Muse · Onchain build log · 7–9 Nov 2026

Token Sniper Build

A small, heavily-filtered bot that watches for new token launches, runs legit checks before it spends anything, and buys tiny amounts through the Privy agent wallet.

Update — 12 Nov 2026: v2 and v3 were retired and archived on 12 Nov 2026 (~12:52 GMT+7); their watcher jobs are off and the code sits in the archive, kept for the record. The copy-trader was already archived on 11 Nov 2026 and stays off — no copy trading. The only sniper running now is the Launch Sniper, live on Arc — real-time and worker-only. Everything below is the 7–9 Nov 2026 build record, kept as history.

Latest — 9 Nov 2026, ~16:40 GMT+7: the copy-trader's sell signal (mirror-exit) is LIVE — when any of the 8 watchlisted wallets sells a token we hold a copy position in, we sell our full position. The copy-trader went live at ~15:32 GMT+7, native-ETH copies at ~16:00 GMT+7, and its first live copy-buy landed minutes later: FAWBO at ~16:03 GMT+7, $0.02 native ETH, agent now holds ~771 FAWBO (tx 0xdcbde778…, block 109937492, confirmed). Its post-buy sell-route sim failed and flagged it sellBlocked, but the onchain check shows the FAWBO→Permit2 allowance is 0 — the native buy sets none — so this is likely the known sim false-positive, not a proven hook-block like debtdog; the real test is the first actual sell. The v1-five cleanup sold four tokens; debtdog stays held, sell-blocked. Gate review landed: min-tokens loosened from 500 to 250 in both v2 and v3. v2 and v3 remain LIVE, watchers running every 2 minutes, 0 open v2/v3 positions — debtdog is the only held v2-era sniping token, and FAWBO is the copy-trader's first position. Strategy from today: sniper only — the daily airdrop DYOR check is paused until the user says otherwise. Buys stay tiny: $0.02 each, cap 5, buy line 45, sell 50% at +100% / sell all at −50% — and from now on every number is reported NET of all fees, with hook-fee income tracked as its own line.

v1 RETIRED — archived 8 Nov ~09:08 GMT+7 · watcher cron disabled v2 + v3 RETIRED — archived 12 Nov 2026 · launch sniper is the only sniper running V1-five cleaned up 9 Nov — 4 sold · debtdog held (sell-blocked) Gate review applied — min-500 → 250 · age <24h · line 45 Copy-trader ARCHIVED — 11 Nov 2026 · no copy trading Sell signal LIVE — mirror-exit since ~16:40 GMT+7 · fee-sane bound 5%
$0.02per buy in v2, v3, copy-trader. Small buys only, by design — cap 5, buy line 45
1 held + 1 copydebtdog. The 9 Nov cleanup sold perp, Quesadilla, BAMBI and eve (all verified 0 onchain). debtdog's 968 tokens (~$0.034) stay held — its pool hook blocks our sell route. v2/v3 counters remain 0/0. Separately, the copy-trader's first live buy — FAWBO, ~771 tokens at ~16:03 GMT+7 — is flagged sellBlocked by the post-buy sim, likely a false-positive (allowance is 0 because the native buy sets none).
<24h · ≥250The two hard gates in v2/v3: the token's first pool must be under 24 hours old, and the $0.02 buy must yield at least 250 whole tokens (price ceiling ~$0.00008/token) — loosened from 500 on 9 Nov after the gate review. v1 never carried these gates.
+100% / −50%Sell half at 2×, sell everything at half price in v2 and v3. Otherwise hold. All sniper figures are now reported net of gas + LP + hook fees.

Signing wallet — Privy agent wallet (public address only): 0x0A1D2730f4Ae736Bac3D772FaB8607BFd7F650C4
No private key is ever in the bot. Signing happens through the Privy Agent CLI — the key is held by Privy and never seen by Muse.

9 Nov: v1-five cleaned up · gate review applied · copy-trader live · sniper-only

  1. V1-five cleanup — 4 sold, debtdog held (~13:15 GMT+7, user order). On the user's order ("Sell it all the tokens from snipping, clean it up"), perp, Quesadilla, BAMBI and eve were sold — all four balances verified 0 onchain afterwards. Proceeds were 39,206,842,446,335 wei WETH, converted to +83,145 USDG raw ($0.083); WETH is back to its pre-existing dust and USDG now stands at 271,426 raw. Gas for the whole cleanup was ~0.0000289 ETH (~$0.065) — so the cleanup nets about +$0.018 after all fees.
  2. debtdog UNSOLD — sell-blocked by its pool hook. debtdog stays held: 968 tokens (~$0.034). Its pool hook 0xDE167EC2… blocks our Universal Router sell at any minimum-out — even 1 wei — and at half size, with both allowances confirmed set onchain. The pool itself is live (287 sells in 24h by others) and GoPlus calls the token clean — the hook blocks our route specifically, not all sells. Only one debtdog pair exists (the WETH pool), so there is no alternative route. It is the second sell-blocked token after ANT.
  3. Gate review applied — min-500 loosened to 250. The review pre-agreed on 8 Nov is now closed: the v1-five verdicts came in 5/5 survived (debtdog +50%; the other four sold at ~−2% token value). So the minimum-tokens rule loosened from 500 → 250 in both evaluators (MIN_BUY_TOKENS_WHOLE = 250, price ceiling ~$0.00008/token). The token-age gate (<24h) and the score line (45) are unchanged.
  4. Smart-money signal — live, but has fired zero times. Live in both evaluators since 9 Nov ~07:11 GMT+7 — and verified at ~14:17 GMT+7 it has fired zero times: 0 nonzero smart-money scores across 419 v2 + 32 v3 evaluations. It is a score boost only (+10/+15), never a buy trigger. The reason is overlap: the watchlist wallets buy established tokens and hold for days, while the sniper looks at minute-old launches.
  5. Copy-trader — LIVE since 9 Nov ~15:32 GMT+7. It mirrors the buys of the 8 watchlisted smart-money wallets on Robinhood, at $0.02 per copy-buy with its own cap of 5, reusing the sniper's proven buy paths. Every copied token still has to pass the same safety gates — GoPlus honeypot check, holder concentration, known-rug veto, and a transfer simulation. Dry-run tested first, then went live after review — 6 real watchlist buys were detected in dry-run and all correctly refused, $0 spent. Native-ETH copies went live ~16:00 GMT+7 (user go-live call): native pools are 58% of new Robinhood launches, and the native buy is a single transaction (router execute with value, no wrap/approve/Permit2) — dry-run proven on the live Steve pool, break-even 1.80× vs 2.18–2.31× for WETH copies. Caveat: the first live native sell is still unproven, and the post-buy sell-route sim can false-positive on a fresh position (no token→Permit2 allowance yet). First live copy-buy landed minutes after go-live: FAWBO at ~16:03 GMT+7 — $0.02 native ETH (tx 0xdcbde7786023293c1f4d8f4d729277db31cbaf6fde7cffb1e65ab9b42e492799, block 109937492, confirmed), agent holds ~771 FAWBO. The post-buy sell-route sim failed and flagged it sellBlocked — but the onchain check shows the FAWBO→Permit2 allowance is 0, which the native buy never sets, so this is likely that known false-positive, not a proven hook-block like debtdog. The real test is the first actual sell.
  6. Sell signal (mirror-exit) — LIVE since 9 Nov ~16:40 GMT+7. When any of the 8 watchlisted wallets sells a token we hold a copy position in, the copy-trader sells our FULL position (current agent balance) via the proven V4 / V4-native / V3 sell builders. Sells of tokens we do not hold are noted-only — the live cron already logged a real FISHIN sell that way. Sell signals attempt the sell even on a sellBlocked position: a real attempt is the only proof the hook blocks, and the post-buy sim is known to false-positive on fresh positions (no token→Permit2 allowance yet). Fee-sane bound (Part 60, live before go-live): mirror sells skip pools whose static fee exceeds 5% (50000) and skip dynamic-fee pools — the fee is unknowable before the sell. Correction: PUSSYCOIN's pool fee 8388608 is not 838.86%; it is the V4 DYNAMIC_FEE_FLAG (0x800000), the hook sets the fee per swap. Proof: sell-signal test suite 41/41. A mirror sell costs only the sell leg — ~$0.0106 gas + pool fee — so a $0.02 copy sold at 2× nets ≈ +$0.029, at 1× ≈ +$0.009 (break-even ≈ 0.53×). Current state: live=true, nativeEthLive=true, mirrorExit=true, 1/5 positions (FAWBO).
  7. Old builds archived — live code verified. 13 old files were moved (not deleted) to archive/v2-old-builds-20261109/: the Nov 7 v2 zip, old state/config backups, the dead v3-trade.js, and the v3 microtest/dry-run scaffolding. The live code was verified loading afterwards; v1 stays retired at archive/robinhood-v1-20261108/.
  8. Strategy — sniper only. From 9 Nov the sniper is the sole growth strategy. The daily airdrop DYOR check is paused until the user says otherwise.
  9. Accounting — everything net of fees. New rule: every sniper number is reported NET of all fees (gas + LP fees + hook fees), and hook-fee income is tracked as its own line. Current live state: v2 + v3 watchers running every 2 minutes; 0 open v2/v3 positions; debtdog is the only held sniping token.

