---
name: trading-edge-hub-12
description: Trading edge hub ledger part 12 (continues trading-edge-hub-11, near its cap): from 7 Oct 2026 11:45 BKK. Read with [[trading-edge-hub]] and [[trading-edge-hub-11]] before any buy/sell call.
sources: [cowork]
---
- 2026-10-07 11:40 BKK (FW Analyst 2 A2.40, http://159.223.88.168/hx-c9ois/fw/a2-ramp-v2.md) | RAMP FIDELITY at $10: buy, stop sell, time stop sell, failed-buy fix, restart with open trade all PROVEN (8 of 8 legs fill/model 1.0000; live minus paper twin mean +0.034x); pull sell UNSEEN | BUG: fm_mc5.py L330 to L332 kills on the FIRST failed sell; pulls are about 9% of trades (14 of 149 half B, about 1 a day), so one pull can halt all buying; fix ordered to PP (sup-to-pp v39 md5 aa67c8476c764c0c458597a1bb67b14c): failed sell on pulled pool = open_unsold, retry, keep buying | per-trade bars at $100/$300/$1k (buy delay 6 s, fill/model 0.995, buy gap under 3%, SELL_SENT 5 s, stop slip -0.05x, time stop live/twin 0.97 (0.95 at $1k), round trip 0.85 at step size); twin must be the SAME size (7 of 176 coins exit differently $100 vs $1k) | depth cost: half B mean x $10 2.26, $100 2.31, $300 2.26, $1k 2.09, $3k 1.78 (so $300 is free, $1k costs about 10%) | CORRECTION: lab.py half B $1k is 2.09x to 6 Oct 12:00 (n149), not 1.69x (to 5 Oct, n146); mean not stable at n about 150, size on fills not on the mean | CROSS-WINDOW: 35 of 70 SW gate trades were co-held by FW; co-held still average 2.32x, do NOT block; proposed cap = one combined drawdown kill (20 tickets of the larger ticket across both wallets), max 1 ticket per coin per window, combined open max 6 (for SW supervisor) | emergency slip 30% to 15% before $1k | FACT, RULE PROPOSED
- 2026-10-07 11:44 BKK (PP pp-to-sup v45) | PULL-KILL FIX LIVE: fm_mc5.py md5 c0210cc1676246ff10569b4e6080cdea; failed sell on pool WETH under 0.1 = open_unsold booked 0x, PULL_UNSOLD log, retry 60 s doubling to 1 h, not counted in max open (buying continues); failed sell on a live pool kills only on 3rd try; self test 4 cases PASS; restart at 0 open, drawdown kept | gmgn-cli quote returns 401 and has no robinhood chain, so the bot's own on-chain sim stays the quote | FIX, FACT (live)
- 2026-10-07 11:47 BKK (FW A2.41 ramp grader http://159.223.88.168/hx-c9ois/fw/ramp-days.txt) | RAMP DAY 1 ($10, 6 Oct 17:26 to 7 Oct 17:26) FAILS: B2 missed trade pool 0xd27657e890fc13ac8a3b9f2e197bbc4ef3afe742 (fee 10000; PoolCreated more than 3.5 days before first swap 6 Oct 20:56; paper 0.93x) because fmmc5 only registers pools from PoolCreated (+4 h 10 min lookback); B1 uptime UNKNOWN (no heartbeat log); B3/B4 PASS; day P&L +$272.40 | TRAP: A3 recon said ALL CLEAR on a PoolCreated-built universe; graders must use the first-swap universe | fixes ordered: PP v40 (T0 = first Swap of any unseen WETH pool; per-minute heartbeat file), A3 v24 (align recon) | count 0 of 3; earliest $100 now 10 Oct 17:26 BKK if days 2 to 4 pass | FACT, FIX ORDERED
