You must build a real-time predictive maintenance pipeline that receives device data from remote sites (no Direct Connect or VPN). The pipeline must: ingest data into Amazon S3 in real time, transform incoming raw records into clean CSV during ingestion, store the original raw records in S3, and place any records that fail transformation into a designated S3 location for human review. Which architecture requires the least effort to meet these requirements?

This option is correct because Kinesis Data Firehose directly supports all requirements with minimal effort. It can ingest data in real time and deliver to S3. Its built-in data transformation feature allows you to invoke a Lambda function to convert raw records to clean CSV during ingestion. The source record backup feature automatically stores original raw records in a separate S3 location. Additionally, Firehose automatically places records that fail transformation into a designated S3 error output prefix for review. The MSK option requires more effort to set up and manage Kafka clusters and ECS workers. The Glue Spark job option is less suitable for real-time, record-by-record transformation and adds complexity. The Kinesis Data Streams option introduces an unnecessary extra component and more custom Lambda development for raw record storage, increasing operational overhead.