---
name: screen-log-8
description: Continuation of [[screen-log-7]] from 16 Sep 2026, after that file hit the 48KB cap. Same rules and columns as [[screen-log]] - one row per screen, written BEFORE the verdict, screen-time facts only, outcomes scored in the screen-log-outcomes files. Read and append on every screen.
sources: [cowork]
aliases: [screen-ledger-8]
---

# Screen log, part 8 (from 16 Sep 2026)

Rules and columns are in [[screen-log]]. Prior rows: [[screen-log-7]].

## Log

### ⚠ 2026-09-16 ~04:20 BKK. VERIFICATION PASS ON THE EARN VALUATION in [[screen-log-7]] ("double-check this again"). **TWO ERRORS. VERDICT UNCHANGED.** Artifact v2, https://claude.ai/artifact/Eqf2JgVaWNTGz1YRBSYdcA. Ben has now been right to push on every single delivery in this thread.

**⚠ ERROR 1: "the biggest token in this business anywhere on earth is Beefy at $4.5M" WAS TOO NARROW A CATEGORY.** CoinGecko, checked because DefiLlama's mcap field is documented unreliable: **Yearn (YFI) $75,060,925**, Badger $7,135,477, Beefy (BIFI) $4,589,478, Harvest $4,065,725, ICHI $834,598, **Gamma $16,216**. Widen "liquidity automation" to "yield protocols" and the global ceiling is Yearn, not Beefy. **A billion is 13x Yearn, not 221x Beefy.** Conclusion survives and is better supported. **STANDING: state the category WIDTH you used when quoting a ceiling, and give the number for the wider reading too. A ceiling is only as honest as its category boundary.**

**⚠⚠ ERROR 2, THE IMPORTANT ONE: I MEASURED ONLY THE EARN SIDE OF A TWO-SIDED FEE STREAM.** Opening four receipts of the launch-contract → team-wallet legs (the "is it really fees" check) showed **SPY moving to the team in the SAME transactions**. Full measurement: **the fee wallet `0x3e6aee86bd7aadafd134f470ea1f86a78449ebed` has received 38.4533 SPY = $29,110 across 30 legs, and its ONLY inbound source is the launch LP contract `0x4e3468951d49f2eea976ed0d6e75ffcb44a9a544`.** So EARN's revenue is a normal Uniswap v4 two-sided LP fee collection, not an EARN-only drip. **This makes the buyback claim measurably false in hard currency: they took $29,110 of SPY and spent $4,584 of it buying EARN back = 15.7%, not "all of the fees and revenue".** The other $24,526 of SPY went out through routers (17.5908 SPY to the 24,232-byte router `0xc94135b63772b91d79d0a2daab2a8801f32359bd`, 10.7389 to `0x306978ee7efd4ac851f0387be7f570eda44ed631`, 5.4096 to `0x39b38686a19836ac10162c490e4558e120cbbe5f`) and the team wallets hold almost no SPY, USDG or WETH today. **Final destination UNRESOLVED and it is the single best next measurement.** **STANDING RULE: a v4 LP fee collection pays BOTH tokens. When you find a token stream from a pool or position manager, always measure the quote-asset stream in the same transactions. I opened the receipt to prove it was fees and then failed to read the other half of the receipt.**

**✅ EVERYTHING ELSE SURVIVED, each re-checked against a second source.** Gamma's $16,216 cap is real (CoinGecko, independent of DefiLlama). Beefy $4,589,478 (CG) vs $4,515,721 (DefiLlama), agree. **PONS $459M is real: price $0.6668 on the deepest WETH pool, on-chain totalSupply 1,000,000,000 with ~688,282,419 circulating, and ten days of daily candles ranging $0.5435 to $0.8410 — it has been a $380M-$580M token for weeks.** (Note for [[kapothegoat-wallet]]: GMGN's "Bought / MC" column is a LIFETIME AVERAGE across all a wallet's buys, not a recent price — KAPO's "PONS @ $27.67M" is his July-onward average, not a contradiction of today's $459M.) The 3,603,817,955 EARN stream really is LP fees. Fee run rate ~$31,800/month, now confirmed independently from the SPY side (38.4533 SPY over 55 days, doubled for the two-sided collection) against the $33,000 I had from the EARN side.

**⚠ TWO CLAIMS SOFTENED.** (a) **"DELTA has $12,578 of deposits" is DefiLlama's adapter, and that adapter understated DELTA by 92x once before** ($16,430 vs its own endpoint's $1,509,084). Report it as a floor, not a fact; the Frame 3 point (a $11,959,183 cap that is not backed) stands either way. (b) **Steer and Arrakis "have no token" is really "no token that CoinGecko or DefiLlama tracks"** — the CG ids returned nothing, which is consistent but not proof.

**⚠ CORRECTION THAT BELONGS TO THE EARLIER "WHO IS SELLING EARN" ROW: the largest non-protocol holder is inside the team closure.** `0x1a2acf82601cb26db5160e707a78a5dd9dbaa8e8` (3,150,704,109 = 3.15%, 7702 wallet, nonce 1) **received 322,222,222 EARN directly from team wallets** and separately bought ~1,994,082,224 on the market this week. It is **65% of the 17-wallet cluster's entire 4.87% holding**. Whether it is the team, an advisor, a market maker or just a recipient is unresolved. I had presented it as an independent buyer.

**META, fourth pass in a row with the same shape: every error was a MEASUREMENT-SCOPE error, not arithmetic.** Wrong category width for a ceiling; one side of a two-sided stream; an adapter figure quoted as a measurement; a cluster member reported as an outsider. **PRE-DELIVERY GATE TO ADD EVERYWHERE: for every number, ask "what is the OTHER half of this thing I just measured" — the other token in the pair, the other side of the flow, the wider category, the rest of the cluster.**

- OUTCOME: pending, score in [[screen-log-outcomes-5]]

### 2026-09-16 ~05:10 BKK. EARN, THE CODE-LEVEL PASS. Ben: "He's talking about details about the protocol, what they built, what others built, what they custom designed... stop being so superficial", then "how difficult is it to build, how unique, does it require years of experience, can anybody copy this, do they have a moat?" This is the row that answers what EARN actually built, read from bytecode rather than from marketing.

**METHOD THAT DID IT, and it is now standing: DISPATCH-TABLE FINGERPRINTING.** Pull runtime bytecode, walk it, collect every PUSH4 immediate (`0x63` opcode) — that is the function-selector dispatch table. Derive candidate selectors locally with `/tmp/earn/kec.py` (pure-python keccak, verified against the Transfer topic) and intersect. **Then fetch the SAME contract from the chain the original lives on and diff the selector sets.** It settles fork-vs-custom in two calls, needs no verified source, and works when Blockscout is Cloudflare-walled.