- 2026-10-07 11:55 BKK (SW supervisor) | A3.13 operator-exit KILL: selling kit hold when the coin's own bundle wallets have sold X% loses to the 50% trail (X=10%: live B -$4,098/day, kit300k B -$5,498; beats random 0 to 35.5%); mechanism: bundle already sold 10%+ before our 5x in 26 of 41 coins, so it becomes "sell at 5x" and cuts runners (0x8e41fd25e85e624560fa50ff9f117a02cd30b8b3 29.15x to 4.48x) | v1.49 FAST EXITS LIVE 11:48:31 (fastexit dRPC websocket + 3-way parallel send, warm conns 258 ms vs 768 ms cold, approve at buy, paper exits priced on crossing block, ticket cap 5% of gate depth from $1k) | Coder v1.50: check SW pool universe for the PoolCreated trap FW found; cross-window caps deferred to $100 step | KILL, FIX (live)
- [stated] 2026-10-07 12:00 BKK (FW Analyst 1 A1.29, queue v2 http://159.223.88.168/hx-c9ois/fw/lab/queue.md, grid http://159.223.88.168/hx-c9ois/fw/lab/grid.md, ledger FW-008 to FW-014) | lab.py base half B $1k +$14,160/day (n 149) | ALL KILLED on half B: kit $300k as 2nd FW book (134 of 196 trades are coins mc5 already holds; new coins +$1,636/day, -$1,901 without top 3); operator flag skip/tilt (+$5,006); operator reputation (+$21,677 but random same-weight p95 +$27,525); add on strength +600 s cap>=1.2x (+$16,143 vs random adds p95 +$19,960); LP top-up; time of day; A1.23 cheap band re-run (-$3,326) | LESSON: on mc5 no filter beats random stake; the money lever is ticket size | operator early entry +2/+3 min NOT testable on lab data (universe cut by cap at +300 s = look-ahead) | TRAP: A1 v3 scripts timed blocks with one constant 0.101283 s/block, first-swap clock off median 33 min, max 114 min; 4 h stop off -2.7 to +5 min; no past verdict flips | overnight grid cron research-box 02:30 BKK /root/fwa1/lab/nightly.sh, main pulls grid.md :40 | KILL x7, TOOL
- 2026-10-07 12:05 BKK (Third Window DC Analyst 3 DC3.4, http://159.223.88.168/hx-c9ois/dc/dc-a3-4-community.md md5 babf85107164450f2f70f8d507a0679d) | COMMUNITY AS HOLD SIGNAL KILLED: 420 pp3 $500k / $1M crossers (chain clock, v4 375 + v3 45, 13 Sep to 5 Oct) with pool standing at cross + 1 h; hold all from +1 h (3x or 48 h, pull = 0) -$1,244 / -$973 a day; only 16% still >= 50% of cross cap at +24 h; no holder / outside-holder feature has L24 AUC >= 0.60, best A-cut beat-random 0.47 | FLIP SIDE: top quarter holder growth +10 min to +1 h after cross (median +738 holders) = 0% alive at +24 h in BOTH halves, 67% pulled, median pull 1.7 h after cross; AUC 0.64 to 0.73 as a SELL sign; same growth predicts a last 3x spike (AUC 0.64 to 0.70) = the crowd is exit liquidity | product: own token code > 10 kB on 1 of 420 (launchpad template), category tags start 4 Oct = not testable | RULE: a fast jump in holders after the cross is not community strength on Robinhood Chain, it is the last call before the pull | DC3.5 SELL overlay pre-registered, untested | KILLED (hold), LEAD (sell), FACT
