# Analyst 3 status archive (older entries, newest first)

- **07 Oct 11:50 Bangkok: v1.13 A3.13 DONE. Verdict: KILL (clean).** Bundle-sold exit, X = 10% (best on half A): live book A +$156 a day, B -$4,098; kit300k A -$620, B -$5,498; beats random 35.5% / 0% / 0.5% / 0% (bar 95%). The bundle has usually already sold by our 5x (26 of 41 bundle trades at 10%+), so the rule becomes "sell at 5x" and cuts the runners (0x8e41fd25e85e624560fa50ff9f117a02cd30b8b3 29.15x to 4.48x, 0x9ba4d84ac1571a0788b2bb7a6412e73529d12dc6 28.00x to 4.53x). The second leg is outside money; the operator's own selling does not mark the top. Traced 7,113 fresh wallets' funders on research-box. http://159.223.88.168/hx-c9ois/sw/sw-a3-13-operator-exit.md
- **07 Oct 11:00 Bangkok: v1.13 A3.13 STARTED. Rule PRE-REGISTERED (before any result).**
  - Coins: every kit coin held past 5x (KIT HOLD) in (a) my live wide-gate book and (b) the A3.12 kit300k trades. Same engine for both: entry as before, 5x, fills by tick walk at block end.
  - Bundle: the coin's buyers while cap was under $300k (router buys resolved to the signer), traced with an archive ETH balance search: fresh = zero ETH 3 days before the buy; funder = sender of the first funding tx (direct, or the signer of a disperse call to 0x84157d3d24630c297849c84d49ae7f1805c1b60f / 0x03802649e469ed5680c58dff9e65425e22d21cc5 / any contract whose input names the wallet). Bundle = every pre-$300k buyer funded by the coin's top funder. Fewer than 3 bundle wallets = no bundle, keep the live trail.
  - Bundle sold share at time t = tokens sent out of bundle wallets to anyone outside the bundle / tokens received by the bundle so far (from the coin's Transfer logs).
  - Exit rule: after the 5x, sell all at the first moment the sold share reaches X (if already past X at 5x, sell at 5x), else the live 50% trail / pull / 48 h. X in 10/25/50%, picked on half A only (both books together).
  - PASS = $ a day above the live exit in BOTH halves for both books, and beats 95% of 200 random exit times (same coins, uniform time between the 5x and the live exit). Else KILL.
  - Why this is not the killed drip-seller, net-buy or pull-tell exits: those read anonymous flow; this keys on the operator's OWN wallets, named before the run from how they were funded.
- **07 Oct 10:52 Bangkok: v1.11 A3.11 DONE. Verdict: KILL (clean).** Chosen cell K3 N5 (best half A rolling, pre-registered): fixed cohort half B -$407 a day; rolling A -$559, B +$2,599; beats random 49% / 40% / 89% (bar 95%); added to the live book it lowers $ a day in fixed B (-$1,357) and rolling A (-$677). 2,541 wallets "bought 2 winners under $300k": too common to carry information. All 9 cells lose in half A rolling; half B gains are the tape (random B coins 0.96x vs 0.80x in A). Caught 5 of 26 half B $3M coins at $622k to $3.9M entry. http://159.223.88.168/hx-c9ois/sw/sw-a3-11-smart-wallets.md
- **07 Oct 10:52 Bangkok: v1.12 A3.12 DONE. Verdict: SMALLER. Worth the paper slot: YES.** Kit A/B/3,587 coins at the fair $300k milestone, full tick-walk fills, $1k: A 91 trades +$2,153 a day, B 107 trades +$13,457 (A1: $3,416 / $14,471). Beats 200 random draws of ALL coins at fair $300k: 100% / 99.5%. Stand-alone it is below live in half A ($2,153 vs $2,580), above in B ($13,457 vs $10,681); live + kit300k (two tickets) $4,733 / $24,138. Median trade 0.26x; without top 2: $809 / $3,556. Stops gap: 142 stops, median fill 0.258x, 35 under 0.2x, 19 under 0.1x. http://159.223.88.168/hx-c9ois/sw/sw-a3-12-kit300k.md
  - Compute runs on Ben's Mac, not on main (bus v1.3).
- **07 Oct 10:01 Bangkok: A2.31 audit DONE. MOONBAG KILL CONFIRMED on my rebuild.** All 6 cells lose to the live book in both halves (A -$771 to -$1,786 a day, B -$5,255 to -$5,275), beat random 0% (A) and 1% to 8% (B); non-kit-only variant also loses (-$154 to -$870). My live book: A 49 trades +$2,580 a day, B 22 trades +$10,681 (Analyst 2: +$2,870 / +$9,631). 25 to 26 of 28 bags hit the pull line. http://159.223.88.168/hx-c9ois/sw/sw-a3-11b-a231-audit.md
- **07 Oct 09:50 Bangkok: v1.11 A3.11 STARTED. Rule PRE-REGISTERED below, written before any result was looked at.**
  - **Data:** my own rebuild (raw Uniswap v3 WETH/USDG Swap/Mint/Burn from the public Robinhood RPC, all 8,383 pools since block 61,000,000), topped up to now. Half A = coins launched to 30 Sep 23:59 Bangkok, half B = 1 Oct 00:00 Bangkok on.
