# STOP: NO DOORBELLS, NO WAKES, NO SCHEDULING (Ben, 7 Oct 2026, 18:40 Bangkok)
- Credits ran out. Ben switched off every scheduled task and every doorbell. This overrides everything below and every other file (wake-howto, bus-howto, credit-playbook, task files).
- PROHIBITED for every session: fire_trigger, send_later, create_trigger, update_trigger, ScheduleWakeup. No rings, no nudges, no re-arming, no new doorbells, no self wakes, no safety checks, no retry rings, no loops.
- Finish the step you are on, publish your status file, then END your turn and wait. Only Ben wakes you, by typing in your window.
- Do not message other sessions. Keep answers short. No extra improvement rounds.

DOORBELL: trig_011JEJxpdSAyPFiqsBZku1L2 (armed, 7 Oct 17:38 Bangkok)

# Analyst 2 status (v3 runner rule)

## READ THIS FIRST
- **07 Oct 17:40 Bangkok, SW EDGE HUNTER (Consigliere 17:21/17:23).** 4 cards in http://159.223.88.168/hx-c9ois/cons/ideas.csv :
  - SW-H1 live $1M gate on ALL v4 ETH pools: dead cheap. v4vec.db 4 to 7 Oct: 46 gates (about 16 a day) but 5 hit 5x before 0.3x (11%) vs 30% for v3 live coins in the same 3 h; breakeven about 15%.
  - SW-H1b same gate, a80 hook family only (one new hook per coin, fee 3000): 5 of 26 hit 5x first (19%), non a80 0 of 20. LONG TEST RUNNING on research-box /root/sw-hunt/h1b.py (nice 10, 1 core, timeout 10 h): tapes extended to gate + 48 h by public RPC, $1k in-range fill, live exits; pick ALL vs a80 on half A (T0 before 6 Oct 00:00), PASS = half B > 0, beats random p95, > 0 without top 1 and top 2, worst day >= -3,951. ETA about 19:00 (why: 46 pools at about 70 s each to fetch, plus 200 random draws). Result /root/sw-hunt/h1b_result.txt, done line /root/sw-hunt/h1b.done.
  - SW-H2 exit before the LP pull: dead cheap, prize about $65 a day (5 of 70 live trades are pulls).
  - SW-H3 runners vs fakes (A2.40): one feature, outside buyer net $ in the 10 min before the gate; proposed, next cheap check.
  - TRAP: in v4vec.db ETH INTO the pool is the buy (18,157 of 18,169 moves), and is0 = 1 means the token is currency0; my first cheap check had the sign backwards (same answer after the fix).
  - cost: about 0.2 M credits (est.), API $0, 0.4 h
- **07 Oct 17:35 Bangkok, A2.39 DONE: FAIL, keep the holder filters.** Frozen cell D1 (coins blocked by top 0.08 or army 0.015 whose blocking holders sold back at most 10% before the gate), picked on half A: +$70 / -$92 a day vs live; worst day half B -$6,480 (bar -$3,951); beats random 2.5% / 6.5% (bar 95%); without top 2 B -$705. The 75 blocked coins: median 0.28x, 5% reach about 5x (live coins 34%). Door catches 3 of the 13 catchable scorecard misses (0x577bc74a99aabbd43a354668dc7b7ed4b5e303ae 3.56x, 0x56fa3670e19396c4ee8f7775b49ceaede04a84ef 4.68x, 0xb6039ac45d5d355dd98f318dfa622d699e9556e8 0.28x); 3 misses were already live trades, 3 are outside my archive. Next block pricing shows a fake +$3,348 a day in half A from one coin (0x79f03c47d26a8c75bcae351804683c80facf53ac 62.56x vs 3.92x at the crossing swap). Late 50 min vs ETA 16:45 (why: a book bug, forced 5x price empty, found and fixed first). http://159.223.88.168/hx-c9ois/sw/sw-a2-39-missed-runners.md cost: about 0.3 M credits (est.), API $0, 2.7 h wall
- **BOTTLENECK NOW:** none. Edge hunt queue is empty after A2.39; next idea needs the Supervisor. faster if: nothing
- **07 Oct 14:52 Bangkok, A2.39 THE MISSED RUNNERS, PRE-REGISTERED (before any result).** ETA priced answer 16:45 (why: script reuses my A2.38 sim and A2.36 slot book on the 944 pool archive, about 45 min to build plus 30 min compute and checks).
  - **What is new (not A2.26, A2.37, A2.38, A1.15):** those changed WHEN or AT WHAT CAP we buy. This keeps the live $1M gate and timing and opens a door for coins the HOLDER filters block (top over 0.08 or army over 0.015). 9 of the 13 catchable missed runners since 1 Oct were blocked by these filters (army or sell check 6, receiver 3). Mechanism: the concentrated holders are the operator crew; while they sit and do not sell, they are the bid that pushes the coin and retail pays us at $2M+. The filter exists because those holders can dump, so the door asks whether they already started selling.
  - **Universe:** $1M sellable gate crossings, age at most 3 h, coverage ok, in my archive (data to 6 Oct 10:58 Bangkok), that FAIL top 0.08 or army 0.015. Entry at gate + 30 s like live.
  - **Cells (3), pick ONE on half A by (live + door) minus live $ a day, freeze, then price half B:**
    - D0 waive: every blocked crossing.
    - D1 sitting: D0 and the blocking holders (army wallets if army over 0.015, the top receiver if top over 0.08) sent back to the pool at most 10% of the tokens they got from it before the gate block.
    - D2 kit: D0 and the coin is Kit A or B.
  - **Pricing:** $1k, real tick depth for every pool (all $1M gate pools are in my tick set), 1% fee each side, buy at gate + 30 s, exits priced AT the crossing swap (Exec X5 fastexit) as the main number, and at the end of the next block as a check. Live exits: 0.3x stop before 5x, non kit sell all at 5x, kit A/B hold on 50% trail or pull or 48 h, half pool pull, 12 h time. Live book re-priced the same way so the delta is fair. Slot rule 3 normal + 2 holds, door trades share the slots with live.
  - **PASS:** half B delta above 0; combined worst day at or above -$3,951; half B door beats 95% of 200 random draws (each door trade swapped for a random tick set pool not in the book, entered at the SAME minutes after its first swap, same exits); half B delta stays above 0 without the top 1 and without the top 2 door coins.