- 2026-10-07 12:10 BKK (Third Window DC Analyst 1 Q5 v1, http://159.223.88.168/hx-c9ois/dc/dc-a1-5-q5.md md5 65fd50cef12f0fc3eed1308b6341363b) | Q5 OVERLAYS on the 70 SW wide trades KILLED: (a) SELL NOW when bundle holds >=50% and sold >=10% since entry: A -$2,121 (-$152/day), B +$696, beats random B 79%; (b) HOLD past 5x on non-kit when bundle not selling and outside net buys > 0: A -$4,335, B +$519, beats random 33%; blind hold non-kit 5x exits A -$6,964 | MECHANISM: at SW entry the launch bundle holds a median 0% of wallet-held supply (p75 0.4%, p90 57%) vs 70% at cross + 1 h, so insider levers are gone by gate time; matches SW A3.13 kill | DATA: /opt/tw/an1/xfer.db on research-box = ERC20 Transfer logs, 74 coins (70 SW wide + 4 FW mc5 live), mint to gate + 50 h, 1,688,214 rows, 0 gaps, reusable | TRAP: the mint (zero address to deployer) must not count as the first market transfer, or the 30-block bundle window lands at deploy and comes out empty; dRPC eth_getLogs 500s on address arrays of 3+ x 100k blocks, use one address per call | KILL x2, DATA, TRAP
- 2026-10-07 11:57 BKK (FW A3 v24) | CORRECTION to the 11:47 line: pool 0xd27657e890fc13ac8a3b9f2e197bbc4ef3afe742 was NOT a live miss. True first swap 3 Oct 04:23 BKK (block 78550157); PP paper book v3mc5 took T0 from its feed (starts 6 Oct 12:53) and booked a fake +5 min check at 21:01; per spec (T0 = first swap on chain) mc5 never buys it | TRAP: any paper book whose feed starts mid-history makes fake T0s on old pools; T0 must come from the chain | PP v46 bot (md5 b24d1986c1a73aaade33e6ca319f5980) adds only true-first-swap late pools (earlier-Swap check) + per-minute uptime http://159.223.88.168/hx-c9ois/fw/mc5-uptime.csv; PP v41 fixes its paper book; A2 regrades day 1 B2 (likely PASS, count not reset) | FACT
- [stated] 2026-10-07 12:25 BKK (FW Supervisor, Ben added a Consigliere: http://159.223.88.168/hx-c9ois/cons/cons-start.md, advice v2 md5 9d367a7f4279ad879ce332751ee674da) | FW reply http://159.223.88.168/hx-c9ois/cons/fw-reply.md md5 5844f103af2c251a0c51e28524a3e689: engine merge (SW w2coder base + mc5 signal module) GO with conditions (1 day shadow inside A2 bars, carry PP pull fix + true-first-swap rule, ramp count restarts at switch); interim cap GO but 1 ticket per coin PER WINDOW, not across both (35 of 70 SW gate trades co-held by FW still average 2.32x); lab.py + ledger.csv stay shared, ownership to SW Harness Engineer after 5% reproduction | OPEN CONFLICT: Consigliere relays Ben's size bar as 2 bug-free days; Ben told FW directly 3 successful days per size; FW keeps 3 until Ben confirms | BOTTLENECK line now tops fw-supervisor.md | fw-to-sw http://159.223.88.168/hx-c9ois/fw/fw-to-sw.md md5 7ea44d607ee74737896f525d9e269327 (FW live state /opt/fm/mc5.db on 178.128.20.136) | PROCESS, DECISION PENDING
- 2026-10-07 12:16 BKK (FW Analyst 2 A2.42, http://159.223.88.168/hx-c9ois/fw/ramp-days.txt) | RAMP DAY 1 ($10) REGRADED: B1, B2, B3, B4 all PASS, count 1 of 3 (provisional; final at the 17:30 BKK server run) | CORRECTION (mine): the 11:47 B2 FAIL was wrong; the misleading number was PP's paper T0 for pool 0xd27657e890fc13ac8a3b9f2e197bbc4ef3afe742 (true first swap 3 Oct, A3) | B1 by hand from mc5.log: 10 restarts in day 1, longest gap 55 s (03:43), none over 2 min, no paper +5 min check inside any gap, kill never fired; 2 min sampler covers from 11:47 | grader fix: early grading no longer counts future samples | earliest $100 flip if days 2 and 3 pass: 9 Oct 17:30 BKK | FACT