v1 RETIRED — stopped and archived

Updated 8 Nov ~09:08 GMT+7. v1 is RETIRED. On the user's order ("Stop and archive v1"), the v1 watcher cron is disabled and the v1 code + state + logs are archived (not deleted) at workspace/tokens/sniper/archive/robinhood-v1-20261108/.

Why: v1 does not carry the Part 51 gates — token-age and min-500 are v2/v3 only — and minutes after the fresh-start re-enable it filled all 5 slots with a batch deploy: 4 of the 5 buys share the unidentified hook 0x2c668601… with identical ~16,251-token amounts. The 5 fresh-start buys were OPEN and UNMANAGED — perp, debtdog, Quesadilla, BAMBI and eve (08:57–09:00 GMT+7, 0.00001 ETH each). v1's sell rules (50% at +100%, all at −50%) no longer ran on them. Resolved 9 Nov ~13:15 GMT+7: on the user's order, four were sold in the cleanup; debtdog could not be sold — its pool hook blocks our route — and stays held. v2 and v3 remain live with the gates.

v2 + v3 LIVE — 0 open positions, debtdog the only held token

Updated 9 Nov ~16:00 GMT+7. Fresh sniping under the new rules — v2 + v3 only. The Part 51 build landed ~08:02–08:46 GMT+7 on 8 Nov, every test suite passed, and the watchers were re-enabled at ~08:56 GMT+7. v1 was retired minutes later (~09:08 GMT+7). v2 (Arc + Base + Robinhood + Polygon) and v3 (Robinhood, via the V3SwapHelper) remain live with the gates, watchers running every 2 minutes. v2/v3 counters are 0/0 with zero open positions; after the 9 Nov cleanup the only held sniping token is debtdog (968 tokens, sell-blocked). Buys stay $0.02 each, cap 5, buy line 45, with sell rules in v2/v3: sell 50% at +100%, sell all at −50%.

V3SwapHelper — Robinhood Chain (Sourcify exact match · owner = agent wallet): 0x3ecc88CBaD515D4F34dD365A109Cb7bB71a2FfcF

The new rules — Part 51 (built 8 Nov ~08:02–08:46 GMT+7)

The FOBO lesson — why these gates exist. FOBO was about 13 days old (its first pool dates to 26 Oct 2026), but its USDG pool was only 2 minutes old — the sniper triggered on a new pool, not a new token. The $0.02 buy received only ~2.43 tokens, and the USDG pool's liquidity was then pulled after the buy (0 liquidity left). Under the new rules FOBO is rejected by both gates. Replays of past candidates under the gates: DELTA V3 75 → SKIP (age 2,378.6h), GC V3 35 → SKIP, SI 10 → SKIP; CLAUDIA 30, FNUT 15, $SLNADO 15 and APU 30 unchanged.

Cleanup done — 8 Nov ~07:50–07:56 GMT+7: all six positions sold. FOBO was sold through the deep FOBO/ETH pool (sell tx 0xfb3d4d90…) because the USDG pool it was bought into had 0 liquidity left; v1's five (bottom, GOOGLECAT, Firstkille, SKIMS, Canary) went through sell-all — 15 transactions. The WETH proceeds were converted to USDG (+$0.108, tx 0x99e6bd52…). Agent wallet on Robinhood now: ETH ~0.000334, WETH dust 0.0000098, USDG $0.188281 — and all six token balances verified 0 onchain.

Registries live — Part 50. The launchpad + known-rug registries are in: 15 launchpads (Pons v1/v2, Flap, o1, Vectr, Hood, HedgeFun, Frontier, Duel, Argus ×5), so launchpad provenance is now a real +5 instead of a guess. The known scammer address 0x333089e0… and the fake-POL token 0xe174c72d… are a live known-rug veto. registries-test 98/98.

Per-chain balances: Robinhood — USDG 271,426 raw ($0.271), 4663 · live. After the 9 Nov cleanup: the four sold v1 tokens' proceeds were converted to USDG (+83,145 raw), WETH is back to pre-existing dust, and debtdog (968) is the only held token. | Arc — 0.818111 USDC, 5042 · live. | Base — 0.0004166 ETH, 8453 · live. Router V2 in place; first buy unblocked. | Polygon — 5.514 POL + 2.2667 WPOL, 137 · live since ~19:35 GMT+7. Buys GAS-SKIP while gas is above the 100 gwei ceiling (~271 gwei at enablement).

Scoring — what's live now

Base rubric (100): social-website 25, liquidity-depth 15, early-traction 15, verified-contract 10, lp-locked 10, deployer-history 10, no-impersonation 5, launchpad-provenance 5 (real, via the Part 50 registry). Layered on top: memory-adjustment ±25 from the deployer-memory loop (hourly follow-up cron; first 24h verdicts due ~09:00 GMT+7 9 Nov), GoPlus second opinion [−10, +5], holder-concentration [−10, +5] with the bundle check, and — since 9 Nov ~07:11 GMT+7 — a smart-money boost (+10/+15) when a watchlisted wallet has bought the token. That signal has fired 0 times so far (0 nonzero scores across 419 v2 + 32 v3 evaluations, verified ~14:17 GMT+7 9 Nov). Buy line 45, $0.02 per buy, caps 5; sell 50% at +100%, all at −50%.

Counters at the fresh start: v2 snipesDone was reset to 0/0 (was 5 on Robinhood / 2 on Arc), v3 at 0. v1 was also reset to 0, filled 5/5 within minutes (08:57–09:00 GMT+7), and was then retired at ~09:08 GMT+7. Its 5 positions stayed open and unmanaged until the 9 Nov cleanup sold four of them; debtdog remains held, sell-blocked. The caps (5 per chain / 5 for v3) still bound v2/v3 exposure.

Three bugs found & fixed during the Router V2 apply (~19:35 GMT+7): Buy minimum-out was 10¹² too big on mixed-decimal pairs (spotOutput decimal error — every spot-fallback buy would have reverted V4TooLittleReceived). Sell spot price had the same decimal error. Sell triggers were inverted (priceQuotePerToken branches swapped — a 2× pump would have read as −50% (sell ALL) and a 50% crash as +100% (sell 50%)). All fixed and verified. Sniper v1 was checked and is unaffected. Also fixed in the same pass: chains.js rawGetLogs now chunks eth_getLogs into ≤5,000-block ranges, because Polygon's publicnode RPC rejects ranges over 10,000 blocks.

Verdict & thoughts — 9 Nov 2026

Muse's honest assessment of the strategy, given to the user today — kept on the record alongside the numbers.

