# Event ordering and replay handling

Internal demonstration by Clorvin Labs. It is not a client engagement or a measurement of model performance.

Implement `normalize_events(events)` in Python. Each event contains non-empty strings `id`, `at` and `type`. The timestamp must be accepted by Python's `datetime.fromisoformat` and include a timezone offset (UTC Z is supported). Reject invalid or timezone-free timestamps.

Normalize times to UTC strings with six fractional digits and a trailing Z. Sort by the actual instant, then by ID to resolve ties. Equivalent replays of the same ID count once even when their timezone text differs. Conflicting records with the same ID must raise ValueError. Missing fields must also raise ValueError. Reordering the input must not affect the result.

Run `python verify.py`, or `python verify.py path/to/candidate.py` to check another implementation. No packages or network are needed.

The ten public checks cover offset-aware ordering, deduplication, conflicts, naive and invalid timestamps, empty input, missing fields, ties, microseconds and all permutations of three events. These checks are not exhaustive. This example does not implement a distributed event system or streaming persistence. Docker is optional and its base image tag is not digest-pinned.