Before assuming a tag is broken, confirm the two reports are describing the same thing. Publisher and third-party systems frequently use different timezones, different late-event allowances, and different definitions of what counts (served vs viewable impressions, deduplicated vs raw, filtered vs unfiltered for invalid traffic).
A single-digit percentage difference is generally considered standard and expected. The signal that points to a real cause is where the gap concentrates: a spike on the first day usually means the flight started at different local times, and a gap that appears at midnight points to a timezone boundary. A gap spread evenly across the flight is more likely a counting-method or filtering difference.
To reconcile, align the timezone, date range, and event definition on both sides first, then compare matched request IDs. Only after those match should you investigate deduplication, invalid-traffic filtering, or attribution windows as the remaining cause.