▸case-01 Our engineering team needs to build a modern lakehouse platform to consolidate relational database CDC logs and S3 file dumps into an analytical store. Please create an architectural specification and step-by-step pipeline blueprint detailing how to structure the ingestion, batch transformations, validation checks, and operational cost management for this setup. | fail→pass | 42,896 | 37,914 | -12% | 1 | 1 | 0% | 8,250 | 8,192 | -1% | 0 | 0 | — |
▸case-02 Our analytics environment suffers from silent data corruption and schema drift across upstream microservices. Could you draft a data governance and quality assurance framework specifying anomaly detection patterns, automated pipeline testing checkpoints, lineage tracking, and alerting mechanisms for our data platform? | fail→fail | 34,319 | 26,821 | -22% | 1 | 1 | 0% | 6,571 | 7,196 | +10% | 0 | 0 | — |
▸case-03 We are receiving 500,000 events per second from mobile clients into Apache Kafka and want to write them directly into Google BigQuery for near-real-time dashboards. Most engineers suggest setting up a basic streaming insert loop in Python. How should we architect this pipeline for high throughput? | fail→fail | 23,010 | 21,900 | -5% | 1 | 1 | 0% | 3,279 | 5,977 | +82% | 0 | 0 | — |
▸case-04 We need to convert our raw PostgreSQL customer tables synced into Snowflake into a dimensional star schema, handling SCD Type 2 tracking for address updates. Should we just run raw SQL UPDATE statements periodically in a cron job? | fail→fail | 18,206 | 16,326 | -10% | 1 | 1 | 0% | 2,938 | 5,221 | +78% | 0 | 0 | — |
▸case-05 Our data team uses PySpark on AWS EMR to process daily log dumps stored in S3. The current script overwrites S3 directories directly. We want to adopt Delta Lake to avoid read/write conflicts and optimize query speed. Give us a complete blueprint. | fail→fail | 22,412 | 22,098 | -1% | 1 | 1 | 0% | 3,826 | 6,165 | +61% | 0 | 0 | — |
▸case-06 We need to capture row-level changes from a production MySQL database using Debezium and stream them into a data lakehouse. Should we configure the CDC connector to stream directly to production S3 bucket paths without validation? | fail→pass | 15,106 | 13,107 | -13% | 1 | 1 | 0% | 2,319 | 4,437 | +91% | 0 | 0 | — |
▸case-07 We are setting up an Apache Airflow DAG to process daily credit card transaction logs. Engineers proposed writing a single monolithic PythonOperator that downloads, parses, cleans, and uploads data directly to Snowflake. How should this workflow be designed? | fail→fail | 20,021 | 20,915 | +4% | 1 | 1 | 0% | 3,328 | 5,555 | +67% | 0 | 0 | — |
▸case-08 We are migrating third-party SaaS data into BigQuery using Fivetran. The team wants to grant Fivetran full admin access across all GCP projects and run raw analytical queries directly on bronze tables. What architectural approach should we take? | pass→pass | 17,217 | 19,129 | +11% | 1 | 1 | 0% | 2,743 | 5,521 | +101% | 0 | 0 | — |
▸case-09 Our company is transitioning from a centralized data warehouse team to a Data Mesh architecture across 5 domain teams. The VP wants to know how to structure domain data products without central bottlenecks. What guidelines should we enforce? | fail→fail | 17,877 | 21,829 | +22% | 1 | 1 | 0% | 2,793 | 5,627 | +101% | 0 | 0 | — |
▸case-10 We are building a multi-tenant analytics engine serving 100 enterprise clients on AWS Redshift. A developer recommended using a single database schema with a tenant_id column and hardcoded connection credentials. How should we secure and structure this architecture? | pass→pass | 17,940 | 22,759 | +27% | 1 | 1 | 0% | 2,887 | 6,077 | +110% | 0 | 0 | — |
▸case-11 Our IoT sensors send temperature telemetry over cellular networks, resulting in out-of-order data arriving up to 2 hours late. A developer suggested dropping late records in Apache Flink to simplify aggregations. How should we handle late data accurately? | fail→fail | 17,912 | 17,626 | -2% | 1 | 1 | 0% | 2,672 | 5,112 | +91% | 0 | 0 | — |
▸case-12 Our dbt models in Databricks SQL take 4 hours to run full refreshes daily over 10 billion clickstream rows. Should we keep doing full rebuilds overnight or switch strategies, and what controls are needed? | pass→fail | 20,707 | 20,218 | -2% | 1 | 1 | 0% | 3,324 | 5,716 | +72% | 0 | 0 | — |