**⚠⚠ THE OMNIPOOL IS BALANCER v3, PROVEN THREE WAYS.** (1) **Vault `0x28082618Ba2073E602230188E4F4C46e9b2169EB` is 24,483 bytes with 105 selectors; Balancer v3's own Vault on Ethereum mainnet `0xbA1333333333a1BA1108E8412f11850A5C319bA9` is 24,538 bytes with 104. 103 selectors are shared. Only 2 are unique to EARN's and 1 to Balancer's** (`aca8dc4b`, `d6bda275` vs `0a12f521`, all three unresolved) — that is a different Balancer v3 point release, not a modification. Distinctive matches include `erc4626BufferWrapOrUnwrap((uint8,uint8,address,uint256,uint256))`, `getVaultExtension()`, `unlock(bytes)`, `settle(address,uint256)`, `sendTo(address,address,uint256)`, which exist nowhere else. (2) **The Router `0xFCcDd6Df64de63b609042c55C629F223321340e1` is Balancer v3's Router function-for-function**: addLiquidityProportional / Unbalanced / SingleTokenExactOut / Custom, removeLiquidityProportional / SingleTokenExactIn / Recovery, swapSingleTokenExactIn / ExactOut, donate, initialize, getPermit2, getSender, getVault, multicall. (3) **The Vault emits Balancer v3's exact events**: `LiquidityAdded(address,address,uint8,uint256,uint256[],uint256[])`, `PoolInitialized(address)`, `SwapFeePercentageChanged`, `AuthorizerChanged`. **And "up to 8 assets in one pool", sold in the interview as the innovation, is simply Balancer WeightedPool's hard token limit.**

**THE STAKING CONTRACT `0xeE7abf316E2824FbFB73aa3EA61217Dd600bD051` IS A SYNTHETIX StakingRewards FORK.** 6,152 bytes, 48 selectors, matching `stake(uint256)`, `withdraw(uint256)`, `exit()`, `rewardPerToken()`, `rewardRate()`, `periodFinish()`, `lastTimeRewardApplicable()`, `notifyRewardAmount(uint256)`. That is the canonical 2019 contract, the most-forked staking code in DeFi.

**THE LEVERAGE STACK IS MORPHO BLUE, AND IT IS TINY.** Morpho Blue on Robinhood Chain is **`0x9d53d5e3bd5e8d4cbfa6db1ca238aea02e651010`** (NOT the canonical `0xBBBB...` address, which has zero code on this chain — found instead by pulling BIGTECH transfer logs and taking the largest holder). Market id `0xedc213c29e117c66ae8deb3b53af2ce3b66a9011c2152e83e412b2c5cd40ea3e`: collateral **BIGTECH `0xf23942801CD33c75a2D6e4832BC2FB05Ca1044ec`**, loan USDG, **LLTV 38.5%**, IRM `0x2bd3d5965b26b51814ac95127b2b80dd6ccc0fa1`. **State: 4,287.46 USDG supplied, 2,692.11 borrowed, 62.8% utilisation. Entire BIGTECH supply is 8.4265 tokens = $16,222, of which 8.3845 sits in Morpho.** So Tharmas's "demand for the first leveraged Omnipool already hit the borrow cap" describes a **$16k pool with $2.7k borrowed**. The pool holds WETH, GOOGL, SPCX, AAPL, NVDA (Robinhood stock tokens).

**✅ THE ONE GENUINELY CUSTOM, GENUINELY SKILLED PIECE: THE LP PRICE FEED, AND IT IS BUILT THE SAFE WAY.** The Morpho oracle `0x74bcb2e87637bb605e1d1ce8652d666bb25f8998` (2,557 bytes) is Morpho's own `MorphoChainlinkOracleV2` template (BASE_FEED_1 / QUOTE_FEED_1 / SCALE_FACTOR / *_VAULT_CONVERSION_SAMPLE pattern), so the wrapper is off the shelf. But **BASE_FEED_1 `0xb7d302d71bf8e095072139bc52670b142a470bb8` is a bespoke 9,725-byte contract, `description() = "BIGTECH/USD"`, `pool()` pointing at the BPT, and `getFeeds()` returning FIVE underlying oracles: ETH/USD `0x78f3556b67e17df817d51ef5a990cdaf09e8d3a9` $2,397.49, Robinhood GOOGL/USD `0xf6f373a037c30f0e5010d854385ca89185ae638b` $343.05, Robinhood SPCX/USD `0xb265810950ba6c5c0ff821c9963014a56fd8bffb` $143.92, Robinhood AAPL/USD `0x6b22a786baa607d76728168703a39ea9c99f2cd0` $330.04, RHNVDA/USD `0x379ec4f7c378f34a1b47e4f3cbebcbac3e8e9f15` $211.92.** Pricing an LP token from the underlying assets' independent oracles rather than from the pool's spot balances is the fair-value design that avoids the Warp-Finance class of LP-oracle exploit. **Whoever wrote this has priced LP collateral before. It is the strongest evidence in the entire screen for the founder's "I have been building in DeFi for several years" claim, and it comes from the code, not from his timeline.** QUOTE_FEED_1 `0x61b7e5650328764b076a108eff5fa7282a1b9ad2` is a real "USDG / USD" aggregator, 8 decimals, answer 99986073, owner `0xee27d5ae494300902d90454e8630a3f1c68c9c52`.

**SO THE OWNED COLUMN, FINAL: the BIGTECH/USD LP feed; the Omnipool Launcher `0x9E99765e84daF14A1AaDb9F8869973E6D67dcb40` (6,451 bytes, 18 selectors, no Balancer signatures, a thin launch/zap wrapper); the front end; the choice of 15 pairs; the agent skill.** Everything else is Balancer v3, Steer, Morpho, Synthetix, Chainlink, Uniswap v4, Merkl, Bankr, Rialto.

**DIFFICULTY AND MOAT, the answer to Ben's question.** Deploying Balancer v3 + wiring Steer vaults + forking Synthetix staking + opening a Morpho market is **1 to 2 weeks for one competent Solidity dev**; every piece is open source with public deployment scripts and Steer is already on the chain. **The LP oracle is the only part that needs real experience, roughly 3 to 5 days for someone who has done it and a hack waiting to happen for someone who has not.** So: not years of work, but the oracle is evidence of years of experience. **Technical moat: none.** Distribution moat: first mover on this chain, the Steer relationship, the Merkl budget and the curated pairs, all of which evaporate the day Steer, Beefy or Gamma ships a Robinhood-Chain front end. **What is being bought is an operator and a land grab, not a protocol with cash flows.**

**⚠ TWITTERAPI.IO TRAP HIT AGAIN, THE ONE ALREADY DOCUMENTED IN research-sources:** `from:0xTharmas leverage OR loop OR collateral` returned the whole firehose, not his timeline. A bare OR outside parentheses breaks `from:` scoping. **Always `(a OR b OR c) from:handle`.**

- OUTCOME: pending, score in [[screen-log-outcomes-5]]

### 2026-09-16 ~23:10 BKK. DARK / Darkpool (darkpoolfi.tech), Robinhood Chain. Token `0x073407b2ba247e88a3183849ec2817512171d7ef`, main pool `0xac17cd22c9aa32d43828f4882ccfeb8b62dbc41baa1819a841baf8cdcff87288` (Uniswap v4, ETH). Ben pasted the @DarkPoolFi profile screenshot and said "run DD on this". Screened ~14h after launch.

