Increasing adoption of federated learning in privacy-sensitive applications, such as healthcare, IoT, Cyber Security, and Financial Systems, calls for strong mechanisms of ensuring integrity, trust, and security of the data samples. Traditional FL frameworks are vulnerable to attacks by adversaries, model poisoning, leakage of privacy, and inefficiencies in distributed training, which severely impose limits on their scalability and applicability. Blockchain technology is a decentralized, immutable, and tamper-resistant solution that makes FL more secure and trustworthy. However, despite a large number of publications on BFL, the current reviews do not provide a comprehensive and systematic taxonomy of methods that iteratively secure FL deployments using blockchain and address key technical trade-offs. This work presents an Iterative Comprehensive Taxonomy of Securing Federated Learning Using Blockchains, systematically surveying 25 state-of-the-art BFL models across diverse applications. Compared based on a PRISMA assessment, the models are looked at according to trust management, hierarchical service optimization, privacy-preserving FL, adversarial defense, and reinforcement learning security sets. The study brings up best models that suit various application domains, like healthcare, IoT security, vehicular networks, and self-learning AI systems. The results show that security and privacy-preserving cryptographic techniques, incentive-driven participation mechanisms, and optimized edge computing frameworks are essential to enhancing the resilience of FL against security threats. This work provides a technical roadmap for future breakthrough in BFL, supporting secure, scalable, and privacy-preserving federated learning deployments. This taxonomy can be used as a benchmark for future research bridging the gap between theoretical breakthroughs and real-world implementation in distributed ML mechanisms.