The correct answer is A: Records that violate the expectation are added to the target dataset and recorded as invalid in the event log.

Explanation

In Delta Live Tables (DLT), when you define a data quality constraint using only the CONSTRAINT ... EXPECT (condition) clause (without an ON VIOLATION clause), the default behavior is "warn" (also called "retain invalid records").

• What happens to violating records?
  They are kept (added to the target dataset) along with the valid records. The pipeline does not drop them or fail.

• What happens for monitoring/auditing?
  DLT records the violation as a data quality metric (number of rows that passed vs. failed the expectation). These metrics, along with details about the violation, are logged in the DLT event log (a Delta table that you can query for pipeline observability).

This is different from the other actions you can explicitly specify:

• EXPECT ... ON VIOLATION DROP ROW (or @dlt.expect_or_drop in Python) → Drops the bad records from the target but still logs the count in the event log (this matches option B, but is not the case here because no ON VIOLATION is present).
• EXPECT ... ON VIOLATION FAIL UPDATE (or @dlt.expect_or_fail) → Causes the update/flow to fail (this matches option C).
• There is no automatic quarantine table unless you implement custom logic yourself (option E is incorrect).
• No extra "flagged as invalid" column is automatically added to the target (option D is incorrect).

Official Behavior Summary (from Databricks docs)

Since the question shows only CONSTRAINT valid_timestamp EXPECT (timestamp > '2020-01-01') with no ON VIOLATION clause, it uses the default "warn" behavior → Answer A.

Quick Tip for the Exam

Always check the exact syntax in DLT questions:

• Plain EXPECT = keep + log (warn).
• Add ON VIOLATION DROP ROW = drop + log.
• Add ON VIOLATION FAIL UPDATE = fail the update.

This is a common exam trap—many candidates assume expectations are strict by default, but they are non-blocking unless you explicitly tell DLT otherwise.