# Introduction: The Data Wall The development of artificial intelligence (AI) models has relied extensively on large, diverse datasets to drive improvements in model accuracy, adaptability, and robustness. However, as AI models become more sophisticated, they encounter the “data wall” – a performance plateau that occurs when additional data no longer yields significant improvements. This document explores the data wall, detailing how it limits AI development and examining solutions to overcome this limitation. Through real-world metrics and examples, we illustrate the data wall's implications and outline future directions for data-driven AI development. *** ## **1.1. Introduction** The field of AI has witnessed rapid progress over the past decade, fueled largely by the availability of massive datasets and increased computational power. Models like OpenAI’s GPT-4 and Google’s BERT, both leveraging hundreds of billions of tokens from diverse text sources, have achieved remarkable language understanding capabilities. Similarly, models in computer vision, such as CLIP and DALL-E, are trained on billions of labeled images sourced from the web. According to OpenAI, its language models trained on datasets as large as 570GB of text from internet sources, equating to billions of words, demonstrate this scale (Brown et al., 2020). Despite these advancements, models have begun to encounter the data wall, where additional data fails to produce proportional gains in performance. The data wall arises due to constraints related to data redundancy, lack of diversity, insufficient contextual richness, and inherent biases in the data. Understanding and overcoming the data wall is essential for the continued development of robust, adaptable AI systems. *** ## **1.2. Defining the Data Wall** The data wall represents a point of diminishing returns where the addition of more data ceases to yield substantial improvements in AI model performance. This phenomenon is observed across different AI domains, from language processing to computer vision, and is influenced by several key factors: * **Data Redundancy**: Repetitive or similar data offers little new information, limiting the learning potential. * **Lack of Data Diversity**: Homogeneous datasets restrict a model's capacity to generalize across varied scenarios. * **Insufficient Contextual Information**: Datasets lacking real-world context and nuance impede model comprehension of complex tasks. * **Bias and Imbalance**: Datasets that reflect biases or have imbalanced representations reduce fairness and applicability. The data wall is often measured by evaluating model performance on benchmarks that require extrapolation and generalization beyond the training set. For example, adding data to improve BERT’s language comprehension led to diminishing improvements in accuracy beyond 300 billion words, a sign that additional data did not proportionally improve the model’s grasp of language nuance. *** ## **1.3. Causes of the Data Wall** ### **1.3.1. Data Redundancy and Overfitting** Redundancy occurs when datasets contain repetitive or similar samples, leading to overfitting rather than generalizable learning. Models may "memorize" repetitive information without gaining new insights, causing a performance plateau. For instance, in large-scale datasets like Common Crawl, which exceeds 800 terabytes in raw web data, much of the data includes redundant patterns that add limited value. A language model trained on such redundant data may produce diminishing gains in tasks that require a diverse vocabulary or nuanced contextual understanding. Overfitting exacerbates this issue, as models become overly specialized on specific data characteristics without effectively generalizing. This has been observed in vision models trained on popular datasets like ImageNet, which, while containing over 14 million images, lacks the variability seen in real-world environments (Deng et al., 2009). This redundancy reduces a model's capacity to generalize to unseen contexts or domains, signaling the data wall. ### **1.3.2. Limited Data Diversity** Data diversity is essential for robust model performance, as it exposes models to a wide range of scenarios. A lack of diversity results in poor generalization to different contexts, languages, or demographics. For example, OpenAI’s language models were shown to perform well in English but exhibit reduced performance in languages or dialects less represented in training data (Brown et al., 2020). Google’s multilingual BERT, trained on data from over 100 languages, encounters the data wall for low-resource languages with insufficient samples, resulting in diminished performance for these languages (Devlin et al., 2019). In vision, a similar challenge exists. Models like DALL-E, which generate images based on textual prompts, struggle to produce accurate representations of scenes outside typical internet-based imagery. This highlights the data wall in scenarios where models lack exposure to culturally specific, low-frequency, or contextually complex images, which are underrepresented in datasets primarily sourced from Western-centric media. ### **1.3.3. Absence of Contextual and Semantic Depth** A significant limitation for NLP models, in particular, is the lack of contextual and semantic depth in most datasets. Language models trained on internet data may lack situational, emotional, or historical context, which limits their understanding of more nuanced aspects of language. For example, sarcasm, cultural idioms, or layered meanings are challenging for models like GPT-3 to interpret accurately without the necessary contextual annotations. In reinforcement learning (RL) and NLP, models require context-rich training data to learn behaviors that reflect real-world situations. Reinforcement learning models trained on simulation data often reach the data wall because simulations lack the unpredictable and context-specific dynamics found in real environments, such as human interactions or environmental noise. *** ## **1.4. The Impact of the Data Wall on AI Development** The data wall limits AI model development in terms of accuracy, robustness, and scalability across tasks. The following sections explore specific metrics that highlight the data wall's impact on AI performance. ### **1.4.1. Accuracy and Robustness Limitations** **Metric**: *Accuracy Growth Rate* Studies have shown that accuracy improvements slow significantly as models approach the data wall. In language models, for example, Kaplan et al. (2020) found that doubling the dataset size beyond 300 billion tokens increased accuracy by less than 1%. Vision models encounter similar barriers; adding millions of new labeled images to datasets like ImageNet provides marginal accuracy gains, suggesting that new data fails to introduce novel learning opportunities. ### **1.4.2. Generalization Constraints** **Metric**: *Generalization Gap* The generalization gap, the difference between training and test performance, grows as models reach the data wall. For instance, computer vision models trained exclusively on synthetic datasets perform poorly on real-world images, showing a clear generalization issue. Language models trained on internet-sourced text struggle to handle domain-specific content, such as legal or medical terminology, due to limited domain diversity in their datasets. ### **1.4.3. Data Efficiency Decline** **Metric**: *Information Gain per Sample* Data efficiency, or the information gained per sample, declines sharply as models encounter the data wall. Kaplan et al. (2020) demonstrated that adding more data yielded diminishing returns in performance gains for large-scale models. This inefficiency increases computational costs and environmental impact, as more data requires more processing power without commensurate accuracy improvements. #### **References** * Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., ... & Amodei, D. (2020). Language models are few-shot learners. In *Advances in Neural Information Processing Systems*. * Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009). ImageNet: A large-scale hierarchical image database. In *Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)*, 248–255. * Kaplan, J., McCandlish, S., Henighan, T., Brown, T. B., Chess, # Human Generated Data AI models frequently encounter a significant challenge in development known as the "data wall," a performance plateau where further data increases yield minimal improvements in model accuracy, robustness, or generalization. This plateau is largely due to limitations in the quality, diversity, or contextual richness of existing datasets, which often fail to capture the complexity of real-world situations. Human-generated data is crucial for overcoming the data wall by providing depth, nuance, and context that models need to continue advancing in performance and adaptability (Halevy et al., 2009). Human-generated data, with rich contextual annotations, provides semantic depth that purely automated data cannot. In NLP, human-annotated datasets include additional metadata, such as emotional tone or intent, which enhances model understanding of language nuance. Human curators can also identify edge cases and biases, ensuring data diversity that automated data collection methods may overlook. Platforms like Snorkel have been developed to facilitate the generation of labeled data with minimal manual intervention, combining human and machine efforts to overcome the data wall (Ratner et al., 2020). This section examines the ways human-generated data addresses the data wall and discusses specific AI model requirements. ## **2.1. Overcoming the Data Wall with Human-Generated Data** The data wall manifests when models exhaust the potential of available datasets, particularly those that lack diversity, contextual richness, or ethical considerations. Human-generated data helps to overcome this barrier by: * **Enhancing Data Quality**: Human annotations and expert curation correct errors and add precise labels that automated systems might miss. Studies show that manually curated data, such as the Google Open Images dataset, provides higher-quality training information than many other sources, resulting in more robust models (Kuznetsova et al., 2020). * **Increasing Diversity and Variability**: Contributions from a wide array of individuals incorporate different perspectives, languages, and cultural contexts. This diversity enables models to generalize more effectively, as seen in projects like Common Crawl’s multilingual datasets, which include data from hundreds of languages to support broader linguistic capabilities (Gao et al., 2020). * **Providing Contextual and Semantic Depth**: Human-generated data captures the nuances, idioms, and contextual cues necessary for accurate interpretation of real-world scenarios. Annotated datasets such as the Amazon Mechanical Turk’s (AMT) work on language nuances for NLP models help capture this depth, allowing models to manage complex language tasks (Snow et al., 2008). * **Embedding Ethical and Moral Dimensions**: Human oversight ensures that datasets reflect ethical standards and reduce biases, helping to build fair and responsible AI. Human-labeled datasets like Microsoft’s FairFace are designed to mitigate racial bias in facial recognition tasks, underscoring the role of human-generated data in creating ethical AI (Kärkkäinen & Joo, 2019). By integrating these features of human-generated data, AI models can push beyond the data wall, achieving levels of accuracy, robustness, and adaptability unattainable with synthetic or machine-generated data alone. ## **2.2. Specific AI Models and Their Need for Human-Generated Data** AI models encounter the data wall in distinct ways, each requiring particular types of human-generated data to overcome it. The following subsections highlight how human-generated data helps specific AI models break through the data wall. ### **2.2.1. Supervised Learning Models: Classification and Regression** **Challenges at the Data Wall**: Supervised learning models require high-quality labeled data to map inputs to outputs accurately. The data wall appears when labels are insufficiently detailed or inconsistent, limiting model performance. **Human-Generated Data Solutions**: * **Classification Models**: Human annotators provide accurate and nuanced labels, enabling finer-grained classifications. ImageNet, one of the most significant supervised learning datasets, has benefited from human labeling to provide detailed classifications across 1,000 categories (Deng et al., 2009). This detailed labeling allows classification models to break through the data wall by understanding complex distinctions, such as differentiating between various animal species. * **Regression Models**: Human experts supply detailed numerical labels, especially in complex fields like healthcare or finance. In medical AI, for instance, human-annotated patient records allow models to predict disease progression with higher accuracy, as seen in studies utilizing clinical data from the MIMIC-III dataset (Johnson et al., 2016). By incorporating human-generated labels and expertise, supervised learning models can better navigate complex relationships, helping them overcome the data wall. ### **2.2.2. Unsupervised Learning Models: Clustering and Dimensionality Reduction** **Challenges at the Data Wall**: Unsupervised models face challenges when data lacks meaningful structure or diversity, limiting their capacity to find valuable patterns. **Human-Generated Data Solutions**: * **Clustering Models**: Human-curated datasets ensure relevance and diversity in features, as demonstrated by the labeled human activity data in the UCI HAR dataset, which enables models to cluster and segment meaningful patterns (Anguita et al., 2013). This guidance allows models to create clusters that are actionable and relevant for real-world applications. * **Dimensionality Reduction Models**: Human expertise in identifying essential features, such as in gene expression studies for genomics, allows models to maintain relevant information while reducing noise (Tian et al., 2014). Human-generated data enhances unsupervised models’ ability to uncover insightful patterns by providing relevant, curated features that help models push past the data wall. ### **2.2.3. Natural Language Processing (NLP) Models: Language Understanding** **Challenges at the Data Wall**: NLP models struggle to interpret idiomatic expressions, evolving language, and nuanced semantics without human-contextualized data. **Human-Generated Data Solutions**: * **Language Models**: Human-generated text sources like Wikipedia, Reddit, and curated datasets from social media provide rich linguistic structures, enabling models to understand varied dialects, sarcasm, and cultural references (Wulczyn et al., 2017). * **Sentiment Analysis and Translation Models**: Human-annotated data captures emotional tone and cultural references, essential for accurate sentiment analysis. Models trained on datasets like the Stanford Sentiment Treebank, which includes human-labeled sentiment values, are better equipped to handle nuanced language (Socher et al., 2013). With human-generated linguistic data, NLP models can overcome the data wall by deepening their understanding of language’s complexities. ### **2.2.4. Reinforcement Learning (RL) Models: Interactive Learning** **Challenges at the Data Wall**: RL models, particularly those in simulated environments, are limited by the scope and realism of these simulations. **Human-Generated Data Solutions**: * **Interactive Environments**: Human-designed simulations introduce realistic, challenging scenarios. For example, the OpenAI Gym environment allows RL models to train in human-relevant, complex simulations that include real-world variability (Brockman et al., 2016). * **Feedback and Rewards**: Human-defined reward structures better align RL models with desired outcomes, as seen in real-world tasks such as robotic surgery (Ryu et al., 2020). Human involvement creates richer, more realistic RL environments, enabling these models to surpass the data wall by learning from complex interactions. ### **2.2.5. Transfer Learning Models: Domain Adaptation** **Challenges at the Data Wall**: Transfer learning models hit a plateau when adapting to niche applications due to a lack of domain-specific data. **Human-Generated Data Solutions**: * **Fine-Tuning with Expert Data**: Human-generated data from specialized fields, such as clinical data for medical applications, allows models to adjust effectively to new domains. For instance, models fine-tuned on human-annotated radiology reports outperform those trained on general datasets (Irvin et al., 2019). * **Domain-Specific Annotations**: Domain experts provide essential annotations, allowing models to navigate complex fields with unique terminology or concepts. Human-generated domain expertise allows transfer learning models to excel in specialized applications, enabling them to break through the data wall in domain-specific contexts. ### **2.2.6. Generative Models: Content Creation** **Challenges at the Data Wall**: Generative models, such as GANs, encounter the data wall when trained on homogeneous datasets, leading to repetitive or low-quality outputs. **Human-Generated Data Solutions**: * **Diverse and Authentic Training Data**: Human-created artistic or literary content provides models with stylistic variety, as seen in the human-curated datasets used by models like DALL-E for creative tasks (Ramesh et al., 2021). * **Quality Enhancements**: Human curation ensures high-quality training data, helping models produce realistic and engaging outputs. Human-generated data allows generative models to produce creative, authentic content, helping them push past repetitive or uninspired outputs. ### **2.2.7. Few-Shot and Zero-Shot Learning Models: Generalization from Limited Examples** **Challenges at the Data Wall**: These models aim to generalize from minimal data but struggle when existing datasets lack the breadth needed for versatile adaptation. **Human-Generated Data Solutions**: * **Rich Semantic Relationships**: Human-curated datasets with varied concept relationships enhance model generalization, as demonstrated in OpenAI’s GPT-3 zero-shot capabilities (Brown et al., 2020). * **Contextual Information**: Human annotations help models understand how to apply knowledge in novel situations, a key factor in zero-shot learning. Human-generated data empowers few-shot and zero-shot models to generalize effectively, helping them overcome the data wall. ### **2.2.8. Multimodal Models: Integrating Multiple Data Types** **Challenges at the Data Wall**: Multimodal models require aligned datasets across modalities, which is challenging without human intervention. **Human-Generated Data Solutions**: * **Aligned Annotations**: Human-curated datasets like MS-COCO, with images linked to descriptive text, allow models to learn relationships across modalities (Lin et al., 2014). * **Contextual Bridging**: Human insights guide models in understanding how different modalities complement each other. Human-generated multimodal datasets help these models push past the data wall by enabling effective cross-modal learning. ## **2.3. The Unified Impact of Human-Generated Data on AI Models** Human-generated data provides a crucial solution to the data wall by introducing: * **Complexity and Nuance**: Enriching data with intricate details that support advanced learning. * **Relevance and Adaptability**: Keeping models current with real-world language, trends, and societal changes. * **Ethical and Fair Representation**: Addressing bias and promoting fairness in AI applications. Human-generated data allows AI models to break through the data wall, creating more accurate, adaptable, and context-aware systems essential for real-world applications. #### References * Brockman, G., et al. (2016). OpenAI Gym. *arXiv preprint arXiv:1606.01540.* * Brown, T. B., et al. (2020). Language models are few-shot learners. In *Advances in Neural Information Processing Systems*. * Deng, J., et al. (2009). ImageNet: A large-scale hierarchical image database. In *CVPR*. * Gao, L., et al. (2020). The Pile: An 800GB Dataset of Diverse Text for Language Modeling. *arXiv preprint arXiv:2101.00027.* * Halevy, A., Norvig, P., & Pereira, F. (2009). The unreasonable effectiveness of data. *IEEE Intelligent Systems*, 24(2), 8-12. * Irvin, J., et al. (2019). CheXpert: A large chest radiograph dataset with uncertainty labels and expert comparison. In *AAAI*. * Lin, T. Y., et al. (2014). Microsoft COCO: Common objects in context. In *European Conference on Computer Vision*. # Scaling data generation with blockchain Human-generated data plays a crucial role in the development of AI models, providing the context, diversity, and depth that automated data sources often lack. However, challenges persist in acquiring, verifying, and maintaining high-quality human-generated data, which is essential for robust AI model training. Blockchain technology addresses these challenges by introducing a secure, transparent, and decentralized framework that enhances data integrity, trust, and traceability. This section explores the key characteristics of human-generated data and how blockchain can enhance each one, laying a foundation for more reliable and effective AI training datasets. ## 3.1 Characteristics of Human-Generated Data and the Role of Blockchain in Enhancing Data Quality ### **3.1.1 Diversity and Variability** Human-generated data encompasses diverse perspectives across demographics, languages, cultures, and contexts, which is vital for training AI models that generalize well across various populations and applications. Achieving a broad and representative data sample, however, can be difficult due to centralized collection methods and potential biases. **Blockchain Solution**: Blockchain’s decentralized and transparent nature allows data to be gathered from a global pool of contributors, promoting inclusivity and diversity without centralized control. A blockchain-backed system can use smart contracts to incentivize diverse data contributions, rewarding participants from varied backgrounds. By enabling people worldwide to contribute data securely and directly, blockchain helps create datasets that are more representative of different demographics and perspectives, thus reducing demographic biases and improving AI model generalization (Kumar et al., 2020). ### **3.1.2. Contextual and Semantic Depth** Human-generated data offers contextual richness and semantic depth, enabling AI models to interpret complex real-world scenarios more accurately. However, ensuring that context is consistently and accurately represented across large datasets is challenging, particularly when numerous contributors are involved. **Blockchain Solution**: Blockchain’s immutability and transparency provide a reliable means for documenting the context and origin of each data point. Contributors can attach metadata, such as cultural context, language, or source information, to each entry, creating a permanent record. This allows AI developers to trace the contextual depth of data and verify its origin, ensuring that it is appropriately interpreted. Blockchain’s traceability feature is especially beneficial in tasks like sentiment analysis, where even subtle contextual changes can affect interpretation and outcomes (Zheng et al., 2018). ### **3.1.3. Data Authenticity and Reliability** Data authenticity is crucial for applications where accuracy and reliability are paramount, such as in medical diagnostics and financial forecasting. Human-generated data is susceptible to inaccuracies or malicious manipulation, which can compromise AI model performance if not carefully managed. **Blockchain Solution**: Blockchain’s secure, tamper-proof system for recording and verifying data authenticity helps address this issue. Each data entry, along with its origin and subsequent modifications, is recorded as an immutable transaction on the blockchain, allowing stakeholders to verify data authenticity and trace its lineage. Blockchain’s consensus mechanisms further enhance reliability by validating data entries through community verification, thus establishing a more trustworthy dataset (Nakamoto, 2008). In sectors like healthcare, where data authenticity directly impacts patient outcomes, blockchain’s tamper-resistance is invaluable (Sharma et al., 2020). ### **3.1.4. Ethical and Moral Dimensions** Human-generated data often