Build a data validation pipeline that quarantines bad records instead of crashing on them

Build a validation layer for the data pipeline below. The goal is a pipeline that tells you exactly which records are bad and why — not one that either crashes on the first bad row or silently accepts everything.

DATA SOURCE
A nightly batch CSV upload of ~50k customer records, processed as one DataFrame.

VALIDATION RULES
email must be a valid format; signup_date can't be in the future; country must be a valid ISO code

FAILURE POLICY
quarantine invalid rows and continue processing the rest, but abort the whole batch if over 10% of rows fail

REQUIREMENTS
1. Define the expected schema explicitly — a Pydantic model if records are processed one at a time, or a pandera DataFrameSchema if this is a bulk DataFrame pipeline. Don't validate ad hoc with scattered if checks; one schema is the single source of truth for what "valid" means.
2. Validate every record or row against the schema and classify each as valid, or invalid with the specific rule or rules it broke — not just "invalid." A record failing three rules should report all three, not just the first one that failed.
3. Apply quarantine invalid rows and continue processing the rest, but abort the whole batch if over 10% of rows fail precisely: if it says quarantine bad records and continue, invalid rows go to a separate output (a rejects table or file with the original data plus the reason) and valid rows proceed — the pipeline does not fail the whole batch for a few bad rows. If it says fail the batch above a threshold, implement that threshold explicitly rather than an arbitrary cutoff you invented.
4. Log a run summary: total records, valid count, invalid count broken down by which rule failed most often — this is what tells a human whether a new failure pattern just appeared upstream.
5. Never let a validation failure raise an unhandled exception that kills the whole pipeline run unless quarantine invalid rows and continue processing the rest, but abort the whole batch if over 10% of rows fail explicitly says that should happen for that condition.

OUTPUT FORMAT
1. The schema definition.
2. The validation and routing function (valid versus quarantined).
3. The run summary logger.
4. One worked example: a batch with a mix of valid and invalid records, showing what gets quarantined and why.