  - **Also reported:** $2M+ runners (fair peak, $20k bought at $2M+) caught after minute 1 and before halfway, live vs live + door, both halves; which of the 13 catchable scorecard misses since 1 Oct the door catches; overlap with coins FW took.
- **07 Oct 11:24 Bangkok, A2.38 DONE (option c, tape only): FAIL.** Buying 5 s after the 5th fresh-wallet buy (bundle) within 2 min of launch: half A +$4,966 / half B +$8,157 a day on paper, but half B beats random entries only 93% (bar 95%) and without the top 2 coins half B is -$1,398 a day; both are Kit B coins the live gate also bought (0x8e41fd25e85e624560fa50ff9f117a02cd30b8b3 32.15x, 0xc38c332012a9116dcade0c7f43b5d3c71799a9f8 27.18x open). Half A worst day -$6,467 (with live -$8,467). Median trade 0.27x; 18 to 28% reach 5x; entry median $327k, bundle lands in the first seconds of the pool. Bundle wallets sold into us within 30 min in 7 of 403 trades (1.7%, those came back 0.18x). Clean cells ($100k to $50M entry, removes my archive's $100k survivor filter) fail the same way (random 80 to 87%, no top 2 negative). Caveat: 54% of trades priced at real depth are pools that later hit $500k (lean to winners); the rest at a constant product approx lost $86k. Next only if wanted: bundle entry on Kit A/B only with real depth for every pool. http://159.223.88.168/hx-c9ois/sw/sw-a2-38-operator-ride.md
- **07 Oct 11:17 Bangkok, A2.38 PRE-REGISTERED as option (c), tape-only (no pick received by 11:15).** Universe: the 944 Robinhood v3 pools in my archive (token born within 72 h of pool, data to 6 Oct 10:58). A 'bundle buy' = a buy whose wallet's FIRST swap ever (across all 6,213 v3 pools in my tape) is this buy, inside the pool's first 6 h. Event: the 5th bundle buy in a pool, with all 5 inside M minutes of the 1st (M = 2 or 5; the disperse size N cannot be seen on tape, so N is dropped and reported as 'bundle buyers seen in the first 30 min'). Entry: the first swap at least 5 s after the 5th bundle buy, $1k, real tick depth, every A2.21 friction; exits exactly as the live book (0.3x stop before 5x, sell all at 5x, kit A/B hold on 50% trail or pull or 48 h, 12 h time exit, block end next block, 1% fee). Run as a separate book. Pick M on half A by $ a day (one cell), freeze, price on half B. PASS: half B $ a day above 0, combined with live worst day at or above -$3,951, half B beats 95% of 200 random draws (each event swapped for a random other archive pool entered at its first crossing of the same entry cap). Also reported: entry cap distribution, share of trades where any of the 5+ bundle wallets sells within 30 min of our entry, overlap with the live $1M gate tokens, and results without the top 1 and top 2 coins. Not repeated: A1.27's $300k kit + fresh + single sender cell.
- **07 Oct 10:33 Bangkok, A2.38 BLOCKED ON DATA (no result yet, nothing pre-registered).** The event needs every call to the two disperse contracts with all recipients. Sizes (Blockscout counters): 0x84157d3d24630c297849c84d49ae7f1805c1b60f 46,071 txs, about 400 internal transfers an hour (43 calls in the last 1.3 h, median 17 recipients, max 100); 0x03802649e469ed5680c58dff9e65425e22d21cc5 15,109 txs. Neither contract emits a log, so eth_getLogs cannot see them. Routes tried: (1) Coder's Alchemy alchemy_getAssetTransfers: 429 on the first page, I stopped at once (one short burst, live bot unaffected, last log line 10:24); (2) Blockscout API from the container and from the Mac shell: Cloudflare challenge; (3) Blockscout API inside the browser pane: works but about 10 pages of 50 a minute, so about 7 h for the 23 day window; parallel pages get throttled (19 errors in 90 s). Options for you: (a) let me run the slow browser crawl in the background (about 7 h, done about 17:30), (b) ask PP or Coder for a disperse table (block, tx, sender, recipients) if they index it, or (c) run a tape-only version now: 'bundle' = a wallet whose FIRST swap ever is this buy, entry 5 s after the 5th such buy in a new coin; N becomes the count of fresh buyers seen so far, so the 'funding seen before the buys' part is lost, but the entry point is the same (after the 5th bundle buy). I can pre-register and deliver (c) in about 40 min. Waiting for your pick; A2.30 watch continues. Note: the droplet did not answer on port 80 from about 10:35 to 10:53 Bangkok (503, then no answer); back at 10:54. If no pick by 11:15, I start (c).
- **07 Oct 10:19 Bangkok, A2.37 DONE: late bloomer door KILLED.** 20 tokens first crossed $1M sellable more than 3 h after launch (of 165 $1M crossings in 944 pools); live exits at $1k, real depth: mean 0.61x, half A -$293 / half B -$456 a day, 0 of 20 hit 5x (best peak 2.91x), beats random 32%. No half A cut is even positive in sample (best: age 36.9 h+ -$66 a day on 8; pair -$5 on 5); frozen rule takes 0 half B trades. Only 1 of 15 real ones reached $3M+ (0xd0ca71118ca21d674f9018a53026a38a0ab54dbf, crossed at $5.0M, trade 0.40x); 5 rows are odd-supply caps, without them still 0.62x / 0.45x. Late $1M = slow buyers = no 5x fuel; the 3 h age cap is doing real work (fits A2.25). Re-run on PP e1-runners.csv when it lands (404 now). http://159.223.88.168/hx-c9ois/sw/sw-a2-37-late-bloomers.md
- **07 Oct 10:15 Bangkok, A2.37 PRE-REGISTERED (before any result): late bloomer door.** PP e1-runners.csv is not published yet (404), so I use my own archive: the 944 Robinhood v3 pools in the s16 set (tokens whose pool opened within 72 h of birth, data to 6 Oct 10:58). Event = a token's FIRST $1M sellable crossing (cap at or over $1M with $20k of buys at that level, same rule as the live gate) that comes MORE than 3 h after its first swap, so it could never pass our live age gate. The book takes these events, not just the ones that later hit $3M (no look ahead); I report how many later reached $3M+. Features measured AT that crossing only: age h, top receiver share, army share (strict, same code as the live gate), net buy $ last 30 min, unique buyers last 30 min, kit A/B tag (code hash, same as live). Rule search on HALF A ONLY (crossings to 30 Sep): every single feature cut at its half A 25/50/75% quantile, both directions, plus the best pair of two single cuts (AND); pick the cell with the highest half A $ a day with at least 6 half A trades. That one rule is then frozen and priced on half B. Pricing: live exits (entry at crossing + 30 s, 0.3x stop before 5x, sell all at 5x, kit A/B hold on 50% trail or pull or 48 h, 12 h time exit before 5x), $1k, exits at the end of the next block, real tick depth where my raw tick set covers the pool (count reported), else cap path with the live fee; run as a SECOND book next to live (no shared slots). PASS: half B $ a day above 0 with the frozen rule, combined live + second book worst day at or above -$3,951, and half B beats 95% of 200 random draws of the same number of late crossings in half B. Also shown: the plain 'take every late crossing' book, and results without the top 1 and top 2 coins.
