Explicit schema defined

Schema applied on read

No Python UDFs

Built-in functions used

No collect() on full data

Partitioned write by region

Coalesce or repartition after filter

Filter applied before aggregation

Column selection before aggregation

Broadcast join used

Broadcast on small table

Skew addressed

AQE enabled

Shuffle partitions tuned

Skew documented

Join strategy documented

Filter/select before join

No collect on large data

Watermark defined

Watermark duration matches late data

Checkpoint location set

foreachBatch used

Correct output mode

Explicit schema on read

Event time window used

Late data handling documented

Fault tolerance documented

Cache called on intermediate df

Materialized with count()

Unpersist called after use

Cached df reused for all reports

Output DataFrames not cached

Partition count printed

Explicit schema defined

No collect() on data

Caching rationale in notes

Configuration recommendations present

5 cores per executor

memoryOverhead configured

Partitioning strategy explained

Performance analysis present

Monitoring recommendations present

Broadcast join for merchant table

AQE enabled in script

Shuffle partitions tuned

No collect() on large data

Explicit schema or parquet read

RDD API used for log parsing

reduceByKey used (not groupByKey)

No groupByKey calls

Coalesce on output

No collect() on full data

Small file rationale in notes

Aggregation approach explained

Correct top-N extraction