  1. The rules are right. Fees counted both ways, safety first on every contract, never reduce the portfolio — exactly correct for a ~$5 bankroll. One careless interaction wipes it out at this size, and you can't know whether you're growing without net-of-fees math. Those three are non-negotiable.
  2. The tension. Sniper-only plus never-reduce plus $0.02 a trade means the engine barely fires — zero buys in two days — and gas eats a quarter to half of each trade. Be clear-eyed: the sniper is a lottery-ticket machine, not a compounder. It only works if it hits one big winner, and that takes a lot of small shots over time. The min-500 → 250 loosening helps a little, but it doesn't change that math.
  3. One reservation. Pausing the airdrop check removes the only thing putting anything in the wallet without paying for it — free capital is the one thing that actually grows a tiny portfolio without risk. The "anything, anywhere, anytime" rule is the safety valve: if a genuinely free and positive opportunity appears, it will still be flagged rather than passed over in silence.

The verdict: "Good rules, right focus, but expect slow — the sniper needs one big hit, and that's a matter of time and volume, not effort."

How it works

Four parts. The watcher never sleeps as a background process — a scheduled job runs it every 2 minutes, so it survives machine replacements. State lives in a file, not in memory.

  1. 01 · WATCH — Watcher (run.js): Scans the Uniswap v4 PoolManager for new Initialize events since the last scanned block. A new pool = a new launch to check.
  2. 02 · CHECK — Evaluator (evaluate.js): Runs all six legit checks below — plus, in v2/v3, the score, the token-age gate and the minimum-tokens rule (250 since 9 Nov). Read-only — it sends no transactions, so dry-run mode is completely safe.
  3. 03 · BUY — Executor (trade.js): Buys via the proven Robinhood v4 swap route: wrap ETH → WETH, approve through Permit2, swap through the Universal Router. Bypasses the v4 quoter, which reverts on hooked pools.
  4. 04 · MANAGE — Position manager: Reads each held pool's current price on every run. Sells 50% at +100%, everything at −50%, otherwise holds and logs the decision.

Key contracts — Robinhood Chain: V4 PoolManager 0x8366a39CC670B4001A1121B8F6A443A643e40951 · WETH 0x0Bd7D308f8E1639FAb988df18A8011f41EAcAD73

The live gate: Live buying is controlled by one flag — state.live in STATE.json. The scheduled watcher can pass "live" all day, but the bot stays in dry-run until that flag is true, and it only became true after an explicit "Go". Setting it back to false stops all buys and sells while scanning continues.

The six legit checks

Every check must pass before v1 bought (v1 is now retired). One failure = skip, with the reason logged. This is the whole point of the bot — most launches fail, and that's correct. Note: the token-age gate and the minimum-tokens rule (500 at launch, loosened to 250 on 9 Nov) are v2/v3 only — v1 never carried them, which is part of why it was retired.

  1. WETH pair only — One side of the pool must be WETH. In practice only about 1 in 9 new Robinhood pools qualifies — the rest are mostly native-ETH pairs.
  2. Not our own token — Never buys $MUSE or $RMUSE, and never buys a token the wallet already holds from an earlier snipe.
  3. Standard ERC-20 — Name, symbol, decimals and total supply must all be readable onchain.
  4. No dangerous functions — Bytecode is scanned for mint, pause, blacklist, max-transaction, max-wallet, tax / fee setters, and trading switches.
  5. Honeypot simulation — Simulates buying, then selling everything. If the sell reverts or returns under 50% of the input, it's a skip. If the quoter reverts because of a hook, a transfer simulation is used instead — a revert alone is not a fail.
  6. Real liquidity, right now — The pool must have active liquidity at the current price. A pool can be created empty — FLOPPY was empty at first check, then had liquidity minutes later. The check caught both states correctly.

v1 — RETIRED / ARCHIVED (stopped 8 Nov ~09:08 GMT+7)

Built 08:31–08:55 GMT+7 on 7 Nov 2026, live from ~08:47 GMT+7. RETIRED 8 Nov ~09:08 GMT+7 on the user's order "Stop and archive v1" — watcher cron disabled, code + state + logs archived at workspace/tokens/sniper/archive/robinhood-v1-20261108/ (not deleted). Why: v1 does not carry the Part 51 gates (token-age + min-500 are v2/v3 only), and minutes after the fresh-start re-enable it filled all 5 slots with a batch deploy — 4 of 5 buys share the unidentified hook 0x2c668601… with identical ~16,251-token amounts. First live attempt failed on SLIGNAL (receipt polling threw on an unmined transaction — fixed with raw polling), then all five slots filled by ~09:33 GMT+7. Each buy was exactly 0.00001 ETH. All five were sold on 8 Nov ~02:26 GMT+7, resetting snipesDone to 0/5 — then five new buys (bottom ~02:28 GMT+7, GOOGLECAT ~02:36 GMT+7, Firstkille ~02:45 GMT+7, SKIMS ~02:54 GMT+7 and Canary ~03:02 GMT+7) took it to 5/5. Those five were then sold in the full cleanup on 8 Nov ~07:50–07:56 GMT+7 (together with FOBO), and v1 restarted at 0/5 when fresh sniping went live at ~08:56 GMT+7 — and filled 5/5 again within minutes (rows 11–15 below) before being retired. Those 5 fresh-start buys stayed open and unmanaged until the 9 Nov cleanup (~13:15 GMT+7), which sold four of them; debtdog remains held, sell-blocked.

#TokenReceivedBought (GMT+7)Token addressStatus
1Alon16,120.3408:580xc483…0087Sold 8 Nov
2GHOST16,146.4709:070x8e53…0087Sold 8 Nov
3KYLIE16,068.2909:160xcf6a…0087Sold 8 Nov
4BAG16,042.3609:240xb7b1…0087Sold 8 Nov
5MAGATARD16,198.9109:330x3bce…0087Sold 8 Nov
6bottom~16.19898 Nov ~02:280xb30b…0087Sold 8 Nov — cleanup
7GOOGLECAT15,969,809,578,688,206,915,8138 Nov ~02:360x7c5e…0087Sold 8 Nov — cleanup
8Firstkille~16.17B8 Nov ~02:450x6488…0087Sold 8 Nov — cleanup
9SKIMS~16.02B8 Nov ~02:540xd792…0087Sold 8 Nov — cleanup
10Canary~16.12B8 Nov ~03:020x2255…0087Sold 8 Nov — cleanup
11perp16,2478 Nov 08:57Fresh-start buy · 0.00001 ETH · tx 0x9d906099…Sold 9 Nov — cleanup
12debtdog9688 Nov ~08:58Fresh-start buy · 0.00001 ETH · tx 0xb1e22fdd…Held — sell-blocked by hook
13Quesadilla16,2528 Nov ~08:59Fresh-start buy · 0.00001 ETH · tx 0xf55b3c68… · hook 0x2c668601…Sold 9 Nov — cleanup
14BAMBI16,2528 Nov ~09:00Fresh-start buy · 0.00001 ETH · tx 0x3ea9e6d2… · hook 0x2c668601…Sold 9 Nov — cleanup
15eve16,2528 Nov ~09:00Fresh-start buy · 0.00001 ETH · tx 0x1f9b04c8… · hook 0x2c668601…Sold 9 Nov — cleanup

Cap decision — 15:09 GMT+7, 7 Nov. By then, 42 more vetted launches had been skipped only because the cap was full. The choice offered was raise the cap or keep it. Decision: keep cap at 5. From then until the sale, v1 was manage-only — it watched the five positions against the +100% / −50% rules and bought nothing new.

All five sold — 8 Nov 2026, ~02:26 GMT+7. Dexscreener showed 0 trades in 6 hours on all five — Alon, GHOST, KYLIE, BAG, MAGATARD. They were dead launches, not rugged: liquidity was still in the pools, there was just no buying interest left. On the user's go-ahead, 100% of each was sold (15 transactions — 3 per token via the sell-all route). Recovered 0.0000486 WETH (~$0.11); gas ~0.0000264 ETH (~$0.06). Token balances verified 0 onchain after. snipesDone reset 5 → 0 — all five slots free again, and the watcher (paused during the sells to avoid a state race) was re-enabled. Five new buys filled all five of those slots minutes later.