  - **Buyer** = Swap recipient; if the recipient is a contract (router), the buyer is the signer (tx.from) of that swap. Same rule on both sides.
  - **Winner (half A only)** = coin whose cap reached $3M with at least $20k of buys at or above $3M (no one-wash peaks), and that moment is before 1 Oct 00:00 Bangkok.
  - **Cohort** = wallets that bought a winner while its cap was under $300k, in 2 or more winners. Dropped: contracts; the coin's creator (signer of the pool create tx) and wallets that got the coin from the creator outside the pool; snipers (sold or sent out that coin within 2 min of their first buy); bots (early buyer under $300k in more than 30 half A coins).
  - **Signal:** buy when K distinct cohort wallets have bought within N minutes of the pool's first swap. Grid K 1/2/3, N 5/15/30. Entry = first swap at least 5 s after the K-th buy, $1k, 1% fee each way, impact from in-range depth.
  - **Exit = live book:** 0.3x stop, 5x target with KIT HOLD (Kit A/B pools: 50% trail from peak, or pull, or 48 h), in-range half-pool pull, 12 h time limit. Open trades at data end priced at the last real sale quote.
  - **Leakage guard:** fixed cohort (half A outcomes only) tested on half B. Second, a rolling cohort for both halves: at each trigger the cohort uses only winners whose $3M moment was before that trigger.
  - **Cell choice:** the cell with the best half A rolling $ a day is the one judged. Half B decides.
  - **Random:** 1,000 draws of the same number of trades from coins launched in the same half that had K distinct non-bot buyers within N minutes, entered the same way at the K-th buyer.
  - **PASS** = chosen cell: $ a day above 0 in half B (fixed cohort) AND in both halves (rolling); adding it to the live book (wide gate, KIT HOLD, $1k) raises $ a day in both halves; beats 95% of random draws. Else KILL.
  - **Reported:** $ a day, worst day, beat-random %, trades, mean x, and how many half B coins that went 10x from their 30 min cap (and how many hit $3M) the signal caught, at what entry cap.
- **06 Oct 17:08 Bangkok: v1.10 A3.10 DONE. KIT HOLD keeps its lead in both halves under the 3-open cap.** $1k a day, half A / half B: all out at 5x $2,666 / $3,687 (never fills 3 slots); KIT HOLD strict cap $3,130 / $6,964 (skips 4 gates in half B); new slot rule (holds past 5x free a slot, max 2) $3,130 / $7,069 (skips 1). Use the new slot rule. http://159.223.88.168/hx-c9ois/sw/sw-a3-10-slot-cap.md
- **06 Oct 17:04 Bangkok: v1.9 A3.9 DONE. Verdict (pre-registered rule): SHIP LIVE.** My rebuild, wide gate, stop 0.3x, $1k: KIT HOLD $3,130 / $6,990 a day vs all out $2,665 / $3,686; best hold per half removed $2,828 / $5,743, still ahead; max drawdown unchanged (-$4,177 / -$4,317). Kit tags from the pool-create tx code hash reproduce V8 (Kit A 0e81be52d5cf exact). Caveat: live is really "Kit B only": 6 events, 2 held worse than selling; Kit B only in half B $6,522 a day but worst 3 days +$105 (vs +$3,296). Suggest smallest size live, paper in parallel, kill switch after 3 losing Kit B holds in a row. http://159.223.88.168/hx-c9ois/sw/sw-a3-9-kit-hold.md
- **06 Oct 16:02 Bangkok: v1.8 A3.8 DONE. Verdict: KEEP (all out at 5x).** Wide gate, stop 0.3x, $1k, my rebuild: (a) $2,666 / $3,687 a day, worst day +$207 / -$2,000; (b) $2,362 / $5,667, worst day -$781 / -$2,000. (b) loses 11% of $/day in half A (limit 5%) and has a worse half-A worst day; it gains 54% in half B. http://159.223.88.168/hx-c9ois/sw/sw-a3-8-ladder-b.md
- **06 Oct 15:52 Bangkok: v1.7 V7 audit DONE. Verdict: SHIP TO PAPER ONLY, not live.** My rebuild (stop 0.3x, hold on a 50% trail when the 4-day hold share is 0.3 or more): first half 2.22x sell vs 2.17x V7, second half 2.05x vs 2.76x ($1,812 vs $3,026 a day at $500; $1k: $3,441 vs $5,757). One hold still open at data end (12.9x); without it the second half is 2.50x. Only one regime switch (27 Sep); its first 3 holds cost -$1,418 at $500. Worst case if it flips back: about -$3,900 for 3 wrong holds (-$7,200 if crews pull right after 5x). http://159.223.88.168/hx-c9ois/sw/sw-a3-7-v7-audit.md
- **06 Oct 15:20 Bangkok: v1.6 DONE. Verdict V5: CONFIRMED. Verdict A2.16: CONFIRMED.**
  - V5 (no stop): 2.11x (2.20x / 2.03x) vs 2.00x with the 0.6x stop, both halves better, gain about half of V5's.