- 2026-10-07 12:20 BKK (SW Supervisor, Consigliere advice v2) | SW reply http://159.223.88.168/hx-c9ois/cons/sw-reply.md: accepted size rule, one live engine (base w2coder, Live Desk = SW Coder + Exec, mc5 1 day shadow, firstmin only if Exec X4 latency not worse than main), harness = FW lab.py owned by SW Harness Engineer (bt.py + SW H1 stopped; must reproduce SW acceptance within 5% else A3 tick-walk fill goes into lab.py), sw-registry.csv frozen into fw/lab/ledger.csv | interim cap agreed with FW: 1 ticket per coin PER WINDOW, combined drawdown kill 20 tickets, max 6 open combined; FW state /opt/fm/mc5.db on 178.128.20.136 | BOTTLENECK line http://159.223.88.168/hx-c9ois/sw/sw-supervisor-status.md | OPEN: Ben size bar 2 days (Consigliere) vs 3 days (told FW) | PROCESS, DECISION PENDING
- [stated] 2026-10-07 12:30 BKK (Ben decision to FW Supervisor, replaces the 11:45 three-day rule) | RAMP RULE = 2 bug-free days in a row at each size ($10, $100, $300, $1k), then Ben flips the next; bug-free = no key bug found that day + fills in line with paper; a found bug is fixed and logged, count starts after the fix | day 1 (6 to 7 Oct) had two bugs fixed 7 Oct (pull kill, late pools), so the count starts with day 2 (7 Oct 17:26 to 8 Oct 17:26 BKK); earliest $100 = 9 Oct 17:26 BKK | elite-plan v3 md5 aac1583e45ff6991e76872632e8c58d5, sup-to-a2 v34, fw-reply updated | DECISION
- 2026-10-07 12:20 BKK (FW A1 v31, queue.md "Second engine", ledger FW-015/016) | SECOND ENGINE (C deep oversold + AB dip in uptrend on older coins) worth about +$37/day at $1k combined if the 4 Oct study holds = about 1/380 of mc5; study used coins alive on 4 Oct (survivor bias up); forward paper book live http://159.223.88.168/hx-c9ois/fw/a1-eng2-paper.txt (research-box, every 4 h, zero Claude credits); pass = 20+ closed per pattern, win rate within 10 pts of study, avg > 0 after fees | A2.42 regraded ramp day 1: 4 of 4 bars PASS (but under Ben's 12:30 rule day 1 had the pull-kill bug found, so the bug-free count starts day 2) | PAPER, FACT
- 2026-10-07 12:22 BKK (FW Analyst 2 A2.43, http://159.223.88.168/hx-c9ois/fw/ramp-days.txt) | GRADER ON BEN'S 2 BUG-FREE DAYS RULE: NEED=2, new bar B5 "no key bug found this day" from /root/fwa2/ramp/bugs.json on research-box (log a bug under dayN, rerun grade N) | day 1 logged with 2 bugs (pull kill PP v45, late first-swap pools PP v46), so not counted although B1 to B4 PASS; count 0 of 2; earliest $100 grade 9 Oct 17:30 BKK | top line says ACTION only on a B1 to B4 fail or when the count reaches 2; a day lost to a logged bug shows OK (not counted), so no wasted wake | TOOL, FACT
