FICTIONAL TRAINING-HISTORY MIGRATION — 12 October 2026
All six activities and destination observations are invented. No account or
workout file was imported. These are reconciliation examples, not service tests.

Run locally with Python 3.9 or later (standard library only):
  python3 verify_migration.py expected.csv first-pass.csv
  python3 verify_migration.py expected.csv reconciled.csv
The first command exits 1 with exactly three problems, although both counts
are six. The second exits 0. Different local dates with equivalent UTC instants
pass; timezone-less timestamps are rejected.

COLUMNS
source_key: a manually reviewed mapping to one original activity. The script
does not discover identities or read destination identifiers.
start_time: ISO 8601 with Z or an explicit UTC offset.
sport: a manually normalized category; it must match exactly.
elapsed_seconds: elapsed duration, not moving time; tolerance 1 second.
distance_meters: known source distance; tolerance 1 metre. A blank source value
means unknown/not required, not zero. Destinations that recalculate distance
need a separately justified reconciliation policy.
has_route / has_heart_rate / has_power: 1 means present, 0 absent. Required source
fields must be present. These flags do not prove values, units, sample counts,
sampling times, sport-specific details, notes, photos or social content match.

Keep your real source/destination ledger private. The checker has no network
or account access. Passing its bounded checks is not proof of upload acceptance
or a complete migration; inspect the destination and important streams too.