- **07 Oct 10:04 Bangkok, A2.36 DONE: FAIL by the rule (one coin, one half), but the extra hold slot never cost anything. Suggest hold slots 2 to 3 as a risk call.** The 3 normal slots never bind (0 gates skipped in 23 days; caps 4 and 5 = live). Hold slots 3, 4 or none change only 0x9ba4d84ac1571a0788b2bb7a6412e73529d12dc6 (Kit B): force-sold at 5x in live (4.69x) vs 10.11x on the trail = +$5,421, half B +$903 a day, half A $0, $0 without that coin. It was forced on 5 Oct 00:09 because 0xee049979960d5e42b389b941ec45790ca7b6e1f7 and 0x8e41fd25e85e624560fa50ff9f117a02cd30b8b3 held both slots. Cost: max $ at risk $5,000 to $6,000 (5 to 6 tickets), worst day unchanged, inside the 8 ticket wallet. http://159.223.88.168/hx-c9ois/sw/sw-a2-36-hold-slots.md
- **07 Oct 10:03 Bangkok, A2.36 PRE-REGISTERED (before any result).** Same 70 wide trades (E70; the task says 69, the archive holds 70; no newer trades have tick data), live sim and every A2.21 friction, $1k. Only the slot rule changes: hold slots (kit coins past 5x) 2 (live) / 3 / 4 / no cap x open normal trades 3 (live) / 4 / 5 = 12 cells. A kit coin hitting 5x with no free hold slot sells at 5x (as live); a gate with the normal book full is skipped (as live). PASS (each cell): $ a day above live in BOTH halves (A to 30 Sep, B from 1 Oct), worst day no worse than live minus $1k, and the gain stays positive in both halves without the top 1 and top 2 coins. No random draw: a capacity change picks no coins, so the top-coin test replaces it. Reported per cell: $ a day, worst day, max $ at risk at once (open tickets x $1k cost), gates skipped, kit coins force-sold at 5x, which coins change.
- **07 Oct 10:01 Bangkok, A2.35 DONE: FAIL all 4 cells. Kit hold is tuned, stop; keep the live 50% trail.** $ a day vs live (A +$2,870 / B +$9,631): trail 40% +$461 / -$1,093; trail 30% +$205 / -$3,365; stepped 50/35 (0.65 x peak once peak hits 10x) +$476 / -$51, beats random 94.5% / 46% (near miss); stepped 50/25 +$802 / -$451. Half B loses every time because tighter trails sell the top runner 0xee049979960d5e42b389b941ec45790ca7b6e1f7 at 5.73x to 12.91x (live 15.78x). With A2.34: looser loses both halves, tighter wins only half A, so 50% is the best single setting. Re-test stepped 50/35 at 25+ kit 5x coins. Moving to A2.30 support. http://159.223.88.168/hx-c9ois/sw/sw-a2-35-trail-shape.md
- **07 Oct 09:59 Bangkok, A2.35 PRE-REGISTERED (before any result).** Same 15 kit coins and same frame as A2.34 (live entry, 0.3x stop, 5x trigger, slot rule, $1k, block-end exits, real depth, 1% fee, 48 h cap, pull exit kept). Cells, only the kit trail changes: trail 40% (sell when cap is at or under 0.6 x peak since 5x), trail 30% (0.7 x peak), stepped 50/35 (0.5 x peak until the peak reaches 10x our entry cap, then 0.65 x peak), stepped 50/25 (0.5 then 0.75 x peak). PASS (each cell): $ a day above live in BOTH halves, worst day no worse than live minus $1k, beat 95% of 200 random draws (same coins, hold exit at a uniform random time between the 5x block and the 48 h horizon or data end). Shown with and without the top 1 and top 2 coins. If none pass: 'kit hold is tuned, stop'.
- **07 Oct 09:57 Bangkok, A2.34 DONE: KILL, keep the live kit hold (48 h cap, 50% trail).** The 48 h cap sold 0 of 15 kit coins: the 50% trail sold 13 (median 2.8 h after 5x, longest 43.3 h after entry: 0xf15aab29c3d22cf2b469d4a9b0c0a72f31ba0711 at 11.69x), 2 still open at data end. So 4 d / 7 d / no cap equal live (+$0 both halves). Wider trails lose in both halves under every cap: 60% A -$444, B -$739 to -$2,178 a day; 70% A -$1,050 to -$1,130, B -$1,516 to -$2,955; same without top 1 and 2. Runners still climbing at the 48 h sale: 0. Watch CHONK 0x717d38587504b79e2e888c860bd1569c0d5984af (paper kit hold peak 23.28x, 14.16x at 09:56, cap sells it 8 Oct 23:52): first live test of the cap. http://159.223.88.168/hx-c9ois/sw/sw-a2-34-kit-longer.md
- **07 Oct 09:55 Bangkok, A2.34 PRE-REGISTERED (before any result).** Clock note: the server clock reads 09:55 now, so my stamps from 09:55 to 11:15 below do not match it; from here I stamp the server clock. Sample: the kit A/B coins among the 70 wide trades (E70) that hit 5x under the live sim (15 coins). New kit coins since the archive ends (6 Oct 10:58 Bangkok, e.g. CHONK 0x717d38587504b79e2e888c860bd1569c0d5984af) have no tick data here: listed as a note, not scored. Only the kit hold leg changes; entry, 0.3x stop, 5x trigger, non-kit exits, slot rule (3 normal + 2 holds), $1k, block-end exits, real depth, 1% fee stay live. Cells: time cap 48 h (live) / 4 d / 7 d / none (cap counted from entry, as live) x trail 50% (live: sell when cap is at or under 0.5 x peak since 5x) / 60% (0.4 x peak) / 70% (0.3 x peak); pull exit (quote reserve at or under 0.5 x entry reserve) kept in all cells. A hold still open when its data ends is sold at a real-sale quote at the last row; count reported per cell. PASS (each cell): $ a day above live in BOTH halves (A gates to 30 Sep, B from 1 Oct), worst day no worse than live minus $1k, and beat 95% of 200 random draws (same coins, hold exit at a uniform random time between the 5x block and that cell's horizon or data end). Shown with and without the top 1 and top 2 coins. 'Still climbing at the 48 h sale' = cap at the 48 h sale within 10% of its peak since 5x; for those I report the x at the sale and the later peak.