Full cleanup — all six positions sold, 8 Nov 2026 ~07:50–07:56 GMT+7. On the user's order ("sell it all, clean up, then fresh sniping on a fresh build"), every open position was closed. FOBO went first, sold through the deep FOBO/ETH pool (tx 0xfb3d4d90…) because the USDG pool it was bought into had 0 liquidity left; the five v1 positions above followed via sell-all (15 transactions). The WETH proceeds were converted to USDG (+$0.108, tx 0x99e6bd52…). End state on Robinhood: ETH ~0.000334, WETH dust 0.0000098, USDG $0.188281, and all six token balances verified 0 onchain. Zero open positions — a clean slate for the fresh build.

Dry-run first. In the 3,600-block dry-run (~2 hours): 18 new pools, 16 skipped as non-WETH, 2 would-buys (SI and FLOPPY). No transaction was sent.
Why TA doesn't help yet. Checked via GeckoTerminal OHLCV on 7 Nov: Alon had 2 hourly candles in its whole history, GHOST and $MUSE had 1 daily candle. RSI, MACD and moving averages all return "insufficient data" at launch age.
All 7 tokens are listed. The five snipes plus $MUSE and $RMUSE all have pages on both DexScreener and GeckoTerminal — useful for price snapshots, not for finding new pools (neither has a real new-pairs trigger).

Hard lessons from v1

  1. A quoter revert proves nothing. Our own $MUSE and $RMUSE pools revert the official v4 quoter because of their fee hook. Treating a revert as "honeypot" would reject every hooked pool, including legitimate ones. The fix is the transfer-simulation fallback.
  2. Receipt polling can throw. The first live buy (SLIGNAL, ~08:49 GMT+7) failed because the library throws when a transaction isn't mined yet, instead of returning "pending". The fix is raw receipt polling that keeps waiting.
  3. Don't trust the log filter. On the Robinhood RPC, the library's log call silently ignored the topic filter and returned hundreds of wrong events. The bot now uses raw eth_getLogs and verifies the event topic itself.

v2 — live on Arc, Base, Robinhood & Polygon

Designed and built 16:15–16:48 GMT+7 on 7 Nov 2026, dry-run first, then live from 17:44 GMT+7 on Arc, Base and Robinhood — and from ~19:35 GMT+7 on Polygon too. Chain-agnostic by config, not a separate bot per chain. It watches v4 pools (v3 pools are covered by the separate V3 build below, live since ~03:46 GMT+7 on 8 Nov), scores tokens 0–100 instead of pass/fail, and adds one new requirement v1 didn't have: a real project should have a website and at least one social link.

The agreed terms — read this first:

PhaseChainsStatusWhy / blocker
1 — LIVERobinhood 4663 · Base 8453 · Arc 5042Live since 17:44 GMT+7All three live=true, $0.02 buys, cap 5 per chain. Funding verified onchain.
1 — LIVEPolygon 137Live since ~19:35 GMT+7live=true after the Universal Router V2 switch (router 0x1095692a…9223). Buys GAS-SKIP while gas sits above the 100 gwei ceiling — it was ~271 gwei at enablement.
2Ethereum 1 · Arbitrum 42161 · BSC 56Designed, disabledStill disabled for trading, but quote lists expanded in the ~18:45 GMT+7 update (Ethereum 2→9, Arbitrum 2→6, BSC 3→7) so they're ready if ever enabled. Ethereum is watch-only by default — gas dwarfs a $0.02 buy. BSC stablecoins are 18 decimals, not 6 — a sizing trap if assumed.
—SolanaOut of scopeNo Uniswap v4 on Solana — Pump.fun / Raydium would be a different bot entirely.

PoolManager addresses differ per chain and are verified onchain before wiring — never assumed. Robinhood & Arc share 0x8366…0951; Base is 0x498581ff718922c3f8e6a244956af099b2652b2b; Ethereum is 0x000000000004444c5dc75cB358380D2e3dE08A90. On Robinhood and Arc, the Ethereum v3 factory address exists only as a decoy forwarder — another reason to verify, not copy.

Universal Router V2 — applied ~19:35 GMT+7: Polygon 137 — Router V2 0x1095692a6237d83c6a72f3f5efedb9a670c49223; Base 8453 — Router V2 0x6ff5693b99212da76ad316178a184ab56d299b43. V2 encoding: 5-field ExactInputSingleParams, no minHopPriceX36. Robinhood's fork router (0x8876…) and Arc's (0x4fcA…) keep the v1 encoding — untouched.

How the switch was proven: Before anything flipped, eth_call against both new routers on liquid pools — Polygon GRT/USDT0 and Base USDC/PEPE — reached AllowanceExpired, meaning execution got all the way to the allowance check instead of reverting on encoding. buildBuy and buildSell were then run through the real code path with the same result. Only then did Polygon go live=true and Base's router change.

What counts as a quote token — USDC pairs allowed, expanded scope live. Decision 18:21 GMT+7, 7 Nov: USDC pairs are ALLOWED. No code change was needed — the evaluator's check is "one side must be in the chain's quote allowlist", and USDC was already in it on all three live chains. It's already evaluating USDC pairs in live cycles (e.g. Reboot/USDC on Base). On Arc this matters most — nearly all Arc pools pair with USDC.

Since ~18:45 GMT+7 the scope is expanded and LIVE (expandedQuotesEnabled=true on Arc, Base and Robinhood): all verified tokens and stock tokens are quotes. Robinhood 3 → 208 — WETH · USDC · USDG + 203 tokenized stock tokens, all verified onchain; Base 2 → 13 — adds 7 B20 stocks: AAPLc · NVDAc · TSLAc · MSFTc · GOOGLc · AMZNc · METAc; Arc 2 → 2 — USDC · EURC, unchanged: no stock-token system exists on Arc; Ethereum 2 → 9, Arbitrum 2 → 6, BSC 3 → 7, Polygon 2 → 5 (Phase 2 — live=false).

In — live now: WETH / WPOL / WBNB — wrapped native per chain; USDC (Circle native) — Base, Arc, Polygon; bridged USDC on Robinhood — pairs explicitly allowed; USDT, DAI, WBTC / cbBTC / BTCB — re-verified onchain per chain before wiring; USDG (Paxos) — Robinhood's dominant stablecoin; EURC on Arc — secondary; Stock tokens — Robinhood's 203 tokenized stocks (ERC-8056) and Base's 7 major B20 stocks.

Out / still gated: ~292 other B20 stocks on Base — only the 7 majors are wired; the rest need per-token verification first; Backed / Dinari stock tokens — not wired; Lookalikes — Base USDbC and Polygon USDC.e are different contracts; a Robinhood impostor trades on the name "Universal Stable Digital Coin"; Any address not checked onchain for code, symbol and decimals first.

Expansion LIVE — stocks + ALL verified tokens (Go 18:25 GMT+7, enabled ~18:45 GMT+7). Decisions 18:23–18:25 GMT+7, 7 Nov: add stock-token pairs, on all EVM chains (not only Robinhood), paired with all verified tokens and stocks — WETH/WPOL/WBNB, USDC, USDT, DAI, WBTC/cbBTC/BTCB, USDG, EURC and any other verifiable token, plus verified stock tokens. Live Go: "Go for all" ~18:25 GMT+7. Status: LIVE since ~18:45 GMT+7 — expandedQuotesEnabled=true on Arc, Base and Robinhood, after build + dry-run verification passed. At that point Polygon stayed execution-blocked and Phase 2 chains (Ethereum, Arbitrum, BSC) stayed live=false — "all" meant all verified tokens + stocks as quotes, not all chains live. Polygon has since gone live too (~19:35 GMT+7) via the Router V2 switch; Phase 2 chains remain off.