- [stated] 2026-10-07 ~12:25 BKK (Ben to SW Supervisor) | "3 good days" per size step; applied to SW Coder v1.51 (md5 ef206336), elite plan v1.3, SW status (earliest bigger size 10 Oct 12:00 BKK), cons sw-reply | CONFLICT: FW ledger says Ben told FW 2 days at 12:30 BKK; asked Ben for one number for both windows | DECISION PENDING
- 2026-10-07 12:55 BKK (SW Supervisor, Consigliere v2.6 "size rule final: 3 days") | 2-vs-3 conflict RESOLVED: 3 good days per size step, both windows | order of work: MERGE FIRST at $10 (plan today, 1 day mc5 shadow, switch ~8 Oct evening), count 3 days on the merged engine (switch resets the count); earliest $100 ~11 Oct evening BKK | interim cap adds $ cap per coin = 2 tickets combined; FW live wallet 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286 | Coder v1.52 md5 41d09f9a; PP split line in sw-tasks-pp.md (PP Data seat open) | BOTTLENECK now valued at next step: ~$260/day ($100 tickets on live-book paper) | DECISION
- 2026-10-07 12:40 BKK (SW Harness Engineer H1-H3, http://159.223.88.168/hx-c9ois/sw/sw-harness.md) | bt.py LIVE, acceptance PASSED within 1% (live book A +$2,580 / B +$10,682; kit300k A +$2,153 / B +$13,482; A2.37 KILL 0.62x); 20 s to 2 min per rule | golden data research-box /opt/sw/harness/gold/ (6,938 traded v3 WETH/USDG pools since block 61M, 5.8M rows, hourly, daily snapshot md5 c92f907cb7a08968c87ff0f16547d661) | FINDINGS: live book half B honest to now = +$8,768/day (not +$10,681); A3.12 had no fee, kit300k with 1% fee A +$2,017 / B +$12,867; kit300k size ceiling ~$3k ticket (falls at $5k), live book still rises to $5k | X4: firstmin 178.128.20.136 same latency as main (seq warm 217 vs 216 ms) but 1 GB RAM, no swap: swapfile + OOMScoreAdjust ordered before merge | harness choice: head to head bt.py vs FW lab.py (mc5 +$14,160 half B) instead of stopping bt.py | FACT, TOOL
- 2026-10-07 12:45 BKK (Third Window DC Analyst 1 Q5b, http://159.223.88.168/hx-c9ois/dc/dc-a1-6-q5b.md md5 3196bbef286e687eb14f2002ed00a100) | A3 HOLDER-GROWTH DEATH SIGN AS SELL on SW (70) + FW mc5 (176) trades KILLED, SIGN FLIPS: sell at entry +15m/+1h/+4h when outside holder growth (A3 loose ruler) >= +528 in 50 min: pooled A -$12,065, B -$42,268, beats random B 1%; FW B alone -$33,752 | coins that fired ended at mean 4.49x (median 2.13x) vs 2.34x (0.46x) for the rest; e.g. FW 0xc38c332012a9116dcade0c7f43b5d3c71799a9f8 sold 1.25x vs real 45.5x | MECHANISM: holder growth is stage dependent: right after an early entry (first swap + 5 min, SW gate) it is the run itself; at cross + 1 h on $500k/$1M crossers (A3) it is the operator's last wave before the pull | RULE: never use a holder-growth rule without stating the coin stage it is read at | LEAD (not pre-registered): growth >= +528 after entry as HOLD LONGER / SIZE UP on mc5, proposed Q5c | DATA: /opt/tw/an1/xfer.db now 203 coins (SW 70, FW mc5 176 rows, FW live 4), 2,697,380 rows, 0 gaps | KILL, LEAD, DATA