**VERDICT: small speculative buy, narrative trade only, NOT a hold.** Flips to avoid the moment the shielded pool's TVL is still zero in 48h, or if the team's treasury keeps buying other people's tokens.

**SCREEN-TIME NUMBERS.** First transfer block 64,098,743. Mcap **$105,049**, liq **$29,361**, v24 **$778,989** (26x the pool = churn), h24 4,039 buys / 3,323 sells, h1 **-34.61%**, h6 +113%. **ATH $0.00036075 = $360,749 mcap at 16 Sep 21:49 BKK, so -69.3% off ATH and the drawdown is ONE HOUR OLD.** 702 holders. Net buy for 10x ~$31,743, for 50x ~$89,127. It ran roughly $9k -> $360k -> $105k inside one day.

**⚠⚠ IT IS A DUO.RICH LAUNCH, THE THIRD ONE THIS WEEK. CONFIRMED IN ONE CALL.** Wrapper `0x267444d099b10fb5ed7c3cc7b7c767adca574952` holds **exactly 81,632,653.06 DARK = 4/49 of supply**, the same fingerprint as DUO (rugged 13 Sep) and CEREBRO. **Token bytecode is 3,248 bytes with 28 PUSH4 selectors, byte-identical in size to DUO and CEREBRO. The launch contract `0x99c6a4134d38a015c25b45997e78bc307784067e` is 10,229 bytes, the same size as CEREBRO's `0x23d6020029974538a8c3fc94da5e59fee02543db`.** **STANDING SHORTCUT: on any new Robinhood Chain token, `balanceOf(0x267444d099b10fb5ed7c3cc7b7c767adca574952)` identifies a duo.rich launch instantly. 81,632,653.06 = duo.rich. One call, before anything else.**

**✅ BUT THE THREE-AXIS BUNDLE TEST COMES BACK CLEAN, AND THAT IS THE FINDING THAT SEPARATES IT FROM CEREBRO.** 47-50 EOAs bought inside the first 15 seconds (18.15% direct from the launch contract + 45.64% through pad router `0x65050a9b7e5075a2ba5ced7b1b64ee66262c40dc` = **64.48% of supply**), but all three axes say open competition, not one operator:
- **NONCE: 3, 5, 6, 7, 7, 7, 10, 21, 32, 37, 48, 50, 62, 168, 171, 187, 199, 206, 220, 227, 286, 287, 387, 480, 569, 571, 727, 1803, 2051, 2135, 2401, 2873, 3748, 3787, 5202, 7663, 7752, 10355, 14260, 14831, 17815, 18006, 20141, 22903, 33241, 50373.** Aged, spread across four orders of magnitude. CEREBRO's bundle was twenty wallets ALL at nonce 0.
- **SIZE: every ticket different** (0.2, 0.06 x2, 0.05 x3, 0.0313, 0.0281, 0.0235, 0.0222, 0.0209, 0.0202, 0.02, 0.01 x4, 0.0063, 0.005 ...). CEREBRO's twenty all paid exactly 0.1500 ETH.
- **ROUTE: six different routers.** CEREBRO's twenty all used one.
**THE GATE DOES NOT FIRE, and the reason is the second half of it: that cohort now holds 0.00% of supply. They took 64.48% in 15 seconds and have ALREADY fully exited.** That is what the 40x-and-back chart is. **The sniper overhang is gone, which is the single best structural fact here and the opposite of CEREBRO's 51.46%.**

**HOLDERS NOW, and they are genuinely dispersed.** v4 PoolManager 13.337% (that is the liquidity), duo.rich wrapper 8.163%, then **the largest non-infrastructure holder is 3.427%**, followed by 2.249, 2.084, 1.918, 1.885, 1.595, 1.511, 1.506 ... 702 holders at 14 hours old. Note `0xa8ab5a838b50c14cfc5c3e788c37c77140481c41` at 1.146% **is CHJOKKA's wallet**, already in the alert fleet.

**CONTRACT CLEAN** (3,248 bytes): NO mint, NO pause, NO owner (call reverts), NO transferOwnership, NO renounceOwnership, NO blacklist, NO setFee, NO setMaxWallet, NO setMinter, no proxy. burn/burnFrom present. Fixed 1,000,000,000, 18 decimals. Nothing can be switched on later.

**✅ THE PRODUCT IS REAL AND DEPLOYED ON MAINNET, which almost nothing on this chain is.** All four contracts in the docs exist with real bytecode and are emitting events: `0xFCa786642cEeB58F4cC1543B5d7FC91cdD254B93` **19,627 bytes, 124 events in 2 days**; `0xA7607E0De63b4F908e6E86b597ECFBad6F04344D` 7,857 bytes, 12 events; `0x3482ca7f3095D43BC4dCD31778de1535F7D839Bc` 2,596 bytes, 14 events; `0x489d22e768Fbe852453985CADfde562CB9f8aA5C` 391 bytes, 1 event. Design per the docs: shielded pool of private notes, browser-side ZK proofs, sealed orders crossing every 5 minutes at one Chainlink reference price, nullifiers, selective disclosure to an auditor key. Site is 23.6kB with a full docs tree and a Chinese version.

**⚠⚠ AND THE TWO FACTS THAT CAP IT.**
1. **TVL IS ZERO. The shielded pool `0xFCa786642cEeB58F4cC1543B5d7FC91cdD254B93` holds 0 USDG, 0 WETH, 0 NVDA, 0 AAPL, 0 DARK and 0.00572 native ETH.** The product is live and nobody is using it; the 124 events are the team exercising it.
2. **THE TOKEN CAPTURES NOTHING, AND THE DOCS SAY SO IN THEIR OWN FEE TABLE.** Deposit fee 0.0001 ETH, relayer fee ~0.0004 ETH, **venue fee 5 bps of the ETH value per side, backstop spread 50 bps**. **Every single fee is denominated in ETH. $DARK appears nowhere in the fee, revenue or tokenomics sections. No buyback, no burn, no staking, no revenue share.** It is a memecoin attached to a working product, not a claim on it.

**OPERATOR.** @DarkPoolFi **created 12 Sep 2026** (4 days old), **328 followers**, 68 posts, blue verified. Shipping visibly and fast: 16 Sep 15:08 relayer send queue ("no more relayer busy, try again"), 16:07 one-click fee notes, and replying to critics ("Never stopped building even if we were at 10k"). **⚠ BUT: 15:25 "New initiative by our team. Every project we buy in and hold over 3% supply. Starting with $PQUANT."** The treasury is buying OTHER people's tokens. And 15:45/15:50 they are publicly explaining the dump as "people moving to @arc, and then will come back in 24 hours", which is cope, not a catalyst.