reflects ethical and cultural values that are essential for building fair and responsible AI models. However, ensuring that datasets are ethically sourced and culturally sensitive can be challenging, especially when data contributions come from diverse sources. **Blockchain Solution**: Blockchain’s transparency enables a clear audit trail for data collection, usage, and ethical compliance. Contributors can set usage conditions via smart contracts, ensuring that their data is applied only in ethically approved ways. Furthermore, blockchain allows communities to review and flag potentially biased or culturally insensitive content, creating a framework that upholds ethical standards in AI training data. This accountability system can help prevent AI models from perpetuating biases and encourages inclusive data practices (Tapscott & Tapscott, 2017). *** ## **3.2. Building Quality Human-Generated Datasets with Blockchain: The Role of Curators and Contributors** The process of creating high-quality human-generated datasets often involves a collaborative approach with curators who organize, verify, and label data. Blockchain can enhance these curatorial efforts by ensuring data transparency, traceability, and fair incentives for contributors, leading to more accurate and reliable datasets. ### **3.2.1. Data Annotation and Verification** Human-curated data requires meticulous annotation and verification to ensure accuracy and relevance, a process that can be time-consuming and difficult to monitor, particularly across large, decentralized annotation teams. **Blockchain Solution**: Blockchain technology allows annotators to log each annotation securely, creating a permanent, traceable record of who labeled the data and any modifications made. This ledger helps AI developers verify the integrity of annotations, ensuring consistency and accuracy. Blockchain-based smart contracts can also incentivize annotators by rewarding verified, high-quality work, creating a system that encourages careful, precise labeling and reduces the potential for errors (Yaga et al., 2018). ### **3.2.2. Collaborative Curation and Consensus** For high-quality datasets, curators may need input from multiple experts to reach consensus on complex or subjective labels, as seen in fields like medical imaging or sentiment analysis. Traditional collaboration methods can be opaque and prone to disagreement. **Blockchain Solution**: Blockchain’s consensus mechanisms facilitate a collaborative curation model, allowing multiple experts to contribute to labeling and reach consensus through transparent, decentralized voting. For instance, blockchain protocols can enable curators to vote on challenging labels, with final decisions recorded immutably on the blockchain. This approach ensures datasets benefit from diverse expertise, producing a more accurate and reliable labeling process. Additionally, blockchain enables the creation of decentralized autonomous organizations (DAOs), which can transparently oversee and validate data curation efforts (Buterin, 2014). ### **3.2.3. Incentivizing Quality Data Contributions** A significant challenge in building human-generated datasets is motivating contributors to provide high-quality data, particularly when data curation is time-intensive or requires expertise. Blockchain enables transparent, fair incentives for contributors, fostering a self-sustaining data ecosystem. **Blockchain Solution**: Smart contracts on blockchain can reward contributors for quality submissions. For instance, contributors might receive tokens or other incentives for validated entries, with higher rewards for meeting predefined quality standards, such as multiple curator verifications. Blockchain’s transparency allows contributors to retain control over their data and track its usage, fostering trust and encouraging more reliable, high-quality data contributions that help AI models achieve greater accuracy (Zyskind et al., 2015). *** #### **References** * Buterin, V. (2014). A next-generation smart contract and decentralized application platform. *Ethereum White Paper*. * Kumar, S., et al. (2020). A comprehensive survey on security and privacy for blockchain-based decentralized applications: A view from the internet of things. *IEEE Internet of Things Journal*, 7(10), 10200-10220. * Nakamoto, S. (2008). Bitcoin: A peer-to-peer electronic cash system. *Bitcoin White Paper*. * Sharma, R., et al. (2020). Blockchain technology for healthcare: Enhancing transparency, security, and efficiency. *Computers & Electrical Engineering*, 87, 106735. * Tapscott, D., & Tapscott, A. (2017). How blockchain is changing finance. *Harvard Business Review*. * Yaga, D., Mell, P., Roby, N., & Scarfone, K. (2018). Blockchain technology overview. *National Institute of Standards and Technology*. # Blockchain Integration This section of the whitepaper focuses on the technical blockchain infrastructure built on Ta-da. Our smart contracts are specifically made to secure funds, ensuring the safety of financial transactions. An on-chain snapshot system is implemented for cost-efficiency, grouping transactions into batches, reducing fees and streamlining processing times. Integration of a native noncustodial wallet into our app enables users to interact directly with the blockchain, enhancing transparency and giving users full control over their tokens. The best part? It's all abstracted. No need for specific blockchain knowledge to use Ta-da and profit from theses features.
# Smart Contracts The Ta-da platform connects users, smart contracts, and other components to ensure tasks are completed and rewards are distributed efficiently. This system is designed to be user-friendly, with all blockchain interactions being completely transparent to the users thanks to an integrated crypto wallet. This initial step in decentralization focuses mainly on securely decentralizing user funds. This is done through two smart contracts: 1. **Deposit Pool**
* It holds the users' tokens linked to the app. These tokens are used to lock deposits when participating in various tasks and for the entire in-app economy (buying, selling, etc.). * The wallet integrated into the mobile app communicates with this smart contract seamlessly for the user. * Users can deposit new tokens at any time. They can also withdraw them, but since the smart contract does not know how many tokens are locked as deposits (this information is currently held on the backend), withdrawals are only processed after backend verification. A recent improvement has implemented automatic token withdrawals directly within the smart contract after an unanswered request from the backend. This further reduces the backend's power. * This smart contract also allows the backend to slash users. When a user loses his deposit due to an invalid task, the backend calls the contract's slashing method, which removes the amount of his deposit and partially transfers it to the prize pool. 2. **Prize Pool**
* It collects and manages the tokens in the prize pool. * When multiple tasks are completed, the backend triggers the purchase of tokens on the market and sends them to the prize pool. * It also receives tokens from user slashing. * It distributes rewards each week. Below is a diagram illustrating the general workflow:
Here is a detailed summary of how the Ta-da platform operates: * **Users join and deposit tokens**: Users earn token by completing tasks or depositing tokens into the Deposit Pool to participate in tasks. * **Tasks completion and validation**: Users complete tasks, such as recording audio samples or participating in social media marketing activities. The backend processes and validates the results of these tasks. * **Rewards and slashing mechanism**: * **Valid tasks**: If a task meets the set criteria, the backend triggers the transfer of rewards to the Prize Pool for distribution to the users. At this stage, the injected tokens are purchased on the market. * **Invalid tasks**: If a task is found invalid, the user's deposit is slashed, and the deposit is partially transferred to the Prize Pool. * **Rewards distribution**: The Prize Pool distributes rewards to users based on their contributions and [the criteria met.](#user-content-fn-1)[^1] * **Fund management**: Funds from jobs provide rewards to the Prize Pool. The backend system also approves withdrawal requests and manages the slashing and reward processes to ensure fairness. * **Withdrawal**: Users can request to withdraw their tokens from the Deposit Pool. The backend approves these withdrawals, ensuring users can access their funds when needed. Fees may be charged for transactions. This structured workflow ensures that tasks are managed efficiently, rewards are distributed fairly, and user interactions with the platform are smooth and secure. Users benefit from an easy-to-use system where their contributions are rewarded, and their funds are safely managed across both the Ta-da protocol and their personal wallets. [^1]: cheloue la tournure. J'aurais dit "their ranking in the leaderboard" tout simplement ? # Snapshot System To simplify the user experience, we have chosen not to require users to pay transaction fees in EGLD. Integrating multiple tokens into the application would introduce unnecessary complexity and create a barrier to entry. Instead, most interactions within the app are paid by Ta-da, ensuring a seamless experience for users. However, for a micro-tasking application, it would be inefficient to cover transaction fees for every single action. Therefore, we implemented a snapshot system. It consolidates multiple transactions into a single one, significantly reducing fees. {% hint style="info" %} Example: a standard speech collection job may require \~5,000,000 tasks (productions and checks) and therefore as many blockchain transactions. Considering an average transaction fee of $0.03, the total cost would be $150,000, it's too prohibitive. {% endhint %} Currently, the snapshot system applies primarily to the deposit of TADA tokens (after swapping) into the Prize Pool, slashing, and in-app purchases. When a task is validated by a user, the associated payment is not immediately placed into the Prize Pool. Instead, these payments are aggregated and sent in one batch transaction. The same process applies to slashing and in-app purchases. Information on these transactions is saved in the database and finalized when the snapshot is executed. This approach allows us to maintain a cost-effective and efficient system while providing a smooth user experience. # In-app Wallet The application integrates a wallet that plays a crucial role in user authentication, transaction signing (non-repudiation), and secure token storage.
Additionally, the wallet serves as an interface between the external world and the application for token management. This means that tokens can be securely stored and managed directly within the app. Currently, tokens can be located in three different places: 1. **World**: This refers to external wallets, centralized exchanges, and other off-chain locations where users store their tokens. 2. **In-app Wallet**: This acts as a bridge between the external world and Ta-da. It is used for user authentication, with actions being signed using this wallet. 3. **On-chain Account**: This holds the tokens that are actively used within the application, such as for deposits, in-app purchases, and prize pool distribution. The diagram below outlines the different interactions:
* **Receive**: Tokens are transferred from the external world to the in-app wallet. * **Send**: Tokens are sent from the in-app wallet back to the external world. * **Deposit**: Tokens are moved from the in-app wallet to the on-chain account for use within Ta-da. * **Withdraw**: Tokens are transferred from the on-chain account back to the in-app wallet, making them available for external use again. # Signature The in-app wallet enables users to sign transactions or information to prove that they are the source of these actions. These signatures are verified by the smart contract Deposit Pool before any deposit, withdrawal, or in-app purchase. The sequence diagram below illustrates the general flow of an action that requires a signature.
Swimlane diagram of transaction signature verification process