- **07 Oct 11:15 Bangkok, A2.33 DONE: TOO FEW TO JUDGE (5 runners, 3 faders; bar 6 each) and no cut helps. Keep holding ALL kit coins past 5x.** 15 kit coins hit 5x; every one-feature cut (buyers 1 h, net buy 1 h, buyers since launch, fast to 5x, depth, kit A, kit B) loses to hold-all in both halves (-$95 to -$3,824 a day), best random beat 91%. Holding beat selling at 5x on 11 of 15 coins: misses cost $1,472 to $3,965 each, hits paid up to $11,289; hold minus sell +$41,935 total. Faders and runners overlap on every feature. Re-test at 20+ kit 5x coins (about a week). http://159.223.88.168/hx-c9ois/sw/sw-a2-33-kit-runners.md
- **07 Oct 10:50 Bangkok, A2.33 PRE-REGISTERED (before any result):** sample = kit A/B coins among the 70 wide trades that hit 5x under the live sim. Outcome label = peak cap multiple (vs our entry) within the kit hold window: RUNNER if 20x or more, FADER if under 10x. Features measured AT the 5x trigger block only: unique buyers last 60 min, net buy $ last 60 min, unique buyers since launch (holder proxy; true holder count is not in our data), minutes gate to 5x, kit A vs B, pool quote reserve $. Rule family: one feature, one cut at the sample median, direction fixed by the obvious prior (more buyers / more net buy / more depth / faster = hold; kit A vs B tested both ways and reported as two cells). Cell = "kit hold when X, else sell at 5x", priced at $1k with every live friction and the slot rule. PASS needs $ a day above live in BOTH halves (A to 30 Sep, B from 1 Oct), worst day no worse than live minus $1k, and beat 95% of 200 random splits of the same size. If fewer than 6 RUNNER or 6 FADER coins, report counts and call it too few to judge.
- **07 Oct 10:35 Bangkok, A2.31 DONE: MOONBAG KILLED (all 6 cells), A2.32 second ticket at $5M FAIL (near miss), $10M FAIL.** Bag 10/20/33% x 3 d/7 d loses to live in both halves (A -$905 to -$2,002 a day, B -$4,378 to -$4,634), beats random exit 0% / 2 to 11%; 25 of 28 bags hit the pull line a median 5.8 h after 5x and came back at about 0.36x; non-kit-only variant also loses (-$283 / -$144 at 10%). $5M add-on: A +$17 a day (beats random 93%, bar 95%), B +$1,977 (97%), without top 2 B +$962; all 9 add-ons 22 to 24 Sep lost, 12 of 14 from 25 Sep won. Suggest a PAPER add-on book, not live. Live book on new halves: A +$2,870, B +$9,631 a day. http://159.223.88.168/hx-c9ois/sw/sw-a2-31-moonbag.md
- **07 Oct 09:55 Bangkok, A2.31 + A2.32 PRE-REGISTERED (before any result is looked at).** Universe: the 70 wide live-gate trades (E70), live entry (gate + 30 s), live frictions (0.3x stop before 5x, exits at the END of the next block, real tick depth, 1% fee each way), $1k ticket, slot rule (max 3 open; a bag or add-on takes no slot). Halves per v1.26: A = gates to 30 Sep, B = gates 1 Oct on. Data ends 6 Oct 10:58 Bangkok, tracking covers 7 days after gate, so 14 d is not testable; a bag still open at data end is marked at a real-sale quote at the last row (count reported).
  - **A2.31 MOONBAG.** At the 5x trigger sell (1 minus bag) at the next block end; keep the bag with NO price stop; bag exits at the first of: liquidity pull (live definition: pool quote reserve at or under 50% of the entry reserve) or 3 d / 7 d after gate. Cells: bag 10%, 20%, 33% x horizon 3 d, 7 d, applied to ALL coins that hit 5x (kit trail replaced). Variant reported, not judged: non-kit only (kit keeps the live 50% trail).
  - **A2.32 SECOND TICKET.** For each traded coin, first moment within 7 d after gate where cap crosses $5M (cell 1) or $10M (cell 2) AND net buy $ in the last 60 min is above 0: buy one more $1k at that moment + 30 s; exit on 50% trail from the add-on peak, pull (reserve at or under 50% of the add-on entry reserve), or 7 d / data end. Added on top of the live book.
  - **PASS (each cell):** $ a day beats the live book in BOTH halves, worst day no worse than the live book worst day minus $1k, AND beats 95% of 200 random draws (A2.31: same bags, bag exit at a uniform random time between the 5x sale and the horizon; A2.32: same number of add-ons in the same coins at a uniform random time in the 7 d window, same exit rule). Every result shown with and without the top 1 and top 2 coins. A clean KILL is written up the same way.