- [stated] 2026-10-07 13:00 BKK (SW Analyst 2, status v56 md5 bed8fa663a4dcbbf228781efe7604d83) | research-box liqbuild.service (PP liqbuild2.py, liquidity owner/lock tables) DEAD: OOM-killed 11:53 BKK 7 Oct after 23 h 3 min, memory peak 14.6 GB, never finished (last log 13:21 BKK 6 Oct "1875 s owner_at for pulls", liqlock.db 21 GB last written 13:22 BKK 6 Oct, no LIQBUILD_EXIT) | ask PP: rerun owner_at step in chunks with progress logging; hold-or-sell 250k cross-check waits | TRAP: a research job silent 20 h at 96% CPU was a memory leak, not slow progress | FACT
- 2026-10-07 13:15 BKK (SW Supervisor) | MERGE STARTED: mc5 signal module runs in SHADOW inside w2coder since 12:40:17 BKK (grading page http://159.223.88.168/hx-c9ois/sw/w2-mc5-shadow.md, auto every 2 min); shadow day ends 8 Oct 12:40 BKK; switch only if every shadow decision matches fm_mc5 within X5 bars; $ cap per coin (2 x larger ticket) live | H4 head to head DONE (http://159.223.88.168/hx-c9ois/fw/lab/h4-head-to-head.txt): ONE command lab.py + ONE event engine (bt.py) for new rules; mc5 in bt half B +$13,130 vs lab +$14,160 (-7.3%); gap = 2 coins (0xa402616f4d05cca33947fabc6ff51ba795f6d2f4 stop 0.49x vs hold 6.30x; 0x3939ce55ec4e0ff8df9aa6469a795c793da06d55 20.3x vs 24.4x), rest within ~1%; Harness v1.3: settle against fm_mc5 live code | mc5 size sweep (bt) half B $1k +$13,130, $3k +$27,959, $5k +$34,741/day, still rising at $5k | FACT, TOOL
- [stated] 2026-10-07 13:20 BKK (FW Supervisor applying Ben's final decisions via Consigliere v3.0 to v3.6; supersedes the FW 12:30 line) | FW: ramp days count ONLY on the merged FW + SW engine after its 1 day shadow (ladder $10, $50, $100, $200, $300, 2 bug-free days each) | FW tasks queued (no idle rule): A1 mc5 reference vectors fw/lab/mc5-vectors.csv + lab.py at $50/$200; A2 shadow grader fw/shadow-grade.txt + ladder update + wallet need per step ($50 about $1,250, $100 about $2,450, $200 about $4,850, $300 about $7,300; wallet now about $384); A3 $50/$200 paper twins + recon of merged-engine shadow | FW supervisor self-wake every 20 min | DECISION
- 2026-10-07 13:15 BKK (Third Window DC Supervisor; Engine EN.5 full universe http://159.223.88.168/hx-c9ois/dc/dc-en-5-an4.md, A2 DC2.4 phase 2 http://159.223.88.168/hx-c9ois/dc/dc-a2-7-pons.md) | E-AN4-1 to -4 v4 ENTRY RULES all FAIL on the launch-based universe (every v4 pool that printed a sellable $100k): E-AN4-1 3x/48 h A -$307 | B +$216 a day; buying every coin at first sellable $100k loses $72 to $77 a trade; the 11:31 PASS was survivor look-ahead | only lead: E-AN4-1 with T15 exit +$75 | +$178 a day, beats random 100%, forward only after a new prereg line | Engine v1.6 fixes time-exit bug (one garbage price gave +$5.5e25) | A2: "operator holds float at entry" = 3x 0 to 11% vs 13 to 56% without, same sign in 4 of 4 universes (v4 non-pullable $100k/$250k, Pons graduated $100k/$250k): SKIP candidate, Pons creator table missing (tokens.db 35%) | A2: creator-held LP pulls 97 to 99% in 24 h, hold only burned or real-locker LP | DC desk main bet now A1 Q5c HOLD LONGER (outside holder growth >= +528 within 60 min after OUR entry), scored only on unseen data (DA.4 + forward from 13:15) | DC seats 11 to 7 (A2, A3, A5, Scout closed) | KILL x4, LEAD x2, DECISION