ERC-8056 pricing verified onchain. For stock tokens the economic balance is balanceOf × uiMultiplier / 1e18 — not the raw balanceOf(). Worked example: a $0.02 NVDA buy prices exactly $0.02000 with this math. All 203 Robinhood stock tokens respond to the ERC-8056 interface. Stock quotes also carry weekend-stale price risk — the underlying market is closed while the chain isn't.

Stock-paired launches are real on Robinhood. In the last 50,000 blocks, 65 of 208 new pools (31%) were stock-paired — all of them memecoin / stock pairs. This isn't a hypothetical market; it's a third of Robinhood's launch flow.

Two research-note addresses corrected (fresh onchain checks). The corrected, verified ones: Polygon DAI 0x8f3Cf7ad23Cd3CaDbD9735AFf958023239c6A063; Arbitrum WBTC 0x2f2a2543b76a4166549f7aab2e75bef0aefc5b0f.

Positions gap — FIXED ~18:48 GMT+7, 7 Nov. run.js now records every live buy into STATE.json positions — buy price read from the pool right after the buy, with the Initialize-event price as fallback — so the sell rules (+100% sell half, −50% sell all) can actually trigger on a v2 buy. Dry-run proof (Test 9): +110% → would SELL 50%, −52.4% → would SELL ALL, 0% control → no sell. Live STATE.json untouched, zero transactions sent, and no v2 buys yet — so nothing was ever at risk while the gap was open.

The social rule — and its timing problem. Minimum to score full marks: a website listed on DexScreener that actually loads, plus at least one social (X or Telegram). DexScreener is the only verified source for this — GeckoTerminal has no social or website fields at all. But here's the catch, proven on our own buys: Alon, hours after purchase, had no DexScreener profile at all. All five v1 buys would have failed a hard social gate at snipe time — profiles are usually submitted after launch, or never. So v2 doesn't hard-gate. It scores, re-checks at +10 minutes, and treats "still no profile at +60 minutes" as information, not an automatic sell. Links are also self-declared and spoofable — a scam can link a real project's site — which is why socials are worth at most 25 of 100 points and sit on top of the onchain checks, never instead of them.

Launchpad + known-rug registries — LIVE (Part 50, 8 Nov ~07:35–08:02 GMT+7). The evaluator now carries two curated registries (registries-test 98/98). The launchpad registry holds 15 launchpads with per-chain addresses — Robinhood: Pons v1 factory + locker, Pons v2 factory + V2MemeHook, Flap Portal, o1 factories/hooks, Vectr factory + migrator + locker, Hood, HedgeFun, Frontier, Duel; Arc: Argus portals ×5 + o1; Base: Flap Portal + o1. Launchpad provenance is now a real +5: it scores when the pool-creation transaction or hook matches a registered launchpad, or the LP-NFT owner matches a registered locker — unknown provenance scores 0, never negative. The known-rug registry is a live veto: the scammer deployer 0x333089e0… and the fake-POL token 0xe174c72d… block a buy outright, whatever the score says.

Two new hard gates — LIVE (Part 51, 8 Nov ~08:02–08:46 GMT+7; min loosened 9 Nov). Token-age gate (hard veto): the token's first pool — the earliest across Dexscreener and GeckoTerminal — must be under 24 hours old. GeckoTerminal matters because it is the only indexer source for Arc; Dexscreener has nothing there. If both sources are unknown the gate fails open and logs it. Minimum-tokens rule (hard gate): the $0.02 buy must yield at least 250 whole tokens — a price ceiling of about $0.00008/token — and an unresolvable expected output blocks the buy too. Update 9 Nov: after the gate review (v1-five verdicts 5/5 survived), the minimum loosened to 250 whole tokens in both evaluators; the age gate is unchanged. Both gates exist because of FOBO: a ~13-day-old token whose brand-new USDG pool looked like a fresh launch and yielded ~2.43 tokens. The same build fixed priceFromSqrt (FOBO's position price now reads 0.007849 USDG/token instead of 0, so sell triggers can fire) and the holders-test timeout (38/38). Replays: DELTA V3 75 → SKIP, GC V3 35 → SKIP, SI 10 → SKIP; CLAUDIA 30, FNUT 15, $SLNADO 15, APU 30 unchanged.

Scoring adjustments — live on top of the base rubric. Memory-adjustment ±25: the deployer-memory loop remembers how a deployer's past launches ended and shifts the score accordingly; an hourly follow-up cron tracks outcomes, with the first 24h verdicts due ~09:00 GMT+7 on 9 Nov. GoPlus second opinion [−10, +5]. Holder-concentration [−10, +5] with the bundle check. Smart-money boost (+10/+15): live in both evaluators since 9 Nov ~07:11 GMT+7 when a watchlisted wallet has bought the token — a score boost only, never a buy trigger. Verified ~14:17 GMT+7 on 9 Nov it has fired zero times (0 nonzero scores across 419 v2 + 32 v3 evaluations). Full scoring stack: social-website 25, liquidity-depth 15, early-traction 15, verified-contract 10, lp-locked 10, deployer-history 10, no-impersonation 5, launchpad-provenance 5 — then memory ±25, GoPlus [−10,+5], holder-concentration [−10,+5], smart-money +10/+15. Buy line 45, $0.02 per buy, caps 5; sell 50% at +100%, all at −50%.

V3 — LIVE on Robinhood

Went live 8 Nov 2026 ~03:46 GMT+7. V3 watches Uniswap v3 pools on Robinhood Chain — the pool type that Flap, Vectr and Robinlaunch graduate into, and where Pons v1 launches land. It has its own watcher, its own state, and its own cap of 5 positions, separate from v2's per-chain cap. It was paused during the Part 51 build and re-enabled ~08:56 GMT+7 under the new rules — 0/5, zero positions. Under those same rules its best historical candidate, DELTA (75/100), now replays as a SKIP — its first pool is 2,378.6 hours old, and the token-age gate vetoes it.

The road here — four parts:

V3SwapHelper — the execution unlock: Contract — Robinhood Chain 0x3ecc88CBaD515D4F34dD365A109Cb7bB71a2FfcF. Sourcify exact match (creation + runtime); owner is the agent wallet. Minimal and owner-only: swapExactInputSingle calls pool.swap() directly and pays in uniswapV3SwapCallback, guarded to the active pool. Transfer-first funding — no approvals; output goes straight back to the owner in the same transaction. No delegatecall, no selfdestruct, no upgradeability, no arbitrary calls. Micro-test on DELTA: buy returned 99.0% of spot (the 1% pool fee); sell returned 98.01%. Total cost of deploy + tests: ~0.0000217 ETH.

How V3 buys work: $0.02 per buy, hard cap 0.00001 ETH per buy — same size as v2. Buy line 45, same scoring and sell rules as v2: sell 50% at +100%, sell all at −50%. The Part 51 hard gates apply to V3 too — token-age gate (first pool <24h) and the minimum-tokens rule (500 at launch, loosened to 250 on 9 Nov in both evaluators); DELTA's replay under them is a SKIP on age. Each leg is 2 transactions through the helper (transfer in, then swap); a reverted swap is recovered via the helper's owner-only rescue. LP-locked check is built for V3 — it reads the position-NFT mint in the pool-creation receipt. DELTA now scores 70/100 (was 60 before the check existed).

What it unlocks — buyable at birth now: Flap graduations — the #2 memecoin launchpad by fees, live on Robinhood; its curve trades are off-DEX, but graduations land in V3. Vectr graduations — Robinhood curve launchpad, graduates to V3, LP locked. Robinlaunch graduations — V3 at 2.5 ETH raised. Pons v1 launches — dormant lately, but now buyable the moment one appears.

Open V3 risks, stated plainly: Flap tax tokens (up to ~11% per trade) may not trip the bytecode tax check — the honeypot simulation shows reduced proceeds, not zero, so a taxed token can pass. The LP-lock check is a T+0 heuristic: a position NFT held by a deployer multisig would also score as locked. And V3 launch volume on Robinhood is low (~5 new pools a day in dry-run windows) — expect a slow drip of candidates, not a flood.