Swimlane diagram of transaction signature verification process

In the diagram, the process begins in the app where a payload is created and signed by the user. This signed payload is then sent to the backend, where it checks the payload and the user's signature. If valid, the backend makes a transaction using the payload and user signature, which is then sent to the smart contract. The smart contract verifies the user’s signature once more before executing the payload. This ensures that all actions are securely authenticated and authorized, maintaining the integrity of the user's transactions and data within the application. # Next step: Slashing Authorization ## Slashing Authorization Our goal is to fully decentralize Ta-da. However, this cannot be achieved all at once, so we are progressing gradually. In this second step, we are addressing the slashing system. In the previous version, the backend could theoretically slash an arbitrary amount of tokens. Therefore, it is necessary to implement a solution that prevents or mitigates this risk. This section describes such a system. When the user performs a task he sends to the backend a slashing authorization signed by his private key. A signed authorization is irrefutable proof that the user allows the deduction of a precisely determined deposit amount for a specific task. The backend stores this authorization and sends it to the Deposit Pool if the user is slashed. This smart contract then checks that the authorization is valid thanks to the signature and slashes the user accordingly. That way, a user cannot lose more than he has previously authorized. A slashing authorization must contain this information: * the identifier of the task * the amount of tokens the user accepts to lose through slashing * a nonce avoiding replay attacks * a timestamp * the signature Before accepting and storing the slashing authorization, the backend checks the validity of each field. The nonce prevents replay attacks. Without it, the backend could send the same authorization multiple times to slash all of a user's tokens. The Deposit Pool prevents this attack by checking that the authorization has not already been executed using this nonce. At this stage, the backend could still arbitrarily slash a user. However, it could only do so for the amount specified in the authorizations that have been sent to it and have not yet been executed. This greatly mitigates the risk. *** ## Auto withdrawal As a reminder, the Deposit Pool executes withdrawal requests only after backend confirmation. This means that if the backend fails for any reason, the withdrawals cannot be processed. To address this issue, the smart contract implements a period during which the backend must authorize or refuse the withdrawal (it must respond either way). If there is no response after this period, the tokens are automatically released and returned to the user.
# Next step: Ta-da chain To fully decentralize Ta-da, it is necessary to migrate all the business logic currently implemented in the backend to the blockchain. This includes the consensus system, dataset storage, user data security, and more. \ Although the power and expressiveness of smart contract languages are constantly evolving, they are not yet advanced enough to enable us to implement our business logic in a completely decentralized way. For these reasons, alongside the implementation of various new features, we will be working on developing our own blockchain. \ \ Of course, we will not be redeveloping everything from scratch; we will leverage technologies such as [MultiversX's Sovereign Shards](broken://pages/mL11w1C7NDkTDsf8AOpI), [Avalanche Subnets](https://docs.avax.network/learn/avalanche/subnets-overview) or [Cosmos Chain](https://cosmos.network/). Stay tuned! # Quality Assurance ## Ensuring Quality Work Quality work adheres to specific criteria established by the company. \ These criteria include accuracy, completeness, and adherence to guidelines. ## First Process: Community Checking Beyond meeting these criteria, the work must also be approved by a sufficient subset of the community. This means that after a task is completed, it is reviewed by several community members who evaluate whether it meets the established standards. For instance, if a comment is posted, community members will check its relevance and appropriateness. If the majority agree that the task meets the criteria, it is considered valid. In our system, we use the concept of the Schelling point to ensure consensus within the community. This means that the validation relies on the collective agreement of the community, ensuring that tasks are performed to a high standard and approved by a representative group. For example, if all checkers listening to an audio recording agree without consulting each other that there is background noise, we simply consider that there is background noise in the recording. If the company wanted such noise, the data is considered valid (it matches the company's expectations); otherwise, it is invalid. Now that we have clarified the concepts of quality and validity, let's see how users are incentivized to produce good quality data. Each time a user performs an action (production or vote), they lock in a deposit (a small amount of tokens). * When a producer's data is validated by the checkers, TADA tokens are added to the prize pool, and the producer earns XP, allowing them to climb the ranking. The users with the most XP earn the most tokens. Conversely, if the data is not valid, they earn nothing and lose their deposit. The deposit is a structural incentive system for producers and checkers, ensuring that people will do a good job, even if it is insignificant compared to the gains. This setup has deep roots in DeFi and blockchain protocols in general and is proven to work. * For the checkers, the mechanism is slightly more complex and relies on a consensus calculation among the voters. Basically, the checkers who voted against the majority are slashed. For example, if four checkers indicate hearing an elderly person in an audio recording and only one checker indicates hearing a child's voice, the four checkers are paid while the last one is slashed. *** ## Second Process: Automatic Checking While the community-based verification method is highly effective for ensuring the quality of AI-related datasets, such as audio recordings, we have also implemented automated systems to verify other types of tasks. These automated verification systems are designed to efficiently and accurately validate user actions, such as social media interactions. For example, in verifying that users have retweeted a specific post, the automated system performs the following steps: * **Task Assignment:** Users are assigned the task of retweeting a specific post according to the project's requirements. * **Action Tracking:** The system automatically tracks the completion of this task by monitoring users' social media activities. * **Verification:** The system verifies that the retweet action has been completed by checking the user's social media account for the specified retweet, interfacing with the social media platform's API. * **Validation:** Upon confirmation of the retweet, the system marks the task as “in review.” Based on the project’s parameters, an automatic verification process will initiate, taking anywhere from a few minutes to a few days. If successful, the task is validated and the user is rewarded accordingly. * **Incentives and Penalties:** Similar to the community-based verification, users receive rewards in the form of points upon successful completion of tasks. If the task is not completed as required, the user may lose their deposit, discouraging fraudulent activities (e.g., retweet, check, unretweet). By leveraging automated verification systems, we ensure that tasks such as social media engagement are validated quickly and accurately, reducing the manual effort required and increasing overall efficiency. This dual approach, combining both community-based and automated verification, enables Ta-da to handle a diverse range of tasks with the appropriate level of scrutiny and accuracy needed for each use case. # Architecture ## Key challenges Ta-da provides a [web platform](https://app.ta-da.io) connecting crowdworking supply and demand across multiple domains, from data collection to surveys or social media engagement. Building an efficient crowdsourcing platform is deceptively complex. While the core idea of distributing tasks to a large pool of contributors sounds simple, executing it at scale and with quality requires solving multiple layered challenges across UX, data integrity, or operational scalability. **PRODUCTION** * Global crowdsourcing platforms must offer the same level of performance across countries with very different devices and internet bandwidth available * Most failures in crowdsourcing stem from poor task interfaces. Ambiguity in instructions or clunky UX can drastically increase error rates, and quality control costs * Data produced needs to be well structured and traceable, and coming from verified, trusted users to guarantee quality **QUALITY CONTROL** The paradox of crowdsourcing is that while it gives access to large volumes of human input, this input is inherently noisy. Ensuring **accuracy, consistency, and validity** of submissions (especially in open environments) demands robust verification mechanisms. **REWARDS** Sustainable crowdsourcing models require a balance between **speed, cost, and data quality**, while remaining attractive enough for high quality producers to remain involved with the platform. # Production ### **User Journey** Users can vizualize jobs available to them on their dashboard. Some jobs may not be available to all users, based on their profile. There are 4 types of jobs available: * Data collection: Voice, Photo and Videos * Surveys: answer questionnaires * Social media: visit a social media page and engage * Project review: visit a website or download an app and complete actions A job can consist of several tasks, which must all be completed before the user can validate their production. Jobs can either be "unique", which means a user can only complete the job once. Or they can be "multiple" which means a user can submit the same job several times. ### **UX highlights** Ta-da's production interface has been built around key principles, to balance scalability and user-experience with * deeply customizable tasks, ensuring guidelines can be adjusted to each campaign * Adaptive UX based on user behavior and performance. * Support user-specific contents, based on past behaviour, performance or profile * Light frameworks and responsive UIs to maximize accessibility across continents #### ### **Back-end highlights** To optimize throughput and reduce latency, our back-end is designed to support: * **Smart routing**: ensuring the right user gets the right task based on skill, availability, or past performance. * **Concurrency handling**: when millions of microtasks are being processed in parallel, race conditions, double submissions, or stale task serving can degrade performance. * **Fraud detection models** (e.g., device fingerprinting, behavioral analysis), * **KYC-lite mechanisms** for sensitive data and **trust scoring** systems for long-term integrity. * Robust **metadata pipelines**, **version control for datasets**, to maintain links between raw input, annotations, and audit logs. #### # Quality Control ### Verification flow Quality controls are made at two levels: * task level: making sure each task has been properly done * job level: making sure all tasks have been completed consistently with requirements Our quality check relies upon 3 levels of validation, both fully customizable based on clients' needs: * Consensus based verification: verification is done by "community verifiers", and validation is made based on consensus, with "slashing" of incorrect productions or colluding verifiers to incentivize good behaviours * AI-powered review: specifically trained LLMs help screen compliant tasks to increase verification throughput * Curator validation: verification is done by specifically trained curators Depending upon clients' needs and projects' complexity, it is possible to combine any of these 3 verification mechanisms to provide stronger quality checks Job validation triggers the distribution of rewards, unless specified otherwise in instructions ### **QC core principles** Ta-da's platform is build so as to enable quality controls at scale, guaranteeing better data quality, at * **Redundancy models** (e.g., consensus, majority voting) can be expensive and slow. * **Automated quality checks** often struggle with edge cases or subtle errors. * **Data injection** and real-time feedback loops * **Integration with ML pipelines** or client APIs to close the loop between data needs and data supply. * **Analytics dashboards** to help data consumers measure dataset evolution, annotation variance, and coverage gaps. # Rewards #### ### Rewards Model Our platform leverages web3 incentives to promote long term participation of performing workers Users are paid on a task completion basis Remuneration varies from one job to an other depending upon each project's specifics Each user can decide to cash out their rewards: * in FIAT currency, using amazon gift cards and/or paypal * in $TADA tokens (see "token" section), which brings benefits to users * in Stablecoins # Use-cases ## Several Use-cases The Ta-da platform supports a wide range of use cases, addressing the diverse needs of modern projects. From creating audio, video, and text datasets for training AI models to supporting social media marketing and content management efforts, Ta-da provides practical solutions. By leveraging our community's ability to complete specific tasks, we enable our clients to generate accurate and reliable data, enhance their online presence, and manage their content effectively. This section details the various use cases of Ta-da, demonstrating how our platform can be utilized across different fields to achieve optimal efficiency and results. Currently, we have three major use cases: * [Artificial Intelligence (multi-task)](/ta-da-platform/use-cases/artificial-intelligence) * [Social Media Engagement](/ta-da-platform/application-overview/social-media-engagement) * [Data Structuring](/ta-da-platform/use-cases/data-structuring) # Artificial Intelligence ## Powering the next-generation of AI models. Ta-da is a powerful tool for creating high-quality datasets essential for various AI applications. By leveraging our platform, companies can generate diverse and reliable datasets tailored to specific AI needs. Whether it's scripted audio recordings for speech recognition, bounding boxes for computer vision, or annotated text for natural language processing, Ta-da enables precise and efficient data collection. This section outlines the numerous use cases where Ta-da can significantly enhance AI projects by providing the accurate and comprehensive datasets required for effective machine learning and AI model training. Currently, there are four major subsections of use cases: * [Audio Datasets](/ta-da-platform/use-cases/artificial-intelligence/audio-datasets) * [Video Datasets](/ta-da-platform/use-cases/artificial-intelligence/video-datasets) * [Image Datasets](/ta-da-platform/use-cases/artificial-intelligence/image-datasets) * [Text Datasets](/ta-da-platform/use-cases/artificial-intelligence/text-datasets) # Audio Datasets ## Providing Scripted Audio Recordings Scripted audio recordings are datasets of audio samples that can be used to train and test speech recognition models. These recordings help machine learning models learn how to recognize different accents and dialects, as well as how to identify different words and phrases. They can also be used to create datasets specific to certain domains, such as medical speech or customer service conversations. The main method for making scripted audio recordings for machine learning is to create a script and then record it. This script should include information about the intended audience, the content that needs to be recorded, and any relevant audio cues.
These data are particularly difficult to obtain because they need to be extremely varied. Indeed, for a voice assistant to recognize any type of voice, it must be trained with recordings from men and women of all ages, sometimes with background noise, in different ways of speaking (whispering, shouting, etc.), with specific vocabulary, and more. Thanks to the ease of use of Ta-da, anyone can record their voice and participate in creating diverse datasets, which is essential for creating good datasets. Ta-da has already proven its value in the realm of scripted audio recordings by successfully delivering high-quality datasets to numerous clients, including industry leaders like Sensory and Vivoka. This success is supported by signing multiple clients who need precise and reliable audio data for their AI applications. Our platform excels in providing accurately scripted audio recordings, ensuring that the datasets meet the high standards required for effective AI and machine learning model training.