- 2026-10-07 13:21 BKK (FW Analyst 2 A2.44, http://159.223.88.168/hx-c9ois/fw/shadow-grade.txt + http://159.223.88.168/hx-c9ois/fw/ramp-days.txt) | SHADOW GRADER LIVE (research-box /root/fwa2/shadow, hourly :05, main pulls :08): per coin MATCH / MISMATCH / PENDING / PRE-START, merged engine shadow (sw/w2-mc5-shadow.md) vs fm_mc5 live (mc5live.txt + A3 chain audit) vs A1 vectors (pending) vs PP paper; a2-ramp-v2 bars | FIRST RUN 1 MISMATCH: pool 0x85b3abf1dd315a9485523fd8b0d1be388b1363b9 (T0 7 Oct 12:47:22, after shadow start 12:40:17) shadow says seen_late, fm_mc5 checked it (cap_out); same action, but seen_late = shadow saw the first swap more than 330 s late, which on an in-band coin is a missed buy; asked SW Coder why + delay / fill / gap columns on the shadow page | TRAP: coins with T0 before the shadow start must be graded PRE-START, not MISMATCH | RAMP GRADER on Ben's final ladder $10/$50/$100/$200/$300, 2 bug-free days each, counted ONLY after the merged engine's live switch (switch.json); WAITING until then; wallet need $50 about $1,250 ... $300 about $7,300 vs about $384 now; hourly auto mode | TOOL, FACT
- 2026-10-07 13:21 BKK (FW A2.44, http://159.223.88.168/hx-c9ois/fw/shadow-grade.txt) | merged engine mc5 SHADOW started 7 Oct 12:40:17 BKK (SW, sw/w2-mc5-shadow.md); FW shadow grader live hourly | first run 1 MISMATCH: pool 0x85b3abf1dd315a9485523fd8b0d1be388b1363b9 (first swap 12:47:22) shadow seen_late (>330 s) while fm_mc5 saw it on time (both no buy, cap_out) = on an in-band coin it would be a missed buy | relayed to SW in http://159.223.88.168/hx-c9ois/fw/fw-to-sw.md v2 md5 f5c6a93e7712e62bdea509603423b025 (why late, add delay / fill-model / buy-gap columns, post switch minute + live board path) | ramp-days.txt now WAITING for merged engine switch | FACT, OPEN
- [stated] FW 7 Oct 13:35 BKK: 5th live mc5 buy 13:25:14, pool 0x86303aff9563213a7c5f877bbe15eacde98d7c50, $10, cap $470k, fill/sim 1.075. Alchemy RPC key hit monthly cap (A3 found; half C stalled 07:15 to 13:25, moved to dRPC); fm_mc5 live feed unaffected (dRPC first, Alchemy last). fm_mc5 late-pool thread logs ~30 '429 late_pause' an hour: PP v42 to fix, A3 v26 to check for misses. A3 v25 twins live at $10/$50/$100/$200/$300/$1k (30x coin: $10 twin 30.39x, $1k twin 26.75x, live 30.18x).
- [stated] FW 7 Oct 13:40 BKK CORRECTION: 13:25 live buy pool 0x86303aff9563213a7c5f877bbe15eacde98d7c50 (token 0xc708284dcccad5360e93216947f6eb296291f13c) chain audit = delay 4 s, fill/model 1.0000 (bot's 1.075 was its own sim ratio, not extra coins). 429s on fm_mc5 late thread came from PUBLIC RPC, not Alchemy; PP v42 build (dRPC first) installs at first 0-open minute. PP paper book v3mc5 skipped this live coin as no_state (public RPC has no history) and was down 12:38 to 13:04 (db lock): PP v43 to move it to dRPC + backfill. Paper trades lost to late_pause: 0.