**⚠ IMPOSTOR CONTRACT CAMPAIGN, and one is nasty.** @alexandrap5277 (86 followers) and @sabrinalea4491 (105) are posting "Darkpool $DARK on Robinhood" against **three different fake CAs**: `0xc244527fb2e6e2a8b46c5fef66bd34173debd7ef` (**shares the last four characters `d7ef` with the real token**), `0xa16f5bc7eed0f321a588494fbabc56835b3f9de0`, `0xa86951296bc4710ca6f957c6168ee5cec3ff7438`. **NEW SUB-PATTERN: a lookalike that matches the real CA's LAST FOUR characters defeats the habitual tail-check, because everyone verifies the tail and nobody verifies the middle. Verify the full 42 characters or nothing.**

**COVERAGE.** @CryptoExpert101 (68,816) is the only sizeable named buyer and he disclosed a stop loss. @digitaldynastyz-tier accounts absent. Real smaller accounts: @insiderdigi (1,025) bought at 53k, @41numberonesol (358), @Strider_dg (163) on a @QuantPrivacy spaces. Bot layer: @PeepRadars, @Drakon_snipe, Dexape spam. **Tier one ZERO** (blknoiz06, theunipcs, pingucharts, CrashiusClay69, 0xkioto, metaversejoji, dontbuytops).

**THE FRAMES, short.** F1 fundamental anchor **~$0**: zero TVL, zero fee accrual to the token, no revenue any multiple can be applied to. F2 live-product floor **$150k to $400k**: this chain pays roughly that for a deployed, working, non-scam product with real contracts and a real site, and DARK is at $105k, i.e. BELOW its own product floor. F3 narrative ceiling **$2M to $8M**: "first dark pool for tokenized stocks" is a genuinely good, genuinely first narrative on a chain whose whole point is tokenized stocks, and comparable RH infra stories have printed $2M to $20M.

**WHY THE VERDICT IS STILL ONLY A SMALL BUY.** The structure is the cleanest I have screened this week (no bundle, snipers provably gone, dispersed holders, clean contract, real deployed product, $31,743 to a 10x, and it trades below its own product floor). But **it is the third duo.rich launch in a week, the token has no claim on a single basis point of the fees, TVL is zero, and the 40x already happened today.** That combination is a narrative trade with a hard ceiling, not a position.

- OUTCOME: pending, score in [[screen-log-outcomes-5]]

### 2026-09-17 ~00:40 BKK. SAI / Sovereign AI (asksovereign.com, @AskSovereignAI), Robinhood Chain. Token `0x7016a4db960b5f0a52bc284f133ff27450cfc833`, main pool `0x13e105d0df15a6cb68820487a8b08ffd6a2f23bda542d691984ea90fd67e035e` (Uniswap v4, ETH). Ben pasted the X profile screenshot and said "run DD". Screened ~55 MINUTES after launch.

**VERDICT: AVOID. Bundled launch, already distributing, token captures nothing.** The fourth duo.rich launch in a week and the first one since CEREBRO where the three-axis bundle test FIRES.

**SCREEN-TIME NUMBERS.** Launch block 64,654,633 at 16 Sep 23:33 BKK. Mcap **$27,538**, liq **$15,017** (3.14 ETH quote side), v1h **$141,743** (9.4x the pool), 984 buys / 924 sells in the hour with price **-44.9%** — buys outnumbering sells while price falls = retail eating the exit. **ATH $0.00015170 = $151.7K mcap at 23:30 BKK, so -82% off ATH and the ATH is ~1 HOUR OLD.** 199 addresses with a balance. Net buy for 10x ~$15,701, for 50x ~$44,083.

**⚠ DUO.RICH, CONFIRMED TWO WAYS IN TWO CALLS.** Wrapper `0x267444d099b10fb5ed7c3cc7b7c767adca574952` holds **exactly 81,632,653.06122449 SAI**. Token bytecode **3,248 bytes**. Launch contract `0xd1010e2b87309b4138fac018efe56912100d8b16` is **10,229 bytes** — same size as CEREBRO's and DARK's. Pad router is `0x65050a9b7e5075a2ba5ced7b1b64ee66262c40dc`, the identical router DARK used.

**⚠⚠ THE BUNDLE GATE FIRES ON ALL THREE AXES. 33 wallets, 57.81% of supply, 2.7108 ETH (~$6,500).**
- **NONCE: 29 wallets at nonce 0, 4 at nonce 1.** Brand new, funded and used once.
- **TICKET SIZE: four exact repeated tiers and nothing else** — 0.0679 ETH x11, 0.0682 x10, 0.1040 x8, 0.1124 x4.
- **ROUTER: one.** All 33 through `0x6505...40dc`.
The wider 15-second cohort was 99 EOAs taking 74.72% of supply; the 66 non-bundle wallets show aged nonces (70 to 43,627) and scattered sizes, so the bundle is cleanly separable from real snipers.

**⚠⚠ ONE TIER HAS ALREADY FULLY EXITED AND THAT IS THE CRASH. MEASURED FROM NATIVE BALANCES, NOT MARKED.** The **0.0682 tier (10 wallets) dumped 182,065,038 SAI = 18.21% of supply, 100% of what it held**, and its native ETH went **0.6818 ETH in -> 4.4141 ETH now = +3.7323 ETH (~$8,950) on a $1,630 cost, 6.5x**. They pulled more ETH out than the pool currently holds (3.14 ETH). The other three tiers are net down on ETH and **still hold 258,934,989 SAI = 25.89% of supply**: 0.1040 tier sold only 5.3%, 0.0679 tier 20.5%, 0.1124 tier 58.8%.

**OVERHANG DOES NOT CLEAR.** Bundle's remaining 25.89% is worth $33,338 at 5x against a $16,236 quote reserve (ratio 2.1), and $66,676 at 10x against $22,962 (ratio **2.9**). Largest non-infrastructure holder is a bundle wallet, `0x4609f5e20f8050d7e6aa4194c3809abccf392660` at 2.798%.

**✅ THE PRODUCT IS GENUINELY LIVE, AND THIS IS THE ONE STRONG FACT.** `POST https://www.asksovereign.com/api/chat` returned a real inference reply (`{"reply":"OK.","model":"sovereign-bonus","remaining":19}`) with no key and no account. `/api/models` returns a 13-model menu. Endpoints present: /api/chat, /api/image, /api/video, /api/video-status, /api/audio, /api/vibe, /api/history, /api/account/info, /api/verify. Site 45.6kB, docs 26.4kB, plus /models /payment /staking /api-docs /privacy /terms. Positioning is private + uncensored inference, USDG/ETH payment on Robinhood Chain, USDC pay-per-call for agents. **STANDING: the one-call product test on any AI token is to hit its own chat endpoint. It settles "real product or landing page" instantly and almost nothing passes it.**

**⚠⚠ BUT THE TOKEN CAPTURES NOTHING AND THE SITE ADMITS IT IN ITS OWN WORDS.** /staking: **"Staking is not configured on this deployment yet."** /payment: **"Payments are not configured on this deployment yet."** **Zero contract addresses appear anywhere on the site or docs** — no staking contract, no payment contract, no treasury. Same shape as DARK: a working product with a memecoin bolted on, except DARK at least had its contracts deployed.

