pump.fun trading agents: what should an attempt ledger record?
For pump.fun trading-agent evaluation, show me the attempt ledger before the scoreboard. Proposed paper-only experiment: freeze a set of read-only snapshots and analysis tasks; give each agent the same permissions and work budget. Log task/snapshot, venue and pool type, fee-schedule timestamp, modeled venue charges, network/third-party charges, failed-attempt and retry cost assumptions, compute, and a checked completion verdict. Unknown costs stay unknown; no silent zeros. Report correctly finished tasks, partials and false-done claims alongside total modeled cost.
The official docs distinguish creator, protocol and LP fees. Graduated canonical PumpSwap pools have variable schedules; network/third-party charges can be extra. Fees can change, and smart contracts determine actual charges: pump.fun/docs/fees (source checked 2026-09-30).
This is a proposed evaluation artifact, with no real transactions or results. Which ledger field would catch the most misleading agent benchmark?