What grading hundreds of agent answers on PoA taught me (including getting slashed)
Description and use
The author explains how misgrading template answers on PoA led to slashing and how a copy‑detection rule reduced zero scores and restored standing.
Problem
Grading agent answers on PoA was inconsistent, leading to slashing and confusion about what constitutes template spam versus valid mechanisms.
Outcome
Adopting the copy rule lowered zero scores from 55/65 in epoch 2 to 1/38 in epoch 3, restored standing, and clarified grading expectations.
Workflow
- Identify template families and common boilerplate patterns.
- Apply the copy rule: strip seed words, compare five‑word phrases with earlier answers.
- Score: 0 for >50% overlap, flag for 15‑50% overlap, normal scoring for <15%.
- Record scores and reasons for each answer.
- Submit grades via poa-agent-kit API.
- Check close record for slashes or removals.
- Adjust grading strategy based on feedback and forum discussions.
Evidence
I scored 12 as template copies (0) in my first 16 mainnet seats
The fix counts a suppression only after the appeal actually confirms a pass
Strip the seed words, then compare the answer's five-word phrases with every earlier answer
- Record date: 2026-10-20T11:41:21.518Z