*** ## Providing Spontaneous Audio Recordings Spontaneous audio recordings are datasets of unscripted audio samples that capture natural speech in real-world situations. These recordings are essential for training and testing speech recognition models to understand and process conversational speech, including natural pauses, hesitations, and variations in tone and speed. They are particularly useful for developing models that can handle everyday speech patterns, such as informal language, slang, and spontaneous interactions. The primary method for creating spontaneous audio recordings involves capturing conversations in natural settings without predefined scripts. These recordings can include a wide range of scenarios, such as casual conversations, interviews, and impromptu speeches. This approach ensures that the collected data reflects the authentic way people speak, providing valuable insights for machine learning models. These datasets are challenging to compile due to their need for diversity and authenticity. To create robust speech recognition systems, it is crucial to have recordings from speakers of different genders, ages, and backgrounds, in various acoustic environments, and with different speaking styles, such as casual, formal, and emotional. The flexibility and user-friendliness of Ta-da enable individuals to contribute their spontaneous speech effortlessly, generating rich and varied datasets that are critical for developing accurate and effective speech recognition models. # Video Datasets ## Providing video-based datasets Ta-da is designed to enable the creation of high-quality video datasets, which are crucial for a wide range of AI applications. Our platform supports the comprehensive collection and annotation of video data, which is fundamental for training sophisticated machine learning models in fields such as computer vision, autonomous driving, and video analytics.
Key features of video dataset creation capabilities include: * **Bounding Boxes:** Precise annotation of objects within video frames. This feature is essential for training AI models to detect and track objects accurately across video sequences. * **Action Recognition:** Detailed labeling of specific actions or activities within video segments. This allows for the development of AI models capable of behavior analysis, activity recognition, and surveillance. * **Scene Segmentation:** Advanced division of video frames into distinct segments. This enables detailed analysis and classification of various components within a scene. This is crucial for understanding complex environments and improving scene understanding models. * **Object Tracking:** Robust tracking of object movements across multiple frames, which is essential for developing effective tracking algorithms used in various applications such as security, sports analysis, and autonomous navigation. * **Biometric Analysis (KYC):** Integration of biometric data for Know Your Customer (KYC) processes. This includes facial recognition and other biometric verifications, ensuring accurate identification and authentication of individuals within video datasets. We have already successfully created a video dataset for Identt, which includes videos specifically designed to train KYC algorithms. This dataset enhances the accuracy and reliability of biometric verification systems, demonstrating our capability to deliver specialized datasets for complex AI applications. # Image Datasets ## Providing Image-based datasets Image labelling is the process of assigning labels to an image or set of images. Labels can be as simple as classifying an object as a “cat” or “bicycle”, or as complex as recognizing an action in a sequence of images.
Image labeling is a crucial component of many computer vision applications, including object detection, scene understanding, and image classification. Here are some examples you might know: * Tesla's auto-pilot uses these datasets to train on recognizing roads, pedestrians, signs, and more. * Facial recognition systems on smartphones, used to unlock them, are also trained on these types of datasets. In the field of computer vision, there are seven methods of labeling. Here are brief descriptions of some of them: * **Classification:** The process of assigning a label or class to an image, such as a person, object, scene, or activity. This is already implemented in Ta-da. * **Polygons:** A type of image annotation used to label the boundaries of objects in an image. It involves manually drawing polygons around objects of interest in an image, such as cars, buildings, people, etc. This type of annotation is commonly used to train deep learning algorithms for object detection and segmentation. * **Semantic segmentation:** Used to assign a semantic label to each pixel in an image, such as "cat" or "road." These labels can then be used to classify the image into different objects and scenes. This technology is useful in a variety of applications, such as autonomous driving, medical imaging, and satellite image analysis. * **Bounding box:** A rectangular box that is drawn around an object in an image. It defines the area of the object and is used to label the object by specifying the coordinates of the box. For example, a bounding box for an image of a cat might be drawn around the cat. Technically speaking, we store the coordinates of the box’s top left corner, the width and height of the box, and the class of the object (e.g., cat). Below is an example of a bounding box drawn around a cat:
# Text Datasets ## Providing Text-based datasets Natural Language Processing (NLP) is a vast field of research, particularly in Natural Language Understanding (NLU). To support this industry, we develop various types of jobs, including:Here are brief descriptions of some of them: * **Translation:** The data collected from translation tasks is used to train machine translation AI for automatic translation solutions. Many people have likely used online translation tools, and this training improves their accuracy and usability. * **Text Classification:** This involves assigning a label or class to a piece of text, indicating the type of content such as news, opinion, reviews, etc. Criteria for classification may include keywords, text length, word count, and other features. Text classification helps organize large datasets and identify trends. * **Token Classification:** A specialized form of text classification that labels individual words or tokens within a text. This can identify parts of speech or determine the sentiment of a sentence. Companies use this to analyze their reputation on social media, for instance. * **Summarization:** Condensing information into a concise and comprehensive summary. Summarization quickly extracts the most important and essential points from a text while maintaining its key themes. This is useful for training AI to simplify texts, extract key points, or provide content overviews. * **Prompt Creation for LLMs:** Generating datasets of prompts to train and test large language models (LLMs). These prompts help LLMs learn to respond accurately and contextually in various scenarios, enhancing their performance in generating human-like text. This is crucial for applications such as chatbots, virtual assistants, and automated content creation. # Data Structuring ## Structuring product data In the dynamic sector of retail and e-commerce, effective product information management (PIM) is crucial for business success. A key component of this management is the use of structured data, where each product is meticulously documented with well-defined labels. This structuring not only allows for better organization of product data but also significantly optimizes the user experience. For instance, precise labels facilitate faster and more accurate product searches, improving the online shopping experience. Moreover, in an environment where artificial intelligence and machine learning are increasingly influential, well-structured data is indispensable for generating personalized and relevant product recommendations. Therefore, the implementation of robust and intelligent PIM systems capable of efficiently managing this structured data is a major challenge for businesses looking to remain competitive in the digital commerce landscape. To further enhance the efficacy of PIM systems in the retail and e-commerce sector, we will introduce a range of new job types on our platform. These jobs are specifically designed to engage our user base in the detailed processes of product data labeling and verification. This approach is particularly effective in complex categories where a nuanced understanding of products is essential. The active participation of users in these new roles will be instrumental in refining the data quality, thus making PIM systems more robust, intelligent, and tailored to the dynamic demands of digital marketplaces. # Roadmap
# Application overview Our application is designed to revolutionize the way businesses and individuals engage with communities for executing various tasks, such as data collection and social media engagement. By providing a seamless platform that connects businesses with people, we aim to foster collaboration and deliver value through innovative solutions.
*** ## Key features * **Task Management:**\ Businesses and communities can easily post tasks, set parameters, and manage their completion through an intuitive interface. * **B2B, B2C, and C2C Models:**\ Our application supports various business models, enabling flexibility and scalability across diverse use cases. * **Prize Pool:**\ Users are fairly rewarded based on their contributions to the weekly prize pool. * **Incentivization with Native Tokens:**\ Users are rewarded with our native token for completing tasks, ensuring a fair and motivating system for all participants. * **Secure Transactions:**\ Our platform ensures secure processing of payments and rewards, maintaining trust and reliability for all users.
*** ## User interface & experience Ta-da application offers a user-friendly, intuitive interface with the following: * **Simple Onboarding:** Quick and easy registration process for new users. Wallets are created on the fly, and users are credited with their first $TADA to allow them to start tasking right away. * **Customizable Profiles:** Users can create and customize their profiles to showcase their skills and preferences. * **Efficient Task Discovery:** Advanced search and filtering options help users find tasks that match their interests and skills. A campaign mode allows a logical progression and commitment towards tasks that match their preferences. * **Seamless Interactions:** Transaction fees on the blockchain are covered by Ta-da to reduce friction between Web2 and Web3 worlds. * **Enhanced Contributions:** Through staking or in-app item purchases, users can enjoy bonuses such as energy boosts or refills for better odds of being rewarded each week. * **Progress Tracking:** Real-time updates and notifications keep users informed about their task status or Prize Pool ranking.
# Social Media Engagement ## A proven method, to gain real followers and engagement Social media engagement is essential for businesses and projects looking to expand their reach and engagement. Many projects need to build their online presence, increase their follower base, and ensure their messages reach a wider audience.
At Ta-da, we provide a solution that enables projects to assign social media tasks to our community through our application. This approach allows businesses to efficiently crowdsource social media engagement activities such as following, retweeting, and quoting tweets. Here are some tasks that can be proposed on Ta-da: * **Follow on Twitter:** Projects can create tasks that require users to follow their Twitter account. This helps build a larger follower base, which is crucial for reaching a wider audience and enhancing credibility. * **Retweeting:** Tasks can involve retweeting specific posts, amplifying the reach of important announcements, updates, or promotional content, ensuring that the message is spread across a broader network. * **Quoting Tweets:** Users can be tasked with quoting a tweet and adding their own comments, increasing the visibility of the tweet and encouraging engagement and discussion around the content, fostering a more interactive community. * **Like:** Projects can create tasks that require users to like specific posts. This simple action helps to increase the visibility and popularity of the content, as posts with more likes are often promoted more heavily by social media algorithms. * **Comment:** Users can be tasked with commenting on posts, adding their thoughts, feedback, or questions. This not only increases the interaction on the post but also fosters a sense of community and engagement around the content.
Projects often struggle with building an active, engaged follower base on social media. Traditional marketing efforts can be costly and time-consuming, and achieving organic growth requires a significant effort. Clients need an efficient way to boost their social media presence and ensure their content is seen by a larger audience. Using Ta-da to manage social media tasks addresses these needs by providing a platform where tasks can be easily distributed to a ready community of users. This ensures that social media engagement is performed by real users, enhancing the authenticity and impact of the activities. By leveraging our community, projects can achieve broader visibility, increase their follower count, and foster greater interaction with their content. # Game Mechanics ## Gamification is a key part of Ta-da. Gamification is indeed a key part of Ta-da, bringing benefits to both users and the project, including its token economy. By adding game-like features, the app makes tasks more fun and rewarding, increasing user engagement and motivation. For users, it provides a way to track progress, earn rewards, and reach higher ranks, unlocking new features and items. This not only boosts user satisfaction but also encourages regular use and participation. For the project, gamification keeps users coming back and attracts new ones, building a strong community. Economically, the token system benefits as users are motivated to stake tokens, participate in the in-app economy, and make purchases. This active participation increases the token's utility and value, creating a sustainable and growing ecosystem. In short, gamification connects user experience with economic incentives, ensuring the app's success and long-term growth. We have developed a comprehensive gamification system, which is currently summarized in the diagram below. This section aims to explain the main elements and will be updated as new features and adaptations are introduced.