Limit orders — general 1inch tool (10 Nov)

Built from the old 1inch limit-order-protocol 4.2.2 source — the older app.1inch.io advanced limit order version, the favorite venue for doing limit orders. One command places a limit order for any ERC-20 token on any supported chain: node place-order.js --chain 8453 --maker WETH --taker <address> --amount full --price 1e-7 --years 3. The tool verifies the router onchain, checks balance and allowance, signs via the Privy agent wallet (EIP-712), triple-checks the order hash (viem == official 1inch SDK == onchain LOP.hashOrder), tries the 1inch orderbook API, and publishes onchain where a registrator exists.

ChainRouterRegistratorPublish
BaseAggregation Router v60x2339…1CCEonchain ✓
PolygonAggregation Router v60x2339…1CCEonchain ✓
Robinhood0x5a70…389C7noneAPI-only
Arc0xe08c…6bdanoneAPI-only

Live orders placed with it: Base NAMI — full WETH balance → 1,426.69 NAMI @ 1e-7, expiry 2027-02-08 and a 3-year version to 2029-11-10, both published onchain via OrderRegistrator (~$0.0000005 gas each). Robinhood NAMI — 0.00015 WETH → 1,500 NAMI @ 1e-7, 3y — signed and verified, not published (no Robinhood registrator) and not yet funded (wrap at publish time). Dry runs proven on Robinhood (NAMI) and Polygon (MUSN). The 1inch orderbook API returns 401 without a key — a KYC'd Business account is needed for the official book; onchain publication via OrderRegistered events is the permissionless path. Code: github.com/laserinu/muse-aigent/tree/main/sniper/limit-orders.

Try the v2 score

The exact rubric from the v2 design. Hard gates (honeypot, dangerous functions, no liquidity, wrong quote token, our own token — and now the token-age gate, the minimum-tokens rule (250 since 9 Nov) and the known-rug veto) still override any score — fail one and there's no buy at any number.

A deployer in the known-rug registry is a hard veto now (Part 50), whatever the total says — and four live adjustments sit on top of this base score: memory-adjustment ±25, GoPlus second opinion [−10, +5], holder-concentration [−10, +5], and the smart-money boost +10/+15 (live 9 Nov, fired 0 times so far).

TOTAL SCORE: 0 — Watch only. Under 45 — log it, re-score at +10 and +60 minutes. Buy only if it crosses 45 and a slot is free.

Honest limit: No score at launch predicts upside. At T+0 there are no holders, no volume and no price history. This rubric ranks "looks like a real project trying" against "anonymous throwaway" — it filters junk. The +100% / −50% sell discipline is what does the risk work, not the score.

v2 dry-run result: 7,200 blocks scanned in the dry-run; 24 new pools evaluated; 0 would-buys — and 0 transactions sent. Watch-only — where all 5 v1 buys would have landed at T+0, for having no profile yet.

Funding — verified for live v2

Funding was verified onchain at go-live (~17:44 GMT+7, 7 Nov) and again after the evening top-ups (~19:35 GMT+7): Arc USDC 0.818111, Base ETH 0.0004166, Polygon POL 5.514 + WPOL 2.2667, Robinhood ETH 0.0002024 — portfolio total ~$4.89. After the 8 Nov cleanup (~07:56 GMT+7) the Robinhood picture changed: proceeds were converted to USDG, leaving ETH ~0.000334, WETH dust 0.0000098 and USDG $0.188281, with zero open positions — enough to fund the fresh run at $0.02 per buy. After the 9 Nov cleanup the four sold v1 tokens added +83,145 USDG raw, taking Robinhood USDG to 271,426 raw ($0.271), with debtdog the only token still held. Funding always comes from the user's own wallet — Muse never moves user funds unprompted.

ChainAt proposed $1 × 5 buysWhat was heldAt the agreed small-buy size
Robinhood~$5.05 needed~$0.05 ETHFirst to go live — short ~$0.14
Base~$4.70 needed~$0.80 ETH~$0.15–$0.25 / chain to start
Arc~$5.00 needed$0.098 USDCUSDC is both gas and quote — simplest to fund
Polygon~$5.45 needed~$0.05 POL at design time; 5.514 POL + 2.2667 WPOL after the evening top-upLive since ~19:35 GMT+7 via Router V2; buys GAS-SKIP above 100 gwei

Phase 1 total at the proposed $1 size: ~$20.20 needed against ~$1.15 held. That $20 figure is the design's math, not the plan — the plan is small buys, one chain at a time.

Still gated / unverified in v2: B20 approve-via-Permit2 is unproven — the first live B20 (Base stock) buy may fail at the approve step, gas only. Only the 7 major B20 stocks are wired on Base; ~292 others need per-token verification before they can be quotes. Backed / Dinari stock tokens — not wired. PositionManager addresses on Base and Polygon. Half-sizing (score 45–69) — not yet wired into the buy builder; half-score buys currently spend the full $0.02. Live price feed for sizing — needed before any size above v1's fixed tiny amount. Polygon gas — live since ~19:35 GMT+7, but buys GAS-SKIP while gas is above the 100 gwei ceiling (~271 gwei at enablement). Positions recording — FIXED ~18:48 GMT+7. v2-router swap encoding on Base and Polygon — RESOLVED ~19:35 GMT+7.

Safety rules that never change

Timeline — 7–9 Nov 2026 (GMT+7)