- **07 Oct 08:02 Bangkok, A2.30: live trade 1 CLOSED (CrawlScan stop). Live 0.195x vs paper twin 0.229x = exit 14.8% WORSE (fails the 10% fill rule on this trade).** Sell tx 0x9ad953b36c2c00c3d89505cc0e42160fe8156b4ec419ad3b671077d220adcf79 landed 07:51:33, 4 s after the paper block, inside a liquidity pull; paper books that sold at 07:51:33 also got 0.19x. Entry was 0.96% better. Gas $0.0138 round trip. Ask Coder: faster stop sell, and price the stop twin at the live sell block. 1 of 10. http://159.223.88.168/hx-c9ois/sw/sw-a2-30-live-vs-paper.md
- **07 Oct 06:05 Bangkok, A2.30: FIRST REAL LIVE BUY FILLED (CrawlScan 0x1ec25a196c49dd3445f1fcc53c37c513af92839c, Kit B).** Tx 0xf4921256fed26e57b8df0890a006b8b83adbf84000d06bdd587bb0899f06df1f success; signing path proven. Fill 20.9 s after gate (paper 19.4 s); live entry 0.96% BETTER than paper (paper $500 moves the pool +1.0%, live $10 does not). Gas $0.0075. Trade open; paper twin 0.79x at 06:03. 1 of 10. http://159.223.88.168/hx-c9ois/sw/sw-a2-30-live-vs-paper.md
- **07 Oct 01:10 Bangkok, A2.30:** Coder ran `DRY SIGN PASS` at 00:13 (signed, not sent), so the checksum fix holds. 0 real live trades yet; no new gate since CHONK. CHONK kit hold still riding. http://159.223.88.168/hx-c9ois/sw/sw-a2-30-live-vs-paper.md
- **07 Oct 00:37 Bangkok, A2.30: the missed CHONK was a WINNER.** Paper w2_v3_wide_s03 sold CHONK 0x717d38587504b79e2e888c860bd1569c0d5984af at the 5x target at 00:22:52 (cap 5.05x, sim 4.78x), 30 min after gate. The bug cost about $37.8 at $10 (about $3,776 at $1k). Kit hold twin rides on (0% sold, Kit B). Coder tested the kill switch with a fake pool at 00:07 (it tripped, LIVE ON again 00:13). Still 0 real live transactions: one tiny real swap needed now. http://159.223.88.168/hx-c9ois/sw/sw-a2-30-live-vs-paper.md
- **07 Oct 00:04 Bangkok, A2.30: first live signal MISSED (CHONK 0x717d38587504b79e2e888c860bd1569c0d5984af, Kit B, gate 23:52:46).** Live buy threw TypeError: router 0xcaf681a66d020601342297493863e78c959e5cb2 not checksummed; no ETH spent. Coder already patched live.py (to_checksum_address, 00:00) and LIVE ON at 00:01:34, but the signing path is unproven until one real buy fills: ask for one tiny real swap now. Paper twin open. http://159.223.88.168/hx-c9ois/sw/sw-a2-30-live-vs-paper.md
- **06 Oct 19:49 Bangkok, A2.30 STARTED (live vs paper): 0 live trades so far.** Live on since 19:25 at $10; no coin has passed the wide gate since CLAUS 13:51. Feed alive (swaps to the minute); 2 coins crossed $1M sellable since, both failed army (0x206211d456a3a9dbe45e13507aac5131ace9c716 9.46%, 0x4fdcc31f5514644bf512b296057f025ddd4f9308 29.4%). 0 gates in 6 h happens 40 to 50% of the time; I flag a possible fault if none by 7 Oct 07:00. Table updates on every bus check; ring after live trade 1, 3, 10. http://159.223.88.168/hx-c9ois/sw/sw-a2-30-live-vs-paper.md
- **06 Oct 19:46 Bangkok, A2.29 DONE (kit hold on operator flag): FAIL both cells, keep hold on the code hash.** $1k vs live +$3,317 / +$6,683: hash OR operator +$2,591 / +$7,085 (-$726 / +$402; without top 2 -$726 / $0; beats random 35% / 88%, needed 95%); operator only +$2,432 / +$4,830 (-$885 / -$1,853). Why: a 50% trail only beats selling at 5x if the coin later reaches 10x; all 6 flagged non-kit 5x coins in half A faded to 2.45x to 3.56x (-$10,150 vs selling at 5x), only 2 in half B ran on (+$3,620). Flag rebuilt walk forward from archive balances (my 165-gate set): non-kit 18 of 34 wide trades flagged (A1.21: 11 on 114 gates). Operator flag = ENTRY signal, not a HOLD signal. Note for Coder: I read its Alchemy endpoint (read only, about 3,000 light calls); a short burst got 429 for a few minutes, normal right after. http://159.223.88.168/hx-c9ois/sw/sw-a2-29-operator-hold.md
- **06 Oct 18:31 Bangkok, A2.28 DONE (size v2, live exit): smallest ticket for $2,000 a day in both halves = $1k (+$3,317 / +$6,683).** All sizes count. $ a day: $500 +1,745 / +3,511; $2k +6,002 / +12,138; $3k +8,153 / +16,577; $5k +11,156 / +23,039. $ a day per $ of wallet (8x) is highest at the smallest ticket and falls every step (0.44 / 0.88 at $500 to 0.28 / 0.58 at $5k): no hump, so pick size by drawdown (about 4.2 to 4.7 tickets at every size). Own buy moves the pool over 10% in 0% of trades at $2k, 89% / 79% at $3k, 100% at $5k; x lost to the pool vs paper 0.19 / 0.30 at $1k, 0.61 / 0.89 at $5k. Wallet: 5x ticket never skipped a trade. Ramp pick: $1k, $2k after live fills match, $3k top. http://159.223.88.168/hx-c9ois/sw/sw-a2-28-size-v2.md
- **06 Oct 18:27 Bangkok, A2.27 DONE (non-kit exit): KEEP LIVE.** Every target change loses ($ a day at $1k vs live +$3,317 / +$6,683): 3x -$947 / -$611, 4x -$761 / -$301, 7x -$109 / +$321, 10x -$1,276 / +$913 (worst day -$4,424), 80% at 5x + 20% runner -$213 / +$48, half at 3x + rest at 5x -$423 / -$285. Stop 0.25x -$38 / -$12. Stop 0.4x passes the rule by a hair (+$81 / +$27, 11 stops get a better price, no non-kit winner lost) but do NOT use it: two kit winners dipped to 0.308x (0x9ba4d84ac1571a0788b2bb7a6412e73529d12dc6) and 0.357x (0xe99c8437b0b29325a4080b8c837ff4e5a6f9fc69) before 5x (about +$12k at $1k), vs $1,380 saved in 23 days; 13 non-kit winners is too few to rule the band out. http://159.223.88.168/hx-c9ois/sw/sw-a2-27-nonkit-exit.md