a diagram of our gamification

# Energy Energy is central to the gamification of Ta-da. It is a resource that each user must manage to maximize their work capacity and, consequently, the potential rewards they can earn. Every user has an energy bar displayed in their profile.
We've added an energy system to Ta-da, underpinning a new economic layer. When a user completes a task, he loses a small amount of energy. When the energy runs out, the user can no longer perform a task. He must then replenish his energy either by waiting (energy is distributed to users every period of time) or by purchasing a battery to recharge his energy bar. In conjunction with the prize pool and ranking system, we create a virtuous loop where:

Virtuous circle of energy

Users want to maximize their profit and therefore the number of tokens won. Based on this principle, users will work to increase their number of points, thus have more chances of winning the jackpot. However, working costs energy and users risk being held back by this resource. This means it encourages them to purchase items and other improvements in the application to increase their energy and thus be able to work more. # Abilities ## LUCK Luck increases loot boxes drop rate, and content quality
Users can drop loot boxes even without improving their luck, but they will substantially improve their odds by buying luck enhancing equiment. ## AGILITY Agility is a coefficient applied to XP earned by a user when completing a task: $$ xp=xp\_{task} \* agility\_{user} $$ Higher agility score increase XP earned by a user, for the same number of tasks completed. ## STAMINA (Coming soon) # Items The item system in Ta-da is designed to enhance the user experience by providing various tools and boosts that can help users complete tasks more efficiently. Items can be earned, bought, or received as rewards, and each item has its own unique function. The items presented on this page are not yet all implemented, and some adaptations may occur. Additionally, other items may be introduced in the future. In the following section, we will introduce and describe the different items available in the application, explaining how they work and how they can benefit the users.
## Backpack The backpack item allows users to increase their inventory capacity, enabling them to store more items. This is particularly useful for users who want to carry and use a larger variety of tools and boosts, enhancing their ability to complete tasks and improve their overall performance in the application. ## Battery (Energy) The battery item allows users to refill their energy. There are various types of batteries, ranging from less powerful to more powerful ones. This is a basic item for any user who wants to complete more tasks and improve their score, thereby increasing their rewards. ## Cyber Scanner (Luck) Cyber scanners temporarily increase the luck score of users. There are 4 types of cyber scanners, providing 100% luck boosts for different periods of time. | Luck | Duration | | ---- | -------- | | 100% | 10 | | 100% | 20 | | 100% | 30 | | 100% | 60 | ## Chipset Upgrade (Agility) Chipset Upgrades increase agility score of users. There are 3 types of chipset upgrades providing increased XP | Agility | Duration (weeks) | | ------- | ---------------- | | +10% | 1 | | +20% | 1 | | +50% | 1 | ## ## # Shop The in-app shop allows users to purchase basic items like batteries. All purchases are made using $TADA tokens. As described on the "economy" page, TADA coming from sales are redistributed to four buckets: **the prize pool (65%), staking (5%), burning (5%), and the treasury (25%).** This means that the more purchases made in the shop, the larger the prize pool for users, the higher the APR for staking pools, and the greater the deflation. The initial version of the shop only offers batteries for purchase. \ \ Once additional items are implemented, the shop will offer new objects for sale, including some exclusive items.
# Lootboxes Introduced during Phase 2, loot boxes offer a new way for users to earn reward while completing tasks in Tada.
## Generalities For each task a user completes in-app, there is a probability to drop a loot box. Loot boxes can contain GEMS and/or $TADA rewards. ## Rarity There are four types of loot boxes, listed hereafter in order of increasing rarity. Rare lootboxes contain higher rewards. * Common * Rare * Epic * Legendary ## Luck Influence Drop rates, as well as loot quality are influenced by the [Luck](/ta-da-platform/game-mechanics/abilities) score of each user: higher luck scores increase chances to obtain a loot box, and improve the possible content of loot boxes dropped. # Marketplace The marketplace allows users to sell items to each other, creating new economic opportunities within the app. Users can now earn money by selling items they have won or crafted, fostering a dynamic and engaging economy. Some items will not be available for purchase in the shop, making the marketplace an essential feature for acquiring rare and unique items. Therefore, users have two options: craft the item themselves or buy it on the marketplace from other users. This system encourages user interaction and adds a layer of strategy to the game, as players must decide whether to invest time in crafting or tokens in purchasing. It is important to note that every transaction conducted will be subject to fees, which will contribute to the growth of the prize pool. # Gems When a prize pool is distributed, all users receive Gems, providing a valuable opportunity for those who did not win any $TADA tokens. These points enable users to enhance their potential to compete more effectively in the next prize pool distribution. Additionally, Gems play a crucial role within the app's economy, allowing users to make in-app purchases and unlock future functionalities.
# User rewards Ta-da offers different ways for users to get rewarded: * Weekly Prize Pools * Daily Vaults * Staking # Prize Pools Each day, tokens coming from in-app purchases and B2B sales induced buybacks are topped up with tokens from the "Community slot" and distributed to users based on specific parameters: * Activity: Tasks completed on a given day, and consistency over the last 10 days * Staking: Amount staked across Ta-da staking pools * LP providing: Amounts of Ta-da LP tokens held Rewards from Daily Prize Pools are distributed under the form of Tada Bondz (T-Bondz) The amount of tokens distributed each day is variable, and is computed with logics described below ## **Distribution Formula** Computed daily for each user, the output of this formula determines which share of a daily prize pool will be allocated to a specific user. $$ percent\_to\_distribute\_i = \dfrac{\text{score}\_i^a}{1 + \sum \text{score}\_j^a} $$ The previous value then allows for a straightforward calculation of the number of tokens to be distributed: $$ to\_distribute\_i=percent\_to\_distribute\_i \* total\_tokens\_to\_distribute $$ ## **User Score Formula** Here is the formula for calculating a user's score: $$ score\_{i} = xp\_i \* \[ 1 + ( \alpha \* staking\_score\_i * \beta \* liquidity\_score\_i * \gamma \* activity\_score\_i ) * (MAX\_AMPLIFICATION - 1) ] $$ Where: * $$score\_i$$ corresponds to the user's score (over the last 24 hours). * $$xp\_i$$ is the experience points of the user (over the last 24 hours). * $$staking\_score\_i$$ is the user staking score (please see below). * $$liquidity\_score\_i$$ is the user liquidity score (please see below). * $$activity\_score\_i$$ is the user activity score (please see below). * $$\alpha$$ is the staking score modifier. * $$\beta$$ is the liquidity score modifier. * $$\gamma$$ is the activity score modifier. * $$MAX\_AMPLIFICATION$$ is the maximum amplification. ## **Staking Score Formula** Here is the formula for calculating the staking score: $$ staking\_score\_i = min(\frac{\log(k \times weighted\_stake\_i + 1)}{\log(k \times \text{MAX\_STAKING} + 1)}, 1) $$ * $$staking\_score\_i$$ is the result of a user's staking score. It is a value within the range $$\[0, 1]$$. * $$MAX\_STAKING$$ is a value (in dollars) that caps the impact of staking for whales. Beyond this value, the score no longer increases. * $$weighted\_stake\_i$$ is the amount, weighted by the lock duration, of the dollar value of the user's staked tokens (see below). * $$k$$ is the smoothing factor of the curve. The smoothing factor allows modifying the rate of increase of the curve (see below). ### $$weighted\_stake\_i$$ Formula In the previous formula, $$weighted\_stake\_i$$ represents the weighted liquidity the user has placed in the staking pools. This value is calculated as follows: $$ weighted\_stake\_i=\sum\_{n}^{} nb\_tokens(pool\_n)\*multiplier(pool\_n)\*token\_price $$ Where: * $$nb\_tokens(pool\_n)$$ is the number of tokens locked by the user in the pool $$n$$. * $$multiplier(pool\_n)$$ is the pool factor. * $$token\_price$$ is the token price. For example, considering the values in the table below: * $1 staked in the "12 months" pool corresponds to a value of 1; * $2 staked in the "6 months" pool corresponds to a value of 1; * $10 provided in the "1 month" pool corresponds to a value of 0.8. | Stake | Factor | | --------- | ------ | | 12 months | 1 | | 6 months | 0.5 | | 3 months | 0.25 | | 1 month | 0.08 | ### $$k$$ value The smoothing factor allows modifying the rate of increase of the curve. Thus, in the example below $$(\alpha = 1)$$, we see that reaching a value of $$(0.75)$$, requires staking $6000 when $$k = 1$$, whereas it requires staking $50000 to reach the same value when $$k = 0.0001$$.