  1. 08:31–08:55 — v1 built. Watcher, evaluator, executor and position manager on Robinhood Chain. Dry-run: 18 pools, 2 would-buys.
  2. ~08:47 — Watcher goes live. Scheduled every 2 minutes. Buying still gated off until the explicit "Go".
  3. ~08:49 — First live attempt fails — SLIGNAL. Receipt polling threw on an unmined transaction. Fixed with raw polling.
  4. 08:58–09:33 — All 5 slots fill. Alon, GHOST, KYLIE, BAG, MAGATARD — 0.00001 ETH each.
  5. ~15:09 — Cap kept at 5. 42 vetted launches skipped on the cap by then. v1 becomes manage-only.
  6. ~15:30 — Charting check. DexScreener + GeckoTerminal deep dive: technical analysis is impossible on tokens this young — too few candles.
  7. 16:15–16:48 — v2 designed, built, dry-run. Multichain, scored, social-aware. 7,200 blocks, 24 pools, 0 would-buys, 0 transactions. Terms settled: one chain at a time, small buys only.
  8. ~17:39–17:42 — v2 Go — Arc, Base, then Robinhood. "From now on, sniping used v2 Build, Go" for Arc and Base, then "Also on RobinHood now. Go." Websites paused — sniping is the priority.
  9. 17:44–17:46 — v2 LIVE — verified in config + STATE.json. Master live=true; 5042 / 8453 / 4663 live=true; Polygon live=false (execution blocked). Funding verified onchain: Base funds 5 buys, Arc 3 (0.098111 USDC), Robinhood only 1 (0.0000223824 ETH). Enforced via snipesDone pre-seed {"5042": 2, "4663": 4}. Watcher cron sniper-v2-watcher-arc-base, every 2 min, staggered 1 min from v1.
  10. 17:45:48 — First live cycle — 0 buys. Nothing scored ≥50. Base CHOOP 10/100 watch, Robinhood Organic 10/100 watch. Latest cycles still 0 buys — Base NAFF and AOP watched at 15/100.
  11. ~18:21 — USDC pairs confirmed allowed. User decision. Verified in code: USDC was already in the quote allowlist on all three live chains (Robinhood WETH/USDC/USD G, Arc USDC/EURC, Base WETH/USDC). No change needed.
  12. ~18:23–18:25 — Expansion ordered — stocks + all verified tokens. "Yes, add stock-token pairs" → "Not only on RobinHood. Do it on all EVM chains." → "All verified tokens and stocks. Paired with." → Live Go "Go for all" ~18:25. ERC-8056 pricing for stock quotes. Polygon stays blocked, Phase 2 stays disabled.
  13. ~18:27 — This page updated — standing rule. Keep this website updated with the sniping work from now on, on every update.
  14. ~18:45 — Expanded pair scope LIVE. expandedQuotesEnabled=true on Arc, Base, Robinhood. Robinhood quotes 3→208 (203 stock tokens, all verified onchain), Base 2→13 (7 B20 stocks), Ethereum 2→9, BSC 3→7, Polygon 2→5, Arbitrum 2→6, Arc 2→2 (no stock-token system on Arc). ERC-8056 pricing verified: economic balance = balanceOf × uiMultiplier / 1e18 — a $0.02 NVDA buy prices exactly $0.02000. Stock-paired launches are real: 65 of 208 new Robinhood pools in the last 50,000 blocks (31%), all memecoin/stock. Two research-note addresses corrected by fresh onchain checks: Polygon DAI and Arbitrum WBTC. First watcher cycle after enabling: 11 pools, 2 WATCH (HONK, HIRED, both 10/100), 0 buys.
  15. ~18:48 — Positions recording FIXED. run.js now records every live buy into STATE.json positions — buy price read from the pool right after the buy, Initialize-event price as fallback. Dry-run proof (Test 9): +110% → would SELL 50%, −52.4% → would SELL ALL, 0% control → no sell.
  16. 19:01 / 19:25 — Polygon Go + "fix it all the problems". User: "Do snipping on Polygon too" → "Go" (19:01 GMT+7), then "Fix it all the problems" (19:25 GMT+7). The work that followed: the Router V2 switch, the three bug fixes, and Polygon going live.
  17. ~19:35 — Router V2 switch APPLIED — Polygon LIVE — first four-chain cycle: 0 buys. Universal Router V2 in on Polygon (0x1095692a6237d83c6a72f3f5efedb9a670c49223) and Base (0x6ff5693b99212da76ad316178a184ab56d299b43) — 5-field ExactInputSingleParams, no minHopPriceX36. Proven by eth_call reaching AllowanceExpired on liquid pools (Polygon GRT/USDT0, Base USDC/PEPE) and by buildBuy/buildSell through the real code path; Base's first buy is unblocked. Robinhood (fork router 0x8876…) and Arc (0x4fcA…) untouched, v1 encoding kept. Three bugs found and fixed in the same pass, all in v2, none ever exercised live (0 v2 buys before this). Log-scan chunking also fixed: chains.js rawGetLogs now chunks eth_getLogs into ≤5,000-block ranges (Polygon publicnode rejects ranges over 10,000). Polygon is live=true — four live chains: Arc 5042, Base 8453, Robinhood 4663, Polygon 137; watcher runs every 2 min with --chains 5042,8453,4663,137. First live cycle 19:35:13 GMT+7, all 4 chains: 0 buys — Polygon 0 new pools; Robinhood PIG watch 10/100; Arc skip not-a-quote-pair; Base skip no-active-liquidity. Polygon gas was ~271 gwei at enablement, above the 100 gwei ceiling, so Polygon buys GAS-SKIP until gas drops. Funding after the evening top-ups: Arc USDC 0.818111, Base ETH 0.0004166, Polygon POL 5.514 + WPOL 2.2667, Robinhood ETH 0.0002024; portfolio total ~$4.89.
  18. 8 Nov ~01:30 — Buy line lowered — 50 → 45. After ~7.5 hours live with 0 buys — 242 cycles, ~2,000 pools evaluated, best score ever 45/100 — the buy line dropped from 50 to 45. The reason: most new launches use a hook, and a hooked pool only ever earns 5 of the 15 liquidity-depth points (partial credit), while LP-locked and early-traction are structurally 0 at birth — so a hooked pool caps at 45 at T+0, and a line of 50 was near-unreachable for the dominant launch type. Bands are now: ≥70 full size, 45–69 half size, <45 watch only. The change applies to future launches only — past near-misses are not bought retroactively.
  19. 8 Nov ~02:26 — All five v1 positions SOLD — slots free again. Alon, GHOST, KYLIE, BAG, MAGATARD — sold 100% of each on the user's go-ahead. Dexscreener showed 0 trades in 6 hours on all five — dead launches, not rugged: liquidity (~$1,367 per pool) was still in the pools, volume was $0. Sold via sell-all (3 txs per token, 15 total) while the watcher was paused to avoid a state race. Recovered 0.000048594 WETH (~$0.11); gas ~0.0000264 ETH (~$0.06). Token balances verified 0 onchain after. snipesDone reset 5 → 0 — all five slots free — and the watcher was re-enabled, ready for new launches. v2 unchanged: live on Arc, Base, Robinhood and Polygon, 0 buys so far, buy line 45.
  20. 8 Nov ~02:28–03:02 — Five new v1 buys — bottom, GOOGLECAT, Firstkille, SKIMS, Canary. 0.00001 ETH each, taking v1 to 5/5. All five sold again in the cleanup — see below.
  21. 8 Nov ~03:26–03:40 — V3SwapHelper deployed + micro-test PASSED (Part 42). V3SwapHelper 0x3ecc88CBaD515D4F34dD365A109Cb7bB71a2FfcF deployed on Robinhood — Sourcify exact match, owner = agent wallet, deploy cost 0.00001402 ETH. Micro-test on DELTA passed first try, both legs: buy got 4.078 DELTA (99.0% of spot, the 1% pool fee); sell got 98.01% of the WETH back. Total cost deploy + tests ~0.0000217 ETH; helper holds nothing. The V3 execution gap was closed at the execution layer.
  22. 8 Nov ~03:38 — V3 cap decided — separate cap of 5. Parent decision ("decide for me"): go live with a SEPARATE V3 cap of 5 positions on Robinhood, independent of v2's per-chain cap.
  23. 8 Nov ~03:46 — V3 sniping LIVE on Robinhood (Part 43) — 0/5, first cycle clean. v3-helper-trade.js built (buildV3HelperBuy / buildV3HelperSell through the helper, transfer-first, 2 txs per leg, hard cap 0.00001 ETH per buy) and v3-live-run.js wired with its own watermark and state (v3-STATE.json): buy line 45, $0.02 per buy, sell 50% at +100% / all at −50% — same rules as v2. LP-locked check built into V3 scoring — DELTA now scores 70/100 (was 60). Dry-run proof first: 3 cycles, 0 transactions; a 50k-block window saw 9 pools, 5 skipped, 4 watched, 0 would-buys. Then live=true flipped: v3SnipesDone 0/5, 0 positions, first live cycle clean (0 new pools, 0 buys). Cron sniper-v3-watcher-robinhood runs every 2 min.
  24. 8 Nov — before the cleanup — The FOBO buy — the expensive lesson. The sniper bought FOBO for $0.02 into a USDG pool that was 2 minutes old — but FOBO itself was ~13 days old (first pool 26 Oct 2026). The buy yielded only ~2.43 tokens, and the USDG pool's liquidity was pulled after the buy, leaving 0. A new pool is not a new token — the two Part 51 gates were designed directly from this.
  25. 8 Nov ~07:35–08:02 — Launchpad + known-rug registries LIVE (Part 50). 15 launchpads curated with per-chain addresses. Launchpad provenance became a real +5. Known-rug VETO went live: scammer deployer 0x333089e0… + fake-POL token 0xe174c72d…. registries-test 98/98. Replays: DELTA V3 70 → 75 with the +5; CLAUDIA 30, FNUT 15, SI 10, $SLNADO 15 unchanged.
  26. 8 Nov ~07:50–07:56 — Full cleanup — all six positions SOLD. User order: sell it all, clean up, then fresh sniping on a fresh build. FOBO sold via the deep FOBO/ETH pool (tx 0xfb3d4d90…) — its USDG pool had 0 liquidity left; v1's five (bottom, GOOGLECAT, Firstkille, SKIMS, Canary) sold via sell-all, 15 txs. WETH proceeds converted to USDG (+$0.108, tx 0x99e6bd52…). Robinhood after: ETH ~0.000334, WETH dust 0.0000098, USDG $0.188281; all six token balances verified 0 onchain.
  27. 8 Nov ~08:02–08:46 — New gates built (Part 51) — token-age, min-500, price fix, holders fix. Token-age gate: hard veto unless the token's FIRST pool is <24h old — earliest pool across Dexscreener + GeckoTerminal (GeckoTerminal is the only indexer covering Arc; fail-open if both unknown). Minimum-500 rule: hard gate, the $0.02 buy must yield ≥500 whole tokens (~$0.00004/token ceiling) — loosened to 250 on 9 Nov after the gate review; unresolvable output also blocks. priceFromSqrt fixed — FOBO's position price now reads 0.007849 USDG/token instead of 0. Holders timeout fixed. Tests: gates-test 27/27, holders-test 38/38, registries-test 98/98. Replays: DELTA V3 75 → SKIP (age 2,378.6h), GC V3 35 → SKIP, SI 10 → SKIP; CLAUDIA 30, FNUT 15, $SLNADO 15, APU 30 unchanged. FOBO is rejected by BOTH new gates.
  28. 8 Nov ~08:56 — FRESH SNIPING LIVE under the new rules. v2 snipesDone reset (Robinhood/Arc) to 0/0, v1 and v3 at 0, zero open positions at that moment — and all three watcher crons re-enabled (v1, v2, v3). Scoring now: base rubric + memory-adjustment ±25 (deployer-memory loop live, hourly follow-up; first 24h verdicts ~09:00 GMT+7 9 Nov), GoPlus second opinion [−10,+5], holder-concentration [−10,+5] with bundle check. Buy line 45, $0.02/buy, caps 5, sell 50% at +100% / all at −50%. Superseded minutes later for v1.
  29. 8 Nov 08:57–09:00 — v1 fills all 5 slots — batch deploy. Within minutes of the fresh-start re-enable, v1 bought 5 tokens at 0.00001 ETH each: perp (16,247, tx 0x9d906099…), debtdog (968, tx 0xb1e22fdd…), Quesadilla (16,252, tx 0xf55b3c68…), BAMBI (16,252, tx 0x3ea9e6d2…), eve (16,252, tx 0x1f9b04c8…). 4 of 5 share the unidentified hook 0x2c668601… with identical ~16,251-token amounts — looks like one batch deploy, not five independent launches. v1 has no Part 51 gates, so the token-age and min-500 checks never ran on these.
  30. 8 Nov ~09:08 — v1 RETIRED — stopped and archived. User order: "Stop and archive v1." v1 watcher cron disabled; v1 code + state + logs archived at workspace/tokens/sniper/archive/robinhood-v1-20261108/ (not deleted). The 5 fresh-start buys remained OPEN and UNMANAGED until the 9 Nov cleanup. v2 and v3 remain live with the Part 51 gates.
  31. 9 Nov ~07:11 — Smart-money signal LIVE in both evaluators. A +10/+15 score boost when a watchlisted smart-money wallet has bought the token — a boost only, never a buy trigger. Verified ~14:17 GMT+7 the same day: it has fired ZERO times — 0 nonzero smart-money scores across 419 v2 + 32 v3 evaluations.
  32. 9 Nov ~09:00 — V1-five 24h verdicts land — 5/5 survived. debtdog +50%; the other four later sold at ~−2% token value. That closed the pre-agreed gate review: min-500 loosened to 250 in BOTH evaluators (MIN_BUY_TOKENS_WHOLE = 250, price ceiling ~$0.00008/token). Age gate (<24h) and score line (45) unchanged.
  33. 9 Nov ~13:15 — V1-five cleanup — 4 sold, debtdog sell-blocked. User order: "Sell it all the tokens from snipping, clean it up." perp, Quesadilla, BAMBI and eve sold — balances verified 0 onchain. Proceeds 39,206,842,446,335 wei WETH converted to +83,145 USDG raw ($0.083); USDG now 271,426 raw; gas ~0.0000289 ETH (~$0.065); net ~+$0.018 after all fees. debtdog UNSOLD: its pool hook 0xDE167EC2… blocks our Universal Router sell at any minOut (even 1 wei) and half size, both allowances confirmed set; the pool is live (287 sells/24h by others) and GoPlus calls the token clean — the hook blocks our route specifically. Only one debtdog pair exists (WETH pool), no alternative route. 968 debtdog (~$0.034) stays held — the second sell-blocked token after ANT.
  34. 9 Nov ~14:03–14:09 — Strategy set — sniper only · fees counted · safety first · grow, never shrink. User decisions: sniper-only focus until they say otherwise (daily airdrop DYOR check paused); every figure calculated net of fees, hook-fee income as its own line; safety first on every smart contract; and always find a way to grow the portfolio from anything, anywhere, anytime — without reducing it.
  35. 9 Nov ~14:21 — Old builds archived — copy-trader in build. 13 old files moved to archive/v2-old-builds-20261109/; live code verified loading. v1 stays at archive/robinhood-v1-20261108/. The copy-trader was being built: mirrors the 8 watchlisted wallets' buys on Robinhood, $0.02 per copy-buy, separate cap of 5, reusing the proven buy paths, gated by GoPlus honeypot, holder concentration, known-rug veto and transfer simulation — dry-run first, live only after review.
  36. 9 Nov ~15:32 — Copy-trader LIVE. User said "Go live" after reviewing the dry-run verdict. State live=true, watcher cron every 2 min. 0 positions at launch; 8 candidates recorded from dry-run.
  37. 9 Nov ~16:03 — First live native copy-buy — FAWBO. Minutes after the native-ETH go-live, the copy-trader bought FAWBO (0xf2cdba01e2a30adee9af48deadd1e362ae728693) for $0.02 native ETH — tx 0xdcbde7786023293c1f4d8f4d729277db31cbaf6fde7cffb1e65ab9b42e492799, block 109937492, confirmed. Agent holds ~771 FAWBO. Post-buy sell-route sim failed, flagged sellBlocked — but the onchain FAWBO→Permit2 allowance is 0 (the native buy sets none), so this is likely the known sim false-positive, not a proven hook-block. The first actual sell is the real test.