- **06 Oct 18:22 Bangkok, A2.26 DONE (earlier entry without the losers): CLOSE, nothing splits.** 0 of 25 pre registered tries pass (kit rule, 11 features cut at the median, both ways). Features beat live in both halves 13 of 25 times = 52%, and a RANDOM half of the coins does it 52% of the time (200 draws); for buyers in 15 min and pool depth BOTH sides of the cut beat live, so it is noise. 0 of 25 beat plain $750k in both halves. Kit A or B at $750k: +$3,601 / +$7,137 a day (live +$3,317 / +$6,683, plain $750k +$3,973 / +$7,177), max drawdown half B -$7,388. Keep live at $1M. http://159.223.88.168/hx-c9ois/sw/sw-a2-26-early-entry.md
- **06 Oct 18:17 Bangkok, A2.25 DONE (more trades): only sellable $750k passes, and it is weak. Keep live as is; close gate knobs.** At $1k, $750k gives +$3,973 / +$7,177 a day vs live +$3,317 / +$6,683, worst day -$3,713 (floor -$3,951), but max drawdown in half B -$4,700 to -$7,041. The 22 extra coins LOSE (-$280 / -$518 a day); all the gain is the SAME coins bought earlier, and without the top 2 coins per half that gain is only +$302 / +$113 a day. $500k, age 6 h or 12 h, top 0.10 or 0.12, army 0.02 or 0.03 all fail. No pair beats $750k alone. Suggest $750k as a tagged paper variant only. A2.11: still only CLAUS in w2.db. http://159.223.88.168/hx-c9ois/sw/sw-a2-25-more-trades.md
- **06 Oct 17:43 Bangkok, A2.24 DONE (money line v2, live kit hold): at $1k +$3,317 / +$6,683 a day vs +$2,733 / +$3,535 all-out (+$584 / +$3,148).** Worst day and max drawdown unchanged; peak open 5 tickets. Second half without the 2 still-open holds' gain over 5x: +$5,446. Wallet: history needed 5 x ticket (8 x is fine); max drawdown 5.7 tickets so the 10-ticket kill never fired; a guarantee that the kill always stops buying before cash runs out needs 15 x ticket. http://159.223.88.168/hx-c9ois/sw/sw-a2-24-honest-money-v2.md
- **06 Oct 17:35 Bangkok, A2.23 step 1 DONE: Base v4 CLOSED.** 107,947 v4 pools (39,987 ETH-quoted), 18 sellable $1M gates in 3 h (bar 30); 10 are real new coins and their median army share is 0.72 (0 of 10 at or under 0.015; Robinhood v3 about 0.001). 7,431 pools had $20k+ buys in 3 h but only 157 passed $500k: bots trading with bots. http://159.223.88.168/hx-c9ois/sw/sw-a2-23-base-v4.md
- **06 Oct 17:10 Bangkok, A2.22 DONE (second chain): CLOSE both.** Uniswap v3 WETH, 13 Sep to 6 Oct, strict sellable $1M gate within 3 h: Base 1 gate (1,721 new WETH pools, 18 with $20k+ buys in 3 h), Ethereum 6 gates (355 pools; caps look like odd-supply tokens). Bar was 30. New memecoins on those chains launch on v4 launchpads, not v3. Base v4 untested (offer as A2.23). http://159.223.88.168/hx-c9ois/sw/sw-a2-22-second-chain.md
- **06 Oct 16:10 Bangkok, A2.21 DONE (honest money line): at $1k it clears $2,000 a day in both halves: +$2,733 / +$3,535.** All frictions in: wide gate, buy at gate + 30 s, 0.3x stop, every exit sold at the end of the NEXT block (cascades in), real depth. $500 +1,440 / +1,862; $2k +4,921 / +6,368; $3k +6,647 / +8,597. At $1k: worst day 0 / -3,161, worst 3 days 0 / +2,684, max drawdown -4,257 / -4,700. 0xa2508b15ab16826c720631b9afa0a90c453ae8a4: a 1% entry-cap difference on a knife edge (my low 0.605x vs Coder 0.593x before the crash); irrelevant under the 0.3x stop. http://159.223.88.168/hx-c9ois/sw/sw-a2-21-honest-money.md
- **06 Oct 16:04 Bangkok, A2.20 DONE (fill delay): suggested check 3 = fill within 30 s.** Wide gate, 0.3x stop, block-end, $1k: every delay 2 s to 300 s keeps $ a day at or above $2,000 in both halves (6 s: 2,721 / 4,091; 30 s: 2,743 / 3,616; 60 s: 2,753 / 3,617; 120 s: 2,686 / 2,617). Up to 45 s the swings are one knife-edge coin (0x6c963ce1aff02c18b2d0ef81b96df77b2f53c4ee, 0.14x or 4.65x by the second); real leak starts at 60 s (second half without that coin 2.01x to 1.88x at 60 s, 1.69x at 120 s). http://159.223.88.168/hx-c9ois/sw/sw-a2-20-fill-delay.md
- **06 Oct 16:01 Bangkok, A2.19 DONE (cascade): no warning works; cascades are a fixed cost.** 9 live-visible warnings priced on all 70 wide trades at $1k: none raises $ a day in both halves (best: -$49 / +$53). Under the 0.3x stop, block-end pricing costs +$12 / -$57 a day (2.06x / 2.09x at $1k). 4 of 33 stops cascade (adds 0xfe0f309cee96912d9926b8f0786fe3347364e678), all in the second half. 3 are fresh-wallet farms (119 to 208 single-use wallets never seen in any other coin, one sell each in the same 0.1 s via SwapRouter02); 1 is a single gatherer wallet. Gate features do not separate them from winners. Only lead left: common ETH funder (needs address history). Ask: 0xa2508b15ab16826c720631b9afa0a90c453ae8a4, one swap takes it from 1.91x to 0.028x in my data. http://159.223.88.168/hx-c9ois/sw/sw-a2-19-cascade.md
- **06 Oct 15:30 Bangkok, v1.13 quick task DONE:** wide cell per-trade list published for V6: http://159.223.88.168/hx-c9ois/sw/sw-a2-16-wide-trades.csv (70 trades: token, pool, gate time Bangkok, gate block, half, entry and exit time, exit, x at $500 with 0.3x stop; 2.11x / 2.16x). Virality Expert rung. A2.11 keeps running (no new Coder trade since CLAUS).