- 2026-10-07 13:35 BKK (Third Window DC Supervisor; A4 AN4.2/4.3/4.4 http://159.223.88.168/hx-c9ois/dc/dc-an4-2-skip.md, http://159.223.88.168/hx-c9ois/dc/dc-an4-4-v3roots.md; Engine http://159.223.88.168/hx-c9ois/dc/dc-en-q5c.md) | KILL: SKIP "operator holds float at entry" on FW mc5 + SW live books ($ a day down 25 to 34% on mc5, skips 9 trades that made 3x+; SW half B fails 36%): on fast-pump books the operator float is part of the pump | v4 good-root list does not reach the books (0 of 248 trades; books trade v3) and a fixed v3 root list from half A does not carry to half B (v3 operators rotate funding chains weekly) | LEAD, LEAK CHECKED: rolling 3-day record of the creator's funder cluster (3+ coins, counting only 5x hits printed before our entry and misses already dead before it, open coins = unknown, 15%): HOT 61% / 63% honest 5x (lift 2.56 / 1.89), COLD 12% / 18% (lift 0.44 / 0.47), both halves, 909 v3 coins with roots in research-box /opt/tw/an4/crew_v3.db | next: E-AN4-5 prereg = hot -> hold longer (50% trail / 48 h), cold -> sell early, scored on FW mc5 + SW live books | ENGINE TRAP: a plain 50% trail beats BASE (5x / 0.3x / 48 h) on ANY open position by +$88 to +$103 a $500 trade, so every HOLD overlay must beat trail-on-all, not BASE | Q5c HOLD LONGER shadow live on every FW position since 13:15; SW per-trade file asked via Consigliere | KILL, LEAD, TRAP
- [stated] FW 7 Oct 13:31 BKK: 5th live trade closed: pool 0x86303aff9563213a7c5f877bbe15eacde98d7c50 stop 0.5x, bot x0.33 (0.001272 ETH vs own quote 0.001432, gap 11.2%, sent 2.7 s after trigger, slip cap 30%), tx 0x4d3c64a0ad37aaf89adfeec1f1c0b1ccdf93010e4f48a1f20b0d99181ab4b8cd. PP tape-only twin 0.488x. First live sell clearly under paper; A3 v27 chain audit (cascade vs own slip), A2 v36 grade vs kill rule. PP v48: v3mc5 on dRPC first; no_state cause was public RPC rate limit.
- 2026-10-07 13:38 BKK (FW Analyst 2 A2.45, http://159.223.88.168/hx-c9ois/fw/fw-analyst2-status.md) | 13:31 LIVE STOP SELL GRADED, pool 0x86303aff9563213a7c5f877bbe15eacde98d7c50: chain 0.341x (0.001289 WETH out / 0.003780 in), fill/model 1.0000 at the block before (A3), SELL_SENT 2.7 s = clean execution | tape rebuild: first touch block 82263885 (mid 0.505), crossing sell 0.855 WETH at 82263894 -> 0.431, trigger about 82263912 (bot quote 0.379x), fill 82263941 at 0.341; crossing to fill about 5.7 s in a run of 0.75 to 0.9 WETH dumps | loss split: gap-through -0.07, feed/decide about 1.8 s -0.05, send-to-fill -0.04, slippage beyond pool state 0 | B3 stop slip vs PP tape twin 0.488 = -0.147 FAIL, vs crossing-block twin 0.427 = -0.086 FAIL, vs own trigger quote -0.038 PASS: the FAIL is the paper side (twins priced at or before the crossing are unreachable in a cascade) | RULE PROPOSED: price stop/pull twins at pool state at crossing block + measured live latency; step-down rule compare against trigger quote | fm_mc5 (md5 a05678a3873dbfd61d8be1f564cd85e7) kill counts only BUY fill/sim under 0.9 twice, so sells never count; this sell is 1 of 2 toward the ramp STEP-DOWN rule (sell_gap 0.112 > 0.10), no effect at $10 | not a key bug; fm_mc5 days no longer count anyway | merged engine fastexit (about 0.5 s send) would save about 0.03 to 0.04x on such a cascade | FACT, RULE PROPOSED
- [stated] FW 7 Oct 13:45 BKK: A2.45 graded 13:31 sell: execution clean (fill/model 1.0000), loss vs paper = cascade + ~4.7 s from stop cross to fill; not a bug, not toward kill rule. Lever: stop latency (send 2.7 s vs 1.4 s on trade 4). PP v44 queued: measure stop path, detect cross earlier, pre-sign sell near stop.
- [stated] FW 7 Oct 13:50 BKK: A3 chain audit of 13:31 sell: 0.341x, own slip 0 (fill/model 1.0000); stop cross block 82263894 at 0.431x, filled 4.7 s later after two more dumps: speed cost ~0.09x. Second gap: paper twins count 1% fee + gas in the stop value (at $10 the $10 twin stopped 2 min earlier at 0.495x); bot uses raw quote. Rule: bot must copy lab.py (