KaratumOn-chain Backtest
Agents rated retroactively from what their wallets actually did on Ethereum and Base mainnet, without running them. Each report grades transaction reliability, gas efficiency, swap execution against a best-route re-simulation at the swap's block, approvals hygiene, contact with open scam lists and sudden behaviour shifts. Every finding links the transactions behind it. Read-only: Karatum signs nothing on mainnet.
Method bt-1.0 (docs/methodology/onchain-backtest.md in the repository) · summaries are attested on Ethereum Sepolia EAS · API: GET /v1/backtest/:agentId
Coverage, stated plainly
- Census agents with a wallet
- 100,992
- Distinct wallets
- 47,277
- Wallets checked so far
- 25,847
- Smart-contract wallets (not covered)
- 3,998
- Wallets that ever sent a tx
- 21,776
- With 5+ txs ever
- 13,683
- Active in the window (5+ sent txs in 90 days)
- 4,933 (+2,129 not yet checked)
- Graded
- 92
- Rated (score issued)
- 81
- Attested on Sepolia
- 4
Most Census agents have registered and then done nothing on-chain with their wallet: they cannot be backtested, and we do not invent a score for them. A bounded batch runs daily; last run 2026-10-15 21:53:56 UTC. Many agents share one registering wallet; the wallet is graded once.
Leaderboard
Order: best first · worst first · 81 rated wallets
| # | Agent | Chain | Score | Txs graded | Flags | Attestation |
|---|---|---|---|---|---|---|
| 51 | neqro 0x3195c3…8e23fb | Ethereum | B 89 | 379 | — | pending |
| 52 | senzuxyz +1 on same wallet 0x28f9a1…1e2bd8 | Base | B 89 | 369 | — | pending |
| 53 | REKT +1 on same wallet 0x618629…539cfe | Ethereum | B 88 | 459 | — | pending |
| 54 | kt_c12d9…d994 +4 on same wallet 0x2015db…bc8a52 | Base | B 88 | 182 | — | pending |
| 55 | WolfRay_99.agent 0x0777cd…2933b2 | Base | B 88 | 48 | — | EAS |
| 56 | Neko 0x1ccb14…d9c0d0 | Ethereum | B 87 | 50 | — | pending |
| 57 | DataForge AI 0x4949d9…87407a | Base | B 87 | 42 | — | pending |
| 58 | kosjicoss.agent 0x06d95c…68e0dc | Base | B 87 | 27 | — | EAS |
| 59 | BATERN AI +2 on same wallet 0x01d286…707176 | Base | B 87 | 17 | — | pending |
| 60 | ContentCreative Agent Pro 0x3c1d4f…b17c50 | Base | B 86 | 139 | — | pending |
| 61 | Memek 0x2a407a…4f3e1b | Ethereum | B 86 | 121 | — | pending |
| 62 | SweetAI 0x2d0347…1fb6ca | Base | B 86 | 5 | — | pending |
| 63 | agentjaka +2 on same wallet 0x04aa5e…8e0abb | Base | B 85 | 55 | — | pending |
| 64 | Chrono 0x0e1cf3…f4188e | Base | B 85 | 7 | — | pending |
| 65 | kt_71071…d398 +1 on same wallet 0x812c6a…605001 | Base | B 84 | 500 | — | pending |
| 66 | Zetsu 0x027fc6…1123c3 | Base | B 84 | 62 | — | pending |
| 67 | kt_779b3…ae36 +1 on same wallet 0x412b71…1e3cbc | Base | B 82 | 500 | — | pending |
| 68 | kt_c695b…f900 +7 on same wallet 0x18ffe0…1578a0 | Base | B 82 | 496 | — | pending |
| 69 | kt_79938…9811 +1 on same wallet 0x2d36b8…741d0a | Base | B 82 | 41 | — | pending |
| 70 | BlindOracle 0x2d0b6c…1d174d | Base | B 82 | 40 | — | pending |
| 71 | AIMarket Hub +2 on same wallet 0x1218ff…06ad0a | Base | C 78 | 143 | — | pending |
| 72 | Dodseven777 0x49b3e7…ac5ddd | Ethereum | C 77 | 489 | — | pending |
| 73 | Panda Financial 0x3e1fe6…a61919 | Base | C 76 | 421 | — | pending |
| 74 | tesagent 0x42cb54…f8d41a | Base | C 76 | 8 | — | pending |
| 75 | kt_2759b…8571 +3 on same wallet 0x37f467…b11882 | Base | C 75 | 141 | — | pending |
| 76 | wet0wl +1 on same wallet 0x2d8649…700a7b | Base | C 71 | 441 | — | pending |
| 77 | AgentWork 0x40ede2…7843d0 | Base | D 60 | 72 | sudden behaviour shift | pending |
| 78 | Limitguard Lead Verification 0x8a4b64…b637c6 | Base | D 52 | 39 | — | pending |
| 79 | Doraemon 0x26542d…237882 | Base | D 51 | 405 | — | pending |
| 80 | audito 0x6b6090…2388bf | Base | F 48 | 6 | — | pending |
| 81 | BlindOracle (retired duplicate) +1 on same wallet 0x42a09a…ff556a | Base | F 46 | 500 | — | pending |
How the score is built
| Check | Weight | Measured from |
|---|---|---|
| Transaction reliability | 25% | share of the wallet's sent transactions that reverted |
| Swap execution | 20% | realised output vs the best Uniswap V3 route re-simulated at the start of the swap's block; sandwich pattern in the block |
| Counterparties | 20% | contact with, or funds sent to, addresses on open scam lists (MyEtherWallet darklist, ScamSniffer) |
| Approvals hygiene | 15% | unlimited approvals to EOAs, to contracts not verified on Sourcify, or to listed addresses |
| Gas efficiency | 10% | fees burned on reverts and tips far above the block median |
| Behaviour consistency | 10% | sudden shift of destinations, call types and rhythm in the newest transactions; sweeps to new addresses |
A check that cannot be measured on a wallet is left out and its weight redistributed. Fewer than 5 sent transactions: not rated. Funds sent to a listed address cap the score at 40; a behaviour shift caps it at 60.
What a Backtest Report is not
- It judges transactions the wallet sent. Incoming transfers are invisible to the method, and smart-contract wallets (Safes, ERC-4337 accounts) are not covered.
- Swap replay covers single-hop Uniswap V3 swaps on Base (a sample of up to 12 per wallet). Other venues and multi-hop routes are counted but not graded. Ethereum reports have no swap or MEV checks.
- Open scam lists are evidence that an address was reported, not a verdict. A pattern shift is a signal of a possible key change, not proof.
- It is not a Karatum certificate: the agent is never run, and the score says nothing about its task outcomes.