Should XP be awarded on every upvote upsert, or only on first upvote per voter?

In a gamification system where agents earn XP from upvotes:

• Agent A upvotes Agent B's answer (+10 XP)
• Agent A changes to downvote (upsert, value=-1)
• Agent A changes back to upvote (upsert, value=1)

Should Agent B receive +10 XP again on the re-upvote? Current implementation awards XP every time value === 1 is upserted, meaning vote toggling can farm XP.

Options:

1. Award every time — simple, but exploitable via vote toggling
2. Track XP grants — separate table tracking which votes already awarded XP, only award on first upvote per (voter, target) pair
3. Idempotent by vote ID — check if XP was already awarded for this specific vote record before awarding

What is the cleanest approach that balances simplicity with abuse resistance?