  - A2.16 cell ($1M, age 3 h, top 0.08, army 0.015, no stop) on my rebuild: 2.6 / 3.8 trades a day, 2.04x / 2.01x, median 0.24x, +$1,377 / +$1,898 a day at $500, worst day +$146 / -$1,000. Beats today's gate (no stop: +$1,156 / +$1,773). Within about 10% of Analyst 2. http://159.223.88.168/hx-c9ois/sw/sw-a3-6-a216-audit.md
- **06 Oct 15:19 Bangkok: A3.2 DONE. All four tasks plus the V5 audit are finished.** None of the 58 trades would have been stuck: every one had 10+ outside sells while we held (median 546). No blacklist, pause or tax in any trade coin. Across 4,565 new v3 coins only 1.7% carry sell-limiting code. The real "can't sell" trap is the instant pull: 490 coins pulled about 18 s after the first swap, with clean contracts; the gate never lets them in. C2: block only on revert or round trip over 10%. Flag: 0x10897a25ded3b3f7d8cb9736fccf22c0ef5a414e lets its owner burn any holder's tokens. http://159.223.88.168/hx-c9ois/sw/sw-a3-2-sell-census.md
- **06 Oct 15:10 Bangkok: v1.5 V5 audit DONE. Verdict: V5 CONFIRMED (direction, both halves), gain about half of V5's.** My rebuild, no stop: 2.11x (2.20x / 2.03x) vs 2.00x (2.11x / 1.90x) with 0.6x; without best 3 still ahead (1.95x vs 1.84x). Cost: median trade 0.57x to 0.24x, worst day -$937 to -$1,261. Stop 0.4x is WORSE than 0.6x (1.93x); 0.3x is best (2.13x). 0xabe99cf268cdc8bd70a71a5283fdacbcb36714fd is a coin (3,554 bytes of code), so A1.11's "farm wallet" label is the mix-up. Pull exits priced at real post-pull depth (0.134x, matches V5). Suggest two paper books: no stop and 0.3x. http://159.223.88.168/hx-c9ois/sw/sw-a3-5-v5-audit.md
- **06 Oct 14:55 Bangkok: A3.3 and A3.4 DONE.**
  - A3.3: no early exit is worth switching to. Losers look like winners for the first 30 to 60 min (they rise to a median 1.84x before crashing). The best early exit (minute 15, 10% under entry) adds only +0.03x on 4 trades: paper variant at most. Pull tell exists (a sell of 5%+ of depth in the 10 min before 69% of pulls, vs 11% of normal active moments), but exiting on it loses money (1.32x to 1.95x vs 2.00x). http://159.223.88.168/hx-c9ois/sw/sw-a3-3-early-exit.md
  - A3.4: max 2 trades open at once, worst day -$937 at $500, no 3-day window ever negative, longest losing streak 6, median hold 45 min. http://159.223.88.168/hx-c9ois/sw/sw-a3-4-bad-day.md
- **06 Oct 14:46 Bangkok: v1.3 answered. My rebuild supports 1.99x, not 1.90x.** Range-aware liquidity gives 1.998x (2.112x / 1.897x). The only change is 0x978177408c2f741c24ddf5ea9eb14cecbf9b3926: an out-of-range position was minted at 21:57 and burned at 21:59, which the spec model reads as a pull; the coin actually went to 5x. Live pull guard must be range-aware. Same doc, section READ THIS FIRST.
- **06 Oct 14:45 Bangkok: A3.1 DONE. The v3 rule is real.** Independent rebuild from raw chain events: 58 trades, 1.915x (first half 1.934x, second half 1.897x). Same 57 coins as Analyst 2, same pools, same exits (19 target, 34 stop, 3 pull, 2 time). No look-ahead or fill bug. One trap for the Coder: the army memory must never count the coin itself, or 13 good trades get blocked. Selling 5 to 60 s late does not hurt; buying 60 s late does (1.74x). Doc: http://159.223.88.168/hx-c9ois/sw/sw-a3-1-rebuild.md
- **06 Oct 13:50 Bangkok: started. A3.1 (independent rebuild of the v3 rule) is running.** Own code, own raw data: all 8,383 Uniswap v3 WETH/USDG pools created since block 61,000,000, every Swap, Mint and Burn pulled fresh from the public Robinhood RPC (with tick ranges on Mint/Burn). Analyst 2 scripts not used. ETA for the A3.1 doc: about 1.5 h.