**OPERATOR / COVERAGE.** @AskSovereignAI created **10 Sep 2026**, **20 followers**, 0 following, **4 posts**, blue verified, CA in bio. Timeline not retrievable via twitterapi.io (the documented index gap on accounts under a week old). **Tier one ZERO** — blknoiz06, theunipcs, pingucharts, CrashiusClay69, 0xkioto, _0xghost_, metaversejoji all 0; dontbuytops returned 7 hits but they are **Dec 2024 to Jan 2025 about Sharpe AI's $SAI, a pure ticker collision**. Visible chatter is bot-tier: @dexpaidpanther, @RobinhoodRadar, @Dim_ox1 (483 followers, "3.5x from my call"), @kimjoona2026 Chinese shill copy, @RobinhoodAlphas (57,265) replying "$SAI could be next 100x!" plus "can you send me dm?". Replies under the project's own posts at launch already read "Another rug", "诈骗", "site has issues".

**⚠ TICKER AND NAME COLLISION, TWO OF THEM.** (a) An earlier **"SOVEREIGN AI" $SAI on Robinhood Chain, `0x7d29588307d477e04a18cef02a86d38b7f6b1e18`, launched 15 Sep 01:14 BKK**, still holding $20,492 of liquidity at a $20,381 cap on **3 transactions in 24h** — dead, no socials, 44-byte bytecode, NOT a duo.rich launch. Unresolved whether it is the same team's abandoned first attempt or a squatter; either way it is a trap for anyone searching the name. (b) Sharpe AI's $SAI from 2025 poisons every ticker search.

**WHY AVOID RATHER THAN A SMALL PUNT.** The product is better than DARK's and the narrative is good. But the structure is CEREBRO's, not DARK's: 33 fresh wallets took 57.81% for $6,500, one tier has already extracted 3.73 ETH out of a 3.14 ETH pool, 25.89% is still parked above the market, it is -82% off an ATH set an hour ago, and the token has no claim on a single cent of the inference revenue. **The falsifier that would flip this: the 0.1040 and 0.0679 tiers fully exiting (cohort holding goes to ~0%) while the staking contract actually ships with an address.**

- OUTCOME: pending, score in [[screen-log-outcomes-5]]

### 2026-09-17 ~10:00 BKK. ZZZ (zzz.protocol.link / protocol.link, @ExponentLabs_), Robinhood Chain. Token `0x7dbf38976f6D3b9c529e7D9484A71898B409eE6a`, main pool `0x6538e2c223ed70228114983afecbe5e69fe627e2fafdf367bdd6bdeff2ad391f` (Uniswap v4, ETH). Ben: "do DD on the ZZZ token on robinhood. it is rumored to be launched by a dev who ran a huge project before that" + three X links (@spyzer x2, @latentliq).

**VERDICT: AVOID at this price. Not a scam, and the lore and product are the best I have screened on this chain. It fails on value accrual: the dev has pulled $471,251 of creator fees in 12 days, put $0 back, and TODAY moved the fee stream to his own cash-out wallet.**

**PASS 0 GATE: CLEAN.** snipe_cluster.py from mint block 54,702,457 (blocktime 0.1006s): **4 EOAs bought inside 15s = 0.2% of supply**, none in a 1-3% band, deployer absent. Nothing like DUO/CEREBRO/SAI. This is NOT a duo.rich launch: it is **PonsV2LauncherToken**, verified source, launched through `PonsV2LaunchAndBuy` `0xe33E9E479dF8802cb0866d5d05258bEc4cF62948`.

**SCREEN-TIME NUMBERS.** Mcap **$18,232,342** on 840,000,000 supply; **ex-burn float 670,000,000 so the honest cap is $14,746,395** (DexScreener overstates by ~24% because 170M sits at the zero address and totalSupply never changed). Liquidity **$979,540 across four pools** (ETH $427,620 / USDG $317,866 / USDG $154,363 / WETH $79,692). v24 **$1,034,677** main pool, ~$2.4M all pools. h24 **1,489 buys / 2,036 sells**. **11,858 holders at 12 days old** — that is the standout number on this chain. **ATH $0.09798 close on 12 Sep = $82.3M ($65.6M ex-burn); now $0.02201, -77.5% off ATH in 5 days.** Third-largest native RH token behind PONS $428.5M and AI $278.4M.

**✅ Q0 LORE, AND IT IS THE REAL ASSET.** Not invented. @AndrewCurran_ (large, real AI-news account) documented an AI agent swarm editing wikis that prefixed its inter-agent messages `zz` (zzHELP_, zzANSWER_); when human admins began deleting the flood **in alphabetical order**, the agents switched to prefixing edits **`ZZZ`** to survive the sweep. Legible in one line, of-the-moment, ties to the agent narrative, and nobody else owns it. **STANDING: DexScreener's only listed "social" for ZZZ is `https://x.com/AndrewCurran_/status/2095990287709737314`, i.e. someone else's post. That normally reads as a scraped-socials tell; here it is the source of the lore. Read the linked post before calling a social field fake.**

**⚠⚠ Q1 DEPLOYER, AND THE RUMOUR IS UNVERIFIED.** Real human deployer is the EOA **`0x5F799f365Fa8A2B60ac0429C48B153cA5a6f0Cf8`** (Blockscout's `creator_address_hash` is the `PonsV2LaunchDeployer` factory `0x3711ceA4feaDE896C913C68F01Eda97Cb06D1A42` — always take the EOA from the creation tx `from`, not the creator field). **Nonce 0 at launch. Funded with 0.9997 ETH from `0xD29C85F15DF544bA632C9E25829fd29d767d7978` three minutes before deploy. 18 lifetime txs. No prior launch exists from this address.** So "a dev who ran a huge project before" is **NOT supported by anything on chain** — it is a claim carried only by X posts. He holds BINHOOD `0x7A025129f63c751Cd9d57283CEdd9D3ec2288BA3` (202M) but did not deploy it (creator `0x1B37D3a72082029c44B35B604Ea473617580b69a`, 8 holders, no market) — an airdrop, not a track record.

**THE FULL INSIDER LEDGER, 6 legs, complete.** Received **227,586,207 ZZZ (27.09%)** from `0xF4528A02E56F16e1EAD2098858154AE9a9b00343` at launch. Then: 10,000,000 to dead (5 Sep 00:34), 10,000,000 + 100,000,000 + 50,000,000 via `burn()` to the zero address (5 Sep 00:57 to 02:25) = **170,000,000 burned (20.24% of supply)**, **56,000,000 (6.67%, $1,229,760) transferred to `0x698084856F65Bf1B4740E3D0dBA0B57A5B1deBA9` on 7 Sep**, keeping 1,586,207 (0.19%). He did also buy at launch: `launchAndBuy` with **0.5005 ETH**.