FAQ / closing

So is the sniper making money? Nothing on this page claims that — and from 9 Nov every figure is reported net of all fees (gas + LP + hook), with hook-fee income as its own line. Every position from the first runs has been closed except debtdog: the original five v1 tokens were sold 8 Nov ~02:26 GMT+7 (0.0000486 WETH recovered), the second five v1 tokens and FOBO were sold in the full cleanup on 8 Nov ~07:50–07:56 GMT+7, and the third batch — v1's fresh-start five — was cleaned up on 9 Nov ~13:15 GMT+7: four sold (+83,145 USDG raw; USDG now 271,426 raw; cleanup gas ~0.0000289 ETH, net ~+$0.018 after all fees), while debtdog (968 tokens, ~$0.034) could not be sold because its pool hook blocks our route. v2 and v3 remain live with the gates and have not bought yet. The copy-trader has made its first live buy — FAWBO, ~771 tokens for $0.02 native ETH at ~16:03 GMT+7; its first sell is still untested. The amounts are deliberately tiny — the point so far was proving the full watch → check → buy → manage loop works safely on mainnet, and learning the FOBO lesson cheaply. The honest verdict (9 Nov): the sniper is a lottery-ticket machine, not a compounder — it needs one big hit, and that's a matter of time and volume, not effort.

Can I copy this build? The pattern is all here: raw eth_getLogs on the v4 PoolManager, read-only legit checks, a quoter-revert fallback, exact tiny sizes, a hard cap, a file-based live gate, and dry-run before any "Go". What you can't copy from a page is the signing setup — that needs your own Privy agent wallet, funded by you, with your own explicit go-live decisions per chain.

Why not just snipe everything, everywhere? Because most launches fail the basic checks, most pools aren't even paired with a token you'd want to hold, and every chain needs its own funded balances, gas and approvals. Breadth without funding and verification is how bots buy honeypots at scale. One chain, small buys, proven first.