- **06 Oct 15:22 Bangkok, A2.18 DONE (size curve, wide gate + 0.3x stop, real tick depth):** SIZE is the big knob: $500 to $3k multiplies $ a day by 4.6 / 4.7; gate + exit changes together add about 40%. $ a day at $500 +1,432 / +2,183; $1k +2,718 / +4,156; $2k +4,901 / +7,539; $3k +6,627 / +10,258; $5k +8,952 / +14,051. Tops out at $7.5k to $10k a trade (+10.2k / +16.6k a day); safe ceiling about $5k (worst 3 days stays positive in the second half). Max 3 trades open at once: wallet = 3 x size. Only 1 losing day in 23 (28 Sep, 2 pulls = 2 x stake). Re-entry under 0.3x stop LOSES in half A: switch it off. Ramp: $500 to $1k to $2k to $3k, $5k only if live fills match paper. A2.11: no new Coder trade since CLAUS. http://159.223.88.168/hx-c9ois/sw/sw-a2-18-size-curve.md
- **06 Oct 15:20 Bangkok, A2.17 DONE (re-entry):** with no-stop exits a second entry adds nothing (new high 0.23x / 1.38x; drop then new $1M gate: 3 trades, all pulled). With the 0.6x stop, 'new $1M gate after 50% drop' pays 1.56x / 2.67x (8 / 4 trades, +$159 / +$372 a day); A2.9 re-entry stays the better version. If the book goes no-stop, drop re-entry. http://159.223.88.168/hx-c9ois/sw/sw-a2-17-reentry.md
- **06 Oct 15:10 Bangkok, A2.16 DONE (gate grid, $ per day, no-stop exits):** best robust change = keep $1M, widen age to 3 h and army to 0.015: trades/day 2.6 / 3.8 (was 1.9 / 3.4), 2.09x / 2.14x, $/day at $500 +1,406 / +2,152 (was +1,158 / +2,027). $750k gate (3 h, top 0.10, army 0.015) gives 3.0 / 5.8 trades a day and $1,589 / $2,171 but 1.75x in half B: own paper book first. Pure walk forward picks $500k and fails (half B 1.18x, $807/day): never $500k. Flat plateau, no knife edge. http://159.223.88.168/hx-c9ois/sw/sw-a2-16-gate-grid.md . A2.17 re-entry next.
- **06 Oct 15:05 Bangkok, A2.11:** Coder's entry cap fix confirmed on CLAUS (entry_cap $956,288 at fill, result 0.566x vs my 0.548x). No new paper trades since 13:51. Replay of the 58 still 404.
- **06 Oct 15:00 Bangkok, A2.15 DONE: no new venue passes, 0 trades added.** Main v3 age 3 to 6 h: 0.74x (2). Pons graduates with filters: 0.56x (3). Plain v4 no-hook (274 sellable $1M gates): blind age ≤ 1.54 h 0.50x / 0.59x (91), still under 1x with every pull removed. Hook 0x4e3468951d49f2eea976ed0d6e75ffcb44a9a544: 1 gate in 49,040 pools. Rule stays v3 WETH only. http://159.223.88.168/hx-c9ois/sw/sw-a2-15-more-venues.md
- **06 Oct 14:20 Bangkok, A2.11 daily line:** 1 trade (CLAUS), fill delay 20.6 s, price −22% gate to fill, delay cost −0.01x (0.548x at the real fill vs 0.558x at +6 s). CLAUS final from the real fill: stop 14:02:03, 0.548x. http://159.223.88.168/hx-c9ois/sw/sw-a2-11-live-check.md
- **06 Oct 14:05 Bangkok, A2.11 started (first trade CLAUS):** gate, age and top share match the Coder to the second. Two entry problems: (1) BUG, entry_cap is stored at the gate cap, not the cap at fill, so target and stop are set from the wrong base (CLAUS stopped at 13:55 at $676k; from the real fill of $898k it would still be open); (2) fill came 20.6 s after the gate instead of 5 s (feed lag; price fell 22% in that time). Coder's replay of the 58 still 404. http://159.223.88.168/hx-c9ois/sw/sw-a2-11-live-check.md
- **06 Oct 13:55 Bangkok, A2.14 DONE:** real tick depth gives the same fills as the R model within 0.02x up to $5k (real ETH at the gate = 95% of virtual). 1.5x is crossed at about $6.8k (second half about $5.8k). Bug: the R model pull guard counted an out-of-range mint/burn as a pull; fixing it turns 0x978177408c2f741c24ddf5ea9eb14cecbf9b3926 from 0x into a 5x, so the main book is 1.99x at $500 (2.09x / 1.90x). STONKROBBERS 7.6% = price 0.7% from its launch-range end. http://159.223.88.168/hx-c9ois/sw/sw-a2-14-real-depth.md
- 06 Oct 13:35 Bangkok, A2.13 done (accepted in v1.7).

## Doorbell
- trig_014fENmVKwUmQUumuuMu3YkT (Doorbell: Analyst 2). Safety check every 30 min.

## Asks for Supervisor
- **For PP (7 Oct 13:00 Bangkok, re-check 11): liqbuild on research-box is DEAD, killed by the out-of-memory killer at 11:53 Bangkok (7 Oct) after 23 h 3 min, memory peak 14.6 GB.** It never finished: last log line still 13:21 Bangkok on 6 Oct ("1875 s owner_at for pulls"), liqlock.db (21 GB) last written 13:22 Bangkok on 6 Oct, no LIQBUILD_EXIT line. Ask: rerun the owner_at step in chunks (per day or per 1,000 pools) so it fits in memory and writes progress; the hold-or-sell cross-check (A2 old task) waits on it. I stop re-checking until PP says it is rerun.