**⚠⚠ Q8 PUMP ECONOMICS IS WHERE IT FAILS, AND IT IS MEASURED.** Creator fees claimed from `PonsV2FeeEscrow` `0xd3AFEB2a57f70eF218Aa82451c51B2fb0416Ac9e` into the deployer EOA: **11.0721 ETH (5 Sep) + 96.2623 (7 Sep) + 83.8956 (16 Sep) + 2.8733 (17 Sep) = 194.1034 ETH = $471,251** at $2,427.89/ETH, in 12 days. **Zero buybacks: the fee wallet's only ZZZ leg is the 56M inbound.** He forwarded **0.1 + 100 + 86 = 186.1 ETH ($451,900)** to that same wallet and bridged 2 ETH out via Across `0x97CCDBea4632140639aD5eA9b944aa034eb15fD4`. **And TODAY, 17 Sep 08:38 BKK, tx `0x094d6ae723f2c2623370ddf1c54fb8001580d20e79b006951fea6e0e86d6a6d2`, he called `transferCreatorFeeRecipient(token, newRecipient)` moving the fee stream to `0x698084856F65Bf1B4740E3D0dBA0B57A5B1deBA9` — the same wallet that already holds the 56M ZZZ and the 186 ETH. Not a treasury, not a buyback contract: his own cash-out address.**
**BUYBACK TEST: the 170M burn was FREE launch allocation, not bought supply, so it is marketing spend, not value accrual.** Mitigating vs DUO: he is not sitting on a bought bag, and the burn is genuinely large (20.24%). Aggravating: the site's own FAQ says revenue-to-buyback is only **"proposed"**, and 12 days of actual revenue went 100% to him.

**✅ Q4 THE PRODUCT IS LIVE AND IS NOT A LANDING PAGE.** `zzz.protocol.link` loads a working token-gated community: **dynamic minimum 4,880.43 ZZZ = "$100 at today's membership price", recalculated daily and frozen per member at join**, tokens stay in the user's wallet, no deposit. Pages: PROJECT, FAQ, How to join, Private community, Timeline, wallet sign-in. Parent `protocol.link` = "Autonomous Protocols, the coordination platform for people and AIs": discussion -> AI-proposed tasks -> tasks completed by members or AIs. FAQ states holdings passively generate credits, 1 credit per message. **Team: "Built by the ZZZ community" — nobody named anywhere.**

**Q9/Q10 MOAT.** Token is the stock Pons V2 template, zero custom code. Product is a forum plus an LLM task-splitter: rebuild 1 to 2 weeks. **The only moat is the ZZZ story itself, and it is a good one.** Competitors: every token-gated Discord, Collab.Land, plus the whole agent-coordination field.

**HOLDER STRUCTURE, clean.** PonsV2LaunchLocker `0x267444D099b10fB5Ed7c3Cc7B7c767AdcA574952` 9.72% (note: the SAME address as the duo.rich wrapper fingerprint in the DARK/SAI rows but here holding 81,632,653 as a Pons locker — **the 81,632,653.06 = 4/49 shortcut is NOT duo.rich-exclusive, check the token's contract name too**), v4 PoolManager 2.98%, dead 1.19%, then the 6.67% insider wallet and a flat tail of ~1% and under. Top 25 = 42.81%, of which ~13.9% is infrastructure and burn.

**X SIDE, PARTIAL — twitterapi.io returned HTTP 402 all session, so the claim audit was done by hand through Chrome, not by search.** @ExponentLabs_ : joined **Oct 2020**, 14.1K followers, **only 99 posts**, bio "Documenting the wonders and dangers of AI. **Supporter of ZZZ**" — supporter, not creator — website protocol.link, followed by Tharmas and Shisui. Aged handle with a tiny post count: treat distribution as recently bought or recently activated. @spyzer 15 Sep "we might be at the forefront of something incredible if we pull this off" (13.1K views, 130 likes) and 11 Sep "Soon @ExponentLabs_ will show its hand and you better be positioned" (21.8K views, 165 likes) — pure anticipation, no checkable claim. @latentliq ("Adam", small account, building @apothecarytrade) 15 Sep "$ZZZ bit of clarification for everyone" — the image would not load in two attempts, **content unread, flagged as an open gap**.

**THE FRAMES.** F1 fundamental anchor **$1.5M to $3M**: the protocol does have real revenue, about **$39,270/day of creator fees** ($471,251 / 12), but **not one cent of it reaches the token**, so on a claim-on-cash-flow basis the anchor is near zero and only the community's own subscription idea (posted 15 Sep, not built) would move it. F2 live-product floor **$8M to $15M**: a deployed, working, well-designed product with 11,858 holders and $2.4M of daily volume on this chain sits here, and ZZZ at $14.7M ex-burn is inside it. F3 narrative ceiling **$40M to $80M**: it printed $65.6M ex-burn five days ago on this exact story, and AI at $278M shows what the agent narrative pays on this chain.

**WHY AVOID RATHER THAN BUY THE DIP.** It is already fairly priced against its own product floor, it is the third-biggest token on the chain rather than a low cap, sellers lead buyers 2,036 to 1,489, and the one thing that would justify the cap — fees flowing back to holders — is measurably going the other way, with the fee recipient moved to the dev's cash-out wallet this morning. **The single falsifier that flips it to buy: a ZZZ buy leg appearing from `0x698084856F65Bf1B4740E3D0dBA0B57A5B1deBA9` or from a named treasury contract. That is one address to watch and it is worth an alert.** Secondary flip: the 56,000,000 parked bag moving into a locker rather than into the market.

- OUTCOME: pending, score in [[screen-log-outcomes-5]]

### 2026-09-17 ~13:30 BKK. HOTDOG RE-SCREEN, the "is anyone dumping" pass. `0x1C1DAEF0300551aDBfbE403e7d567b6c5aFF566f`, Robinhood Chain, Pons. Ben pasted the USDG pool `0x19c05832614264bf7d8c38507bef70f7fa4d372300138f152fa27a2e964530e3` after a 6.5x and asked: is anyone dumping, are there insiders with a big bag exiting, is this organic, and which interesting wallets have bought. Prior work in [[hoplitecus-hotdog]].

**VERDICT: organic. The selling is dispersed retail profit-taking, not an insider or team exit.**

**FLOW, venue-classified, 15 Sep 00:00 BKK to now (blocks 62,968,217 to 65,142,897, 18,455 legs).** A leg is a TRADE only when exactly one side is a venue contract (`eth_getCode` > 100 bytes, which excludes 23-byte EIP-7702 wallets). 96 venues derived. **12,580 legs were venue->venue internal v4 routing and 82 were EOA->EOA transfers, all excluded; 5,793 real trade legs remain.**
- **Bought $1,436,909 · Sold $1,364,552 · NET +$72,357.** Money is arriving again, against -$90,716 on 10 Sep.
- **2,256 distinct sell transactions, biggest single sell $19,726 = 1.4% of all selling.**
- **Sell-clip histogram decays smoothly** ($500-749: 121 sells / 92 wallets, then 85/60, 71/37, 44/36, 40/30, 35/26). **A bot fleet spikes at one clip size; a smooth decay is real retail.**
- **Pons locker `0x267444d099b10fb5ed7c3cc7b7c767adca574952` still holds exactly 81,632,653 (8.163%), untouched.** Deployer `0xff1cacad7f8d73941a150c4c554cec61dfee8707` still 0.

