHktQZS7ocPEu85qkATgonjv-gMtFwVDhN7oS2GqSO1BZ7M7P374WzjquS7NwkjZWFb4ui06r4edZ21TR0ZDEfQj1q4xF1L2785rcAiL_EpRUvAEngZdWvNgUjHglZ13cy8Fuws2uOkBhwo3Hi02Wl0fPYRABK0MY-6tgZ4-a_f3cLgFfcIwpLQt9nDfdwJjfd_kNbUDr)
54. [wordpress.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFisFGBp_QGaUSWvujKNY9FnKwvnQ4bDZxLKfecbL8lWT-N5tSyain5xkLZyMIYGe_Y1TCorazL6KbGMLvwvLRREXOG7Rd8-CmpvQOeup_uG9FIZvidkEsUWvlDkqxCJErTuHXI9Z26sJniqMPnhRt9wrvHzfyopG8rb79zDANXPbbe41LVOtOj0Mw3OMI=)
55. [scu.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFM3nIq1B9ze2ZGIvZHgvfHO5079hEf28pUafqxKVA42F3R9idCJV-_AjdH8yKCGpO7mwE5BOeUeTFJILpZZUkXV9wOZhJMOmPhoP79EYUS_vEbHIl4dc1Epf68xFQ0ygrpZO5t-Y1EZEmt5ywgaLYxCGNXAUvFz3XUpoKi01gvyr8OYu7hhjRRe448HK7xSsXGrmE=)
56. [reddit.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFvkATt0mNKjF8DjtOEUib6lBrujnbW0PW923KGq_1ugdsuvYxJfPF9eT0JkQwLAFrFPUnJ4Ig33MgB9HDMfgh7LAK65hEuE88c7eD-iNhl1XgkKn05dwJUR0LcaycBTO-vKPXUeE1hyD7KEk0GHerMM_AB38VVr0mGQbl948LMpe4taphnRYBx2t2yN0HMiB3-BgGDnncE9wxJm3M=)
57. [journoresources.org.uk](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFf2FKMlo5WkWETHSwaKPMYcvom4oy8J-ieIEYZ9cBM9HyP3sWPlTUNph8DwlG5ok423Ev4AYefOftmQA44URA2dWcGhdv8rJ78XfmJr0wWuPxW8jYGkipiN859BqavVMwxcNtnkBUPaWakz7evkR75io7bTN3gEfkPHOhh_dnCFy8FCldpgxnd0cU=)