- Coder fix: entry_cap = cap at the fill event (not the gate), fill at the first event at or after gate + 5 s.
- (done in v1.7) in-range liquidity for R.

## Needs Ben
- Nothing.

| Task | Status | Doc |
|---|---|---|
| A2.30 live vs paper | RUNNING: 0 live trades yet | http://159.223.88.168/hx-c9ois/sw/sw-a2-30-live-vs-paper.md |
| A2.29 operator hold | DONE: fail, keep hash kit hold | http://159.223.88.168/hx-c9ois/sw/sw-a2-29-operator-hold.md |
| A2.28 size v2 | DONE: $1k smallest for $2k a day; efficiency falls with size; $3k top | http://159.223.88.168/hx-c9ois/sw/sw-a2-28-size-v2.md |
| A2.27 non-kit exit | DONE: keep live (0.4x stop passes by a hair, do not use) | http://159.223.88.168/hx-c9ois/sw/sw-a2-27-nonkit-exit.md |
| A2.26 earlier entry split | DONE: nothing splits (features = random), closed | http://159.223.88.168/hx-c9ois/sw/sw-a2-26-early-entry.md |
| A2.25 more trades (gate knobs) | DONE: only $750k passes, thin; keep live | http://159.223.88.168/hx-c9ois/sw/sw-a2-25-more-trades.md |
| A2.24 money line v2 (kit hold) | DONE: $1k +$3,317 / +$6,683 a day | http://159.223.88.168/hx-c9ois/sw/sw-a2-24-honest-money-v2.md |
| A2.23 Base v4 buyers | DONE: 18 gates, army median 0.72, closed | http://159.223.88.168/hx-c9ois/sw/sw-a2-23-base-v4.md |
| A2.22 second chain | DONE: Base 1 and Ethereum 6 gates in 23 days, closed | http://159.223.88.168/hx-c9ois/sw/sw-a2-22-second-chain.md |
| A2.21 honest money line | DONE: $1k clears $2,000 a day both halves | http://159.223.88.168/hx-c9ois/sw/sw-a2-21-honest-money.md |
| A2.20 fill delay | DONE: 30 s is safe; leak from 90 s | http://159.223.88.168/hx-c9ois/sw/sw-a2-20-fill-delay.md |
| A2.19 cascade warning | DONE: none works, fixed cost about -$57 a day at $1k (second half) | http://159.223.88.168/hx-c9ois/sw/sw-a2-19-cascade.md |
| A2.18 size curve (new config) | DONE: safe ceiling ~$5k, peak $ a day at $7.5k to $10k, 3 open max | http://159.223.88.168/hx-c9ois/sw/sw-a2-18-size-curve.md |
| A2.17 re-entry | DONE: nothing to add under no stop; small gain only with the 0.6x stop | http://159.223.88.168/hx-c9ois/sw/sw-a2-17-reentry.md |
| A2.16 gate grid by $ a day | DONE: $1M + age 3 h + army 0.015 beats today's rule in both halves; $750k book as paper; never $500k | http://159.223.88.168/hx-c9ois/sw/sw-a2-16-gate-grid.md |
| A2.15 more venues | ACCEPTED (v1.11): all 4 venues lose in both halves, 0 trades added | http://159.223.88.168/hx-c9ois/sw/sw-a2-15-more-venues.md |
| A2.14 size table on real depth | ACCEPTED (v1.8): same fills within 0.02x; 1.5x at ~$6.8k; false pull fixed, main 1.99x | http://159.223.88.168/hx-c9ois/sw/sw-a2-14-real-depth.md |
| A2.13 confirmed stop | ACCEPTED (v1.7): keep the instant stop + re-entry (+$31.5k, both halves ~1.9x, losers not bigger). 2 min confirm alone pays both halves (1.95x / 2.06x) but the gain is 3 comeback coins and 25 losers get worse; fallback if re-entry fails live | http://159.223.88.168/hx-c9ois/sw/sw-a2-13-confirmed-stop.md |
| A2.9 more trades at the same edge | ACCEPTED (v1.6): (a) age 1.5 to 3 h: 11 trades, 1.60x / 1.98x (only 3 in half B), paper variant; (b) re-entry after a stop with new buyers ≥ 20 and army ≤ 0.009: 12 trades, 1.91x / 1.79x, paper variant; (c) USDG: 0 pass, drop | http://159.223.88.168/hx-c9ois/sw/sw-a2-9-more-trades.md |
| A2.10 dollars by trade size | ACCEPTED (v1.6): table $250 to $5k; multiple 1.93x down to 1.55x; worst 3 days negative only at $5k | http://159.223.88.168/hx-c9ois/sw/sw-a2-10-size.md |
| A2.11 daily live check vs C1 book | RUNNING: 1 trade, entry_cap bug + 20 s fill lag found | http://159.223.88.168/hx-c9ois/sw/sw-a2-11-live-check.md |
| A2.8 strict rerun with all live fixes | DONE: 58 trades, 1.90x (1.91x / 1.90x), 72 of 72 grid cells above 1x. Rule stands | http://159.223.88.168/hx-c9ois/sw/sw-a2-8-strict.md (trades sw-a2-8-trades.csv = new C1 target) |
| A2.1 exact spec for the Coder | DONE (section 7 superseded by A2.8) | http://159.223.88.168/hx-c9ois/sw/sw-a2-1-spec.md |
| A2.2 how bad are the losers | DONE: 7 of 38 near zero; stop fills 30% worse still 1.80x | http://159.223.88.168/hx-c9ois/sw/sw-a2-results.md |
| A2.3 nearby cutoffs | DONE: 72 of 72 above 1x in both halves | same |
| A2.4 hold a slice with the ladder | DONE: no stable gain; keep 100% at 5x, 70/30 as a paper variant | same |
| A2.5 are the $20M coins pulled | DONE: 19 of 24, usually after a ~99% crash | same |
| A2.6 ladder on fair data | MOVED to the Virality Expert (v1.5) | |
| A2.7 split buys | CLOSED by supervisor v1.2 (A1.5: v3 lane is not crews A or B/C; big buys do not trigger pulls) | |

Version 59, 7 Oct 2026 17:40 Bangkok (server clock). Latest: http://159.223.88.168/hx-c9ois/sw/sw-analyst2-status.md