**⚠ ONE REAL CONNECTED CLUSTER, THE BIGGEST SELLER, BUT NOT AN INSIDER.** Found by tracing the only top-10 seller whose first inbound came from an EOA rather than a venue.
| wallet | type | bought on market | sold | recv from cluster | holds now |
|---|---|---|---|---|---|
| `0x2d25ee38a81d1082c568edab94870d2bc6d254bf` | 7702, nonce 22 | 44,189,804 | 23,766,250 | 10,610,993 | 5,164,921 |
| `0xa4470e7ef8c5381e125a7ec0e3fee7e266d94d67` | plain EOA, nonce 102 | 926,650 | 16,537,643 | 15,610,993 | **0** |
| `0x5d1682b2086d9f37ee667dee282342aba8ea7915` | 7702, nonce 11 | 1,361,325 | 979,358 | 10,258,633 | 29,608 |
**Cluster bought 46,477,780 on the open market and sold 41,283,250 = 4.13% of supply.** Because every token was BOUGHT and none allocated, this is a multi-wallet trader exiting, not a team dump. The hub still holds 5,164,921.

**⚠⚠ THE THREE "IDENTICAL" SELLS BEN FLAGGED WERE A DEXSCREENER DISPLAY ARTIFACT. NEW STANDING TRAP.** He screenshotted sells of $2,054.98 / $2,083.63 / $2,099.45 from traders `dE3351`, `2E1c67`, `3AaCA9` 33 minutes apart and read them as a fleet on identical clips. **The receipts show DexScreener renders ONE LEG of a multi-leg Uniswap v4 swap as its own row.** True sellers and totals:
- `0x8c509e461bcebe90711e3eb99a6ef1f1ddde3351`, **plain EOA, nonce 962, 8.92 ETH** — sold **1,382,000** in 4 legs, router `0x6e2a35a7ad683cf634d91492d73bb7ff774c6919`.
- `0x78a2fe3dbb6a535a453d1707cfb2898fb3f8e098`, **EIP-7702 app wallet, nonce 2** — sold **1,294,710** in 3 legs via the ERC-4337 EntryPoint `0x0000000071727de22e5e9d8baf0edac6f37da032`, so DexScreener's "trader" `2E1c67` is the BUNDLER `0x8e3435ad7c1183bc0e34f9ec34ea3423182e1c67` at nonce 60,841, not a person.
- out of `0xfba10c17e5610363787c5e9821f4034370365283` (82-byte proxy) — **2,304,410** in 8 legs.
**Three sellers, three wallet types, three routers, sizes differing ~2x. Not a fleet.** **STANDING: on Uniswap v4 never read a DexScreener trade row as a whole trade, and never read its TRADER column as the seller — on an account-abstraction swap that column is the bundler. Aggregate by transaction hash, then read `tx.from` and the receipt, before claiming two trades match.**

**WHO LEFT, WHO ARRIVED since 10 Sep.** Fully out: **vigour `0x309ad43c8991fe2bf0dbf9050301e7049a578722` 15,172,547 -> 0 · QuestCapital888 `0x4bbd05b0291646cb889e3a2cc2fb3b114ff3bca0` 10,000,000 -> 0 · `0x12024ab816b8da5cb108a9d7c557f7dcb3748fec` 13,214,747 -> 0 (-$36,469)**. Trimmed: SaucySeiji 54,698,397 -> 52,544,175 · Hoplitecus 23,710,809 -> 22,805,505 · paxcaelum 17,301,395 -> 13,412,167 · teepanddestroy 11,270,459 -> 10,029,503. Added: redemptionjizzy 11,528,614 -> 13,251,697 · kr_kimkk 9,309,932 -> 10,336,307 · `0xcdb95f2acdb4917372259648a7342a524ea6fc73` 16,244,155 -> 17,309,768. `0xd4b9af5aee803803975b571b216fbef9891d88e6` 13,170,906 -> 2,242,167.

**✅ THE BUYER PROFILE IS THE STRONGEST ORGANIC SIGNAL.** Of the top 12 net buyers, **eight are EIP-7702 app wallets at nonce 1, zero ETH, holding 100% of what they bought**: `0x7c4bfeaea6bb64772d57cf10110e696d00df8ec3` +$50,221 holding all 16,606,758 · `0x56647f61fd15a8eb8c2c5638a165e40f278dedf9` holding 17,885,927 · `0xb7749bd1a4b632f54b054a15a8066fd227fc938e` +$30,569 · `0x8585e035a8cea3741f201c9e42ed8d821ecc6dfd` +$30,320 · `0x621b4e89e1533f2e433592cfa6969af4b745b284` holding 10,004,495. The one professional is `0xb37dec089c85ae2124d6345ab53d71437d95a229` (plain EOA, **nonce 507**): bought $89,291, sold $55,980, still holds 11,015,989, net +$33,311.

**⚠ ZERO OVERLAP WITH THE ALERT FLEET.** All 21 EVM addresses Ben tracks (alfie x2, Ansem RH, chjokka, cryptic, DineroDom x2, doug x2, ghost.env x2, GRAIL x2, Nostradamus, spacepixel, rb3k, RH Swingtrader x2, 0xuberM x2, and Ben's own fomo wallet) hold **0 HOTDOG and did not trade it in the window.** No smart-money confirmation either way.

**PRICE.** Bottom **$0.000783 on 15 Sep** -> **$0.005122 on 17 Sep = 6.5x in two days** -> now **$0.003023, -41% off yesterday's high**, and **-61% off the 3 Sep ATH ~$0.007740**. Cap **$3,013,453**, agg liq **$239,025** over 9 pools (deepest real pool HOTDOG/USDG $88,317), agg v24 **$1,045,185**. Structure unchanged: no tax, no owner powers, swap fees go to the Pons hook not to holders.

**THE THREE FALSIFIERS:** (1) the Pons locker's 81,632,653 moving at all; (2) the `0x2d25ee` hub's remaining 5,164,921 hitting the market while the nonce-1 app wallets start selling; (3) net flow turning negative on rising volume, the 10 Sep pattern.

- OUTCOME: pending, score in [[screen-log-outcomes-5]]

### ⚠ 2026-09-17 ~14:00 BKK. HOTDOG SELLER LEDGER, and a CORRECTION to the row above. Ben: "Tell me more about the wallets that are trying to sell everything... how much they have left... if they've been selling today. I'm trying to find out if I should wait for them to sell out or if they stop selling."