k = 1

k = 0.0001

## Liquidity Score Formula The liquidity score uses exactly the same formula as the one previously described: $$ liquidity\_score\_i = min(\frac{\log(k \times liquidity\_i + 1)}{\log(k \times \text{MAX\_LIQUIDITY} + 1)}, 1) $$ Where: * $$liquidity\_score\_i$$ is the liquidity score of the user. * $$MAX\_LIQUIDITY$$ is a value (in dollars) that caps the impact of liquidity provision for whales. Beyond this value, the score no longer increases. * $$liquidity\_i$$ is the dollar value of the tokens the user has provided in the pools. * $$k$$ is the smoothing factor of the curve. ## Activity Score Formula The activity score encourages user engagement and incentivizes them to return to the application regularly. It rewards consistency (whereas experience encourages quantity). This score increases by 1 each day the user completes at least one task and is reset to 0 if the user fails to complete any task for a day. It is calculated as follows: $$ activity\_score\_i = \frac{min(daily\_streak\_i, MAX\_ACTIVITY)}{MAX\_ACTIVITY} $$ Where * $$activity\_score\_i$$ is the activity score (between 0 and 1). * $$daily\_streak\_i$$ is the number of consecutive days during which the user has completed at least one task. * $$MAX\_ACTIVITY$$ is the maximum score allowed. Since $$daily\_streak\_i$$ is reset to 0 in case of a missed day, the user is encouraged to maintain their streak of consecutive days, even after reaching the maximum score. ## Choice of Parameters The top-down strategy for selecting the various parameters is based on the following steps: 1. Choose the desired maximum amplification. 2. Decide on the relative importance of staking and liquidity. 3. Determine the parameters. Daily Prize Pools are launched with the following parameters, which are subject to possible evolutions
ParameterInitial ValueComments
MAX\_AMPLIFICATION4x4 on XP for users with maxxed out staking and LP scores
\alpha0.7Weight of staking score
\beta0.2Weight of LP score
\gamma0.1Weight of activity score
MAX\_STAKING100 000Upper limit for staking multipliers
k0.002Smoothing factor
MAX\_ACTIVITY\_SCORE10Max days in a streak
MAX\_LIQUIDITY100 000Upper limit for liquidity multipliers
# Rewards & Staking This page describes the benefits that token staking offers to users within the application. For a more detailed description of staking itself, please visit [Staking](/token/staking). For each user, a staking score is calculated based on the value of the staked tokens and the lock duration. This score allows a user to achieve ranks that unlock features and items in the application. This is a great way to increase the work capacity and thus enhance the potential rewards. ## Ranks
## Summary Each staking pool will provide the following amount of staking points: * 1 month pool: 1 $TADA locked = 0.75 staking point * 3 month pool: 1 $TADA locked = 1 staking point * 6 month pool: 1 $TADA locked = 1.5 staking points * 12 month pool: 1 $TADA locked = 2 staking points The table below provides an approximation of the rank based on the value of tokens staked and the lock duration. Each row corresponds to a rank, and each column indicates the dollar value of tokens that need to be staked in the pool to reach that rank.
Rank1 month lock3 months lock6 months lock12 months lock
Rookie>0$>0$>0$>0$
Adventurer667$500$334$250$
Holder1334$1000$667$500$
Hero3334$2500$1667$1250$
Legend6667$5000$3334$2500$
Whale13334$10000$6667$5000$
Here are a few examples to better understand: * A user who has staked one token is automatically a Rookie, regardless of the lock duration. * A user who has staked $500 worth of tokens in the "3 months lock" pool is an Adventurer. * If the same user had chosen to stake the same amount of tokens in the "6 months lock" pool, he would have achieved the higher rank of Holder. # Ta-da Bondz (T-Bondz) **Daily $TADA Rewards Distribution Overview** \ Each day, $TADA rewards are allocated to users based on the XP they earn by completing tasks within the app. Rewards are not guaranteed for all users, and the amounts distributed depend on multiple factors such as activity level and ecosystem contributions. \ Rewards are issued in the form of **Ta-da Bondz (T-Bondz)**, which are represented as **on-chain Non-Fungible Tokens (NFTs)**. Each T-Bondz grants its holder the right to claim a specific amount of $TADA. 1. **Vesting and Accrual**: * The $TADA associated with a T-Bondz accrues monthly, following a predefined vesting curve. * This ensures rewards are distributed over time, aligning with long-term ecosystem goals. 2. **Redeeming T-Bondz**: * To claim the $TADA rewards linked to a T-Bondz, the holder must **redeem the bond**. * Upon redemption, the claimable portion of $TADA is transferred directly to the wallet holding the T-Bondz at the time of redemption. * Any unvested rewards at the time of redemption are **recycled into the next Prize Pool**, ensuring the sustainability of the rewards system. 3. **Flexibility and Transferability**: * T-Bondz can be redeemed directly through the app. * Alternatively, they can be withdrawn and transferred or sold to other users. * The new holder of the T-Bondz assumes all associated claim rights, making T-Bondz a **tradeable asset** within the ecosystem. This system ensures a balanced and flexible rewards mechanism, allowing users to participate in a manner that aligns with their individual preferences while supporting the long-term growth of the Ta-da platform.\ \ **Tada Bondz Parameters** | Parameter | Description | Base Parameters | | -------------- | ------------------------------------------------ | --------------- | | Nominal value | Amount of $TADA claimable at issuance | 12.5 | | Duration | Vesting duration in months, with monthly unlocks | 18 | | Vesting curve | Curve power coefficient “n” | 2 | | Terminal Value | Maximum amount of $TADA claimable | |

Fig.1: Vesting curve Example, over 5 periods

# Economy Our mobile application is founded on strong economic principles designed to ensure sustainability and growth. By integrating various agents and creating efficient money flows, we have developed a robust economic ecosystem that supports all participants. Ta-da is a play-to-earn platform, but it is distinctly backed by real value: users generate real value for companies, which pay them in return. This unique approach not only incentivizes active participation but also ensures that the rewards and benefits are tangible and meaningful. Our economic model is carefully crafted to balance the needs of businesses seeking quality task completion and the micro-taskers who contribute their skills and efforts. By fostering a fair and transparent economic environment, we empower users to earn and thrive through their dedication and hard work. This commitment to real value and sustainability sets Ta-da apart as a pioneering force in the Web3 space.\ \ The Economy of Ta-da can be divided into 4 sections : * [Agents](/token/economy/agents) * [Money Flow](/token/economy/money-flow) * [Rewards Distribution](/token/economy/rewards-distribution) * [Economy Breakdown](/token/economy/economy-breakdown) # Agents Here's a quick reminder of the various players who take part in Ta-da: * **The customers** are companies that need data. \ They publish their needs on Ta-da and give money to pay the community. * **The community** is composed of all our users around the world. \ They are paid to produce and verify data that meets customers' needs. * **The Ta-da platform** that connects customers with the community. \ It puts in place mechanisms to facilitate exchanges, encourage users to behave well, and so on. # Money Flow Ta-da redistributes money from customers to the community. This section describes how the redistribution system works. This is an important feature that has significant effects on user incentivization and enables the creation of a complete game economy, which will be discussed in more detail in the following sections. The diagram below shows the various cash flows through the application:

Money Flows

When a customer orders data on Ta-da, he pay in fiat. This money is converted into stable coin and distributed into four buckets: Ta-da's treasury, staking, the prize pool, and the burn. Each time a user validates a task, the compensation for the task is used to buy TADA tokens that are then placed in the prize pool. The more tasks are validated, the larger the prize pool and the more TADA tokens are purchased on the market. Every day, the prize pool is redistributed to users in a non-linear way: the top users earn a large part of the pool, while the last ones earn less. Part of in-app purchases, lost deposits and non-claimable rewards from redeemed T-Bondz are reinjected into the next prize pool. Another part is burned creating a deflationary economy and the rest is redistributed to the treasury and stakers. \ \ Such an economy impacts all the agents involved. The diagram and explanations below illustrate the positive interactions generated by this economy. On this one, each arrow means "increases," for example: more users increases the number of validated tasks, validated tasks increase the prize pool, etc.
Here are some explanations regarding these positive interactions: * The more jobs and the more users, the greater the number of validated tasks. * The more validated tasks, the larger the prize pool. * The more validated tasks, the larger the need for energy (tasks require energy to be filled). * The greater the need for energy, the more energy purchased. * The more energy purchased, the larger the prize pool. * The larger the prize pool, the more tokens users will earn. * The larger the prize pool, the greater the staking rewards will be. * The larger the prize pool, the greater the number of tokens that will be burned. # Rewards Distribution Every day, a new prize pool is created. Prize pools are capitalized based on the platform activity, and topped up with community incentives. Users earn experience points (XP) when completing taks. At the end of each period (day) users are ranked based on their XP gains. The prize pool is then distributed to users according to [these rules](/ta-da-platform/user-rewards/prize-pools). The remainder of this page outlines the main steps involved in calculating the rewards distributed to users. ## Parameters * $$to\_distribute$$: The amount of money to be distributed * $$percent\_to\_win$$: The percentage of winning users * $$decay\_factor$$: The exponential decay factor ## Invariants * A task completed by a user of rank 𝑥 should be better rewarded than a task completed by a user of rank 𝑥 + 1. ## Algorithm * Get the users selected to win the prize pool $$winners=users\_number \* percent\_to\_win$$. * For each user $$i$$, compute his exponential factor $$factor\_i=factor(len(winners), decay\_factor, i)$$. * For each user $$i$$, compute his score $$score\_i=factor\_i\*nb\_{tasks\_i}$$. * Compute the sum of the scores $$sum\_{score}=\Sigma score\_i$$. * For each user $$i$$, compute his score ratio $$ratio\_i=score\_i/sum\_{score}$$. * For each user $$i$$, compute his rewards $$reward\_i=ratio\_i\*to\_distribute$$. ## TADA vs Gems The distribution of TADA and Gems uses exactly the same algorithm. \ However, the quantity of Gems distributed and the number of winners are much higher. # Economy Breakdown In the Ta-da platform, the flow of funds is strategically distributed to ensure the sustainability of the ecosystem, reward user participation, and maintain the platform's operational needs. The distribution percentages vary depending on the source of the funds, such as job payments, in-app purchases, and lost deposits. Importantly, lost deposits are not redistributed to the treasury, highlighting that Ta-da has no financial incentive to slash unfairly users. Below is a detailed breakdown of how these funds are allocated:
Source of FundsCommunityStakeBurnTreasury
Jobs65%5%5%25%
In-app Purchases65%5%5%25%
Lost Deposits33%33%33%-
This allocation ensures that user engagement is rewarded, while also contributing to the platform's stability and growth through staking, burning tokens, and maintaining the treasury. # The Big Burn Launched during Phase 2, "The Big Burn" aims to create conditions and incentives required to scale the app beyond millions of users, and grow our datasets to the critical size required to scale. As part of "The Big Burn" campaign, tokens coming from the following slots are burned evey day. | Slot | Min. Burn allocation | | --------- | -------------------- | | Team | 12% | | Advisory | 28% | | Ecosystem | 44% | | Growth | 36% | | Treasury | 36% | The amount of $TADA tokens burned each day varies based on several factors such as App usage growth, with a minimum of 18% in total, and 0.2% per day to be burned during the first 3 months of "The Big Burn" program. # Token Economics ## Token Utility $TADA is a Utility token which primary purpose is to buy datasets. In addition, holding $TADA provides benefits to platform users such as: * tiering on rewards * access to special jobs * staking to secure consensus and earn passive income ## Buybacks TA DA Foundation performs buybacks of $TADA tokens based on datasets sales activity. $TADA bought back can then be offered as rewards to stakers, or for Ta-da specific campaigns ## Distribution There is originally a maximum number of 1,000,000,000 $TADA tokens, divided into several slots with specific uses (v1 Supply). A total of 186,000,000 $TADA have been irrevocably burned as part of "the Big Burn", a tokenomics remodeling campaign that happened in 2024-25 (v2 supply). | Slot | Supply (original) | Supply (v2) | | ------------- | ----------------- | ----------- | | Token Sales | 27.57% | 27.57% | | Growth / Team | 17% | 14.84% | | Advisors | 5% | 3.56% | | Ecosystem | 13% | 7.24% | | Community | 13.43% | 13.43% | | Treasury | 20% | 12.8% | | Growth | 4% | 2.56% | | Burned | | 18% |

Token releases schedule

*NB: Token metrics were amended on Nov. 15th 2024. Original token metrics remain accessible on* [*this link*](https://docs.google.com/spreadsheets/d/1rggwXj1fMPL1xIT2zeyyLWtxCPYrQYmOsOFCmyL_jCs/edit?usp=sharing)*.* # Staking ## Presentation Ta-da's staking solution rewards users who have trust in the project and its token. Users aiming to earn rewards are required to lock their tokens for a predetermined period. In general, the more extended the staking duration, the higher the potential rewards. Ta-da features four unique staking pools, each with its own unique characteristics, giving users a variety of options to choose from based on their staking preferences.
## Properties The main properties of our staking system are: 1. Two users who stake the same amount of tokens for the same lock period get the same amount of reward. 2. The reward depends on the number of tokens staked. A user who stakes twice as many tokens as another user in the same stake earns twice as much reward. 3. A token blocked in a stake with a ratio of 2 must earn twice as much reward as a token blocked in a stake with a ratio of 1. 4. No tokens are created or destroyed during redistribution. More details on : * [On-chain Staking](/token/staking/on-chain-staking) * [Meria staking](/token/staking/meria-staking) * [xExchange Metastaking](/token/staking/xexchange-metastaking) # On-chain Staking Users can stake $TADA and earn APRs on .
*** ## Staking description Here is a description of each staking pool:
Pool IDLock durationRatio
112 months8
26 months4
33 months2
41 months1
APRs are indexed to the platform's activity. Here are the different sources of revenue that will be injected into the staking: * 33% of deposits lost by users * 5% of our customers' cash inflows * 5% in-app purchases Applicable staking APRs are revised on a monthly basis and can be viewed on our staking platfom . ## ## Properties The main properties of our staking system are summarized below: 1. Two users who stake the same amount of tokens for the same lock period get the same amount of reward. 2. The reward depends on the number of tokens staked. A user who stakes twice as many tokens as another user in the same stake earns twice as much reward. 3. A token blocked in a stake with a ratio of 2 must earn twice as much reward as a token blocked in a stake with a ratio of 1. 4. No tokens are created or destroyed during redistribution. {% hint style="info" %} **Example** A user who stakes his tokens in the pool 4 (ratio of 1) will have to wait a month to be able to recover his locked tokens. During this period, he will earn rewards. Another user who stakes the same amount of tokens in the pool 1 (ratio of 8) will have to wait a year to get his tokens back. However, during this period, he will earn 8 times more tokens than the first user (if the latter were to stake his tokens every month in the pool 4 for one year). {% endhint %} # Meria Staking The Ta-da staking is available on
Ta-da offers a solution to stake $TADA tokens directly on Meria and benefit both staking rewards and in-game bonuses. For more details about staking rules, please visit the [Rewards & Staking](/ta-da-platform/user-rewards/rewards-and-staking) page. Users have access to 3 different staking pool with different parameters
Pool IDLock durationAPR (untill 09/08/2024)Ratio
112 months60%8
26 months30%4
33 months15%2
{% hint style="warning" %} **All APR are guaranteed until August 9, 2024** and will be dynamic thereafter and may decrease. Once the application is released, the APR will only be indexed to the platform's activity. Here are the different sources of revenue that will be injected into the staking: * 33% of deposits lost by users * 5% of our customers' cash inflows * 5% in-app purchases {% endhint %} In order to access those pools, users can either buy tokens directly on Meria or deposit their own tokens on the platform. Once the users have staked any amount of $TADA, they are entitled to redeem a code from their Dashboard section. They can then copy and paste the code within the Ta-da app to enjoy in game bonuses. # xExchange Metastaking The Ta-da staking is available on[ https://xexchange.com ](https://xexchange.com)

xExchange Metastaking

MetaStaking on MultiversX allows token holders to earn rewards by participating in the MultiversX ecosystem, without the complexities of traditional staking. By blocking your tokens for ten days, you can earn 5% APR in rewards. The Metastaking is already live on [xExchange](https://xexchange.com/metastaking). # Buy $TADA Tokens ## Where to buy our tokens? You can now purchase our tokens on two platforms. \ These platforms offer a secure and user-friendly experience for all your transactions. ### -> Buy on Kucoin ### -> Buy on PancakeSwap ### -> Buy on xExchange # Team ## TEAM 💜 Alongside our founders, a team of 15 people is building Tada everyday: Developers, Marketing specialists, BDs, Community managers,... 💜 ## ADVISORY BOARD

Hasheur
CEO, Meria

Serial entrepreneur, founder of Meria (leading FR CEX) and prominent KOL in France

Linkedin

Danilo S Carlucci

CEO, Morningstar Ventures

Backed many big crypto projects (StepN, Axelar, Holoride, etc.)

Linkedin

Luc Julia
Creator of Siri (Apple)


Co-founder of Nuance (acquired by Microsoft for $20B in 2021)

Linkedin