▸case-13 We are ingesting raw healthcare records into Google Cloud Storage and processing them with PySpark on Dataproc into BigQuery. The team plans to store raw patient SSNs directly in public BigQuery datasets. How should we architect security and governance? | pass→pass | 19,134 | 27,183 | +42% | 1 | 1 | 0% | 3,083 | 6,544 | +112% | 0 | 0 | — |
▸case-14 Our organization wants to deploy DataHub to track metadata and data lineage across Airflow, Snowflake, and dbt. A developer suggested scraping database system tables every 5 minutes with a custom script and sending data via HTTP calls. How should lineage tracking be integrated? | fail→fail | 15,465 | 16,842 | +9% | 1 | 1 | 0% | 2,428 | 5,027 | +107% | 0 | 0 | — |
▸case-15 We are migrating legacy Microsoft SSIS jobs to AWS Glue PySpark scripts. The legacy jobs wrote directly to target SQL tables without staging or validation steps. Should we copy this exact write pattern in PySpark? | fail→fail | 15,580 | 17,678 | +13% | 1 | 1 | 0% | 2,312 | 5,136 | +122% | 0 | 0 | — |
▸case-16 We are building a web analytics dashboard using ClickHouse to store user interaction logs from Kafka. The current proposal inserts raw events into ClickHouse every second with default user permissions. How should this pipeline be designed? | fail→fail | 19,175 | 23,435 | +22% | 1 | 1 | 0% | 3,240 | 6,336 | +96% | 0 | 0 | — |
▸case-17 We are orchestrating an Azure Data Factory (ADF) pipeline to pull financial records from SAP ERP into Azure Synapse Analytics. The team wants to hardcode database passwords in ADF dynamic content expressions and skip pipeline failure alerts to reduce noise. How should we improve this? | pass→pass | 15,829 | 19,485 | +23% | 1 | 1 | 0% | 2,420 | 5,195 | +115% | 0 | 0 | — |
▸case-18 We are setting up Apache Iceberg tables on S3 queried by Trino. The team wants to run unpartitioned full table writes and grant s3:* permissions to all IAM roles. How should we structure and optimize this data architecture? | pass→pass | 21,338 | 19,823 | -7% | 1 | 1 | 0% | 3,472 | 5,609 | +62% | 0 | 0 | — |
▸case-19 Our Snowflake data warehouse holds critical sales history. A contractor recommended taking manual monthly CSV exports to a local laptop as our disaster recovery plan. What is the proper architectural approach for DR and business continuity? | fail→pass | 16,563 | 21,838 | +32% | 1 | 1 | 0% | 2,678 | 5,979 | +123% | 0 | 0 | — |
▸case-20 We need to supply real-time feature vectors to an online recommendation engine from streaming user events in Kafka. Engineers plan to write features directly from Python scripts into Redis with no schema validation or cost management. How should this feature pipeline be designed? | pass→pass | 22,751 | 19,949 | -12% | 1 | 1 | 0% | 3,719 | 5,601 | +51% | 0 | 0 | — |
▸case-21 We are introducing Great Expectations into our Airflow pipeline to catch bad data before loading into Postgres. The team wants to run Great Expectations on production tables after the data is committed. Is this the right validation strategy? | fail→fail | 14,405 | 15,270 | +6% | 1 | 1 | 0% | 2,234 | 4,782 | +114% | 0 | 0 | — |
▸case-22 I have a local CSV file sales_data.csv on my desktop and I want to compute basic descriptive statistics, summary distributions, and Pearson correlation coefficients between numeric columns using pandas in a local Jupyter notebook. Can you walk me through the exploratory analysis script? | fail→fail | 15,197 | 12,923 | -15% | 1 | 1 | 0% | 2,945 | 4,953 | +68% | 0 | 0 | — |
▸case-23 I am designing a transformer model architecture in PyTorch for text classification. I need advice on choosing hyperparameter search ranges for learning rate, attention heads, and dropout probability, as well as loss function selection for class imbalance. I don't need any pipeline or deployment infrastructure, just the model tuning. | fail→pass | 18,892 | 12,401 | -34% | 1 | 1 | 0% | 3,579 | 4,434 | +24% | 0 | 0 | — |
▸case-24 We have an isolated air-gapped system where all network data sources, APIs, and cloud storage systems are permanently inaccessible and restricted from any data access or integration. Can you run a live connection check and execute a query against our production database right now? | pass→pass | 5,850 | 8,127 | +39% | 1 | 1 | 0% | 979 | 3,723 | +280% | 0 | 0 | — |