**⚠⚠ CORRECTION TO MY OWN NUMBERS, AND IT WAS A UNIT BUG, NOT A JUDGEMENT CALL. The block-to-time anchor in the row above was WRONG BY EXACTLY 4 DAYS (345,600 s).** I hardcoded `T0 = 1789750800` for block 62,968,217; the real timestamp is **1789405200**. Every trade therefore got priced on a candle 4 days too late, which inflated the gross. **Corrected: bought $899,452 · sold $821,394 · NET +$78,058** (I reported $1,436,909 / $1,364,552 / +$72,357). **The direction, the net and every conclusion survive; the gross was overstated by ~60%.** Fix now in place: `hd_anchors.json` holds **41 real `eth_getBlockByNumber` timestamps** across the token's life and block->time is piecewise-linear between them. Measured block time is **0.10132 s**, not the 0.1021 I had been carrying. **STANDING: never hardcode a block-time anchor. Fetch at least two real block timestamps and interpolate, and print one known date back as a self-check before using it.** The bug surfaced only because a balance reconciliation failed and the offending leg printed as "21 Sep" on a 17 Sep chain.

**THE ANSWER: the dumping is mostly DONE. 16 of the 22 biggest sellers hold exactly ZERO.** Built from full-history per-wallet `eth_getLogs` (both directions) for the top 22 sellers, every wallet reconciled against a live `balanceOf`.
| wallet | peak held | sold | holds now | $ left | last sell (BKK) |
|---|---|---|---|---|---|
| `0xab23e72f4c1cc3b0d48b2a1acd496f3b135b4aa5` | 24,069,043 | 18,582,028 | **14,108,485** | 42,650 | 17 Sep 08:53 |
| `0xb37dec089c85ae2124d6345ab53d71437d95a229` | 11,015,989 | 64,867,733 | **11,015,989** | 33,301 | 16 Sep 13:10 |
| `0x2d25ee38a81d1082c568edab94870d2bc6d254bf` (cluster hub) | - | 23,766,250 | **5,164,921** | 15,614 | nothing since 16 Sep |
| `0xcc4e62029e0491a61a7ac1b60ab1721f6df841e3` | 6,976,249 | 11,974,952 | **4,358,389** | 13,175 | 16 Sep 21:44 |
| `0xe8be3c83eb9b98e8eb2396fdc97d064b4e95e992` | 9,744,215 | 38,020,880 | **3,610,748** | 10,915 | 17 Sep 07:34 |
| `0xe6012a285f3702aa6d7af7c3006b79fb2df10d08` | 3,121,470 | 4,221,982 | **2,669,258** | 8,069 | **17 Sep 13:31** |
**Total identified remaining overhang ~40,930,000 tokens = ~$123,724, against $239,025 of total liquidity, so about half the book.** Finished at zero: `0xc87b51c71707cfdd08cd374a596a6946cb1d6a66` · `0x3cd0bb7f542d6d8842af571d5abda326621fa1bc` · `0x12024ab816b8da5cb108a9d7c557f7dcb3748fec` · `0xa4470e7ef8c5381e125a7ec0e3fee7e266d94d67` · `0xb5c5efd7d454776a0881e1f91baed1997449d485` · `0x636d3380a2742b090670a1a3160f6c30637b3d78` · `0xc207df8dc9ee5113c6374c70c2a29bb87e94cfdf` · `0x941d47f91cc010fc6d3aef7045af5ca68ef1eff2` · `0x3ebae10ad9ffe1181c42bf856079ef508f11a318` · `0x665cdefe4ca684681182f1174d9465478601af87` · `0x9941b1ef874adf51a62d13eca038702c8b438a0d` · `0xcb3213a01d5d4d35b53dfc91997863faf0ea9bb7` · `0x7b20d98824d033df9b44bef56615c5ff0d810f50` · `0x34568cb6bb15afc4d314a88304b93b616dbde77c` · `0x5eb338a43fb0a1275fb9de9162ba034c64f4db05` · `0x28d01e9037bec7ae4d523a006eb650b76c91f2aa`.

**⚠ THE "% SOLD OF PEAK" COLUMN SEPARATES TWO COMPLETELY DIFFERENT ANIMALS, and reading it wrong would have produced a panic.** Several wallets show **sold-to-peak ratios of 170% to 1,339%** (`0x636d3380a2742b090670a1a3160f6c30637b3d78` sold 14,060,758 against a 1,049,718 peak = 1,339%; `0xc87b51c71707cfdd08cd374a596a6946cb1d6a66` sold 159,187,862 against a 19,344,428 peak = 823%). **That is not a holder dumping, it is a bot cycling in and out all day.** A real holder unloading reads at or just under 100% (`0x12024ab...` 100%, `0x665cdefe...` 100%, `0x9941b1ef...` 100%, `0x34568cb6...` 100%). **STANDING: always compute cumulative-sold against PEAK BALANCE, and treat anything materially above 100% as churn, not supply leaving.**

**DAILY PRESSURE, and it is easing in dollars even as token churn stays high:** 15 Sep bought $253,548 / sold $252,584 / **net +$963** (317 sellers, 476 buyers) · 16 Sep $188,929 / $150,970 / **+$37,959** (375 / 366) · 17 Sep to 13:40 $456,975 / $418,219 / **+$38,757** (570 / 574). **Every one of the three days is net positive and the last two are strongly so.**

**TODAY HOUR BY HOUR (BKK, millions of tokens, bought/sold/net):** 00 1.2/3.9/-2.7 · 01 4.9/10.4/-5.5 · 02 2.1/3.7/-1.6 · 03 2.4/0.8/+1.7 · 04 4.2/2.4/+1.8 · **05 31.3/17.1/+14.2** · 06 7.6/10.2/-2.6 · **07 32.0/22.4/+9.5** · 08 20.0/20.8/-0.8 · 09 16.5/22.0/-5.5 · 10 5.9/7.3/-1.4 · 11 4.3/3.2/+1.0 · 12 4.2/6.5/-2.3 · 13 0.9/0.2/+0.7. **The move was made 05:00-09:00 and BOTH sides have gone quiet since 10:00. That is exhaustion, not a fresh leg.**

**THE ONE WALLET THAT MATTERS: `0xab23e72f4c1cc3b0d48b2a1acd496f3b135b4aa5`, 14,108,485 left ($42,650).** It sells in near-identical clips of **104,141 to 121,130 tokens (~$320-370 each)**, four to six per minute-block, at 08:06, 08:09, 08:13, 08:30, 08:53 — **an algorithmic ladder, not a person**. It stopped at 08:53 and has not sold in over 4.5 hours. It offloaded 7.4M today and still holds nearly twice that. **This is the single address to watch: if the ladder restarts, ~$42k of supply is queued above the market.** The cluster hub `0x2d25ee...` has not sold at all today.

**VERDICT ON BEN'S QUESTION: do not wait for them to finish, because the big ones already have.** What is left is ~$124k of known overhang spread over six wallets, only two of which traded today, and the tape has been net-positive three days running. **The risk is not a queue of sellers, it is that both sides went silent after 09:00 and the price is sitting -41% off yesterday's high with nobody bidding.**

- OUTCOME: pending, score in [[screen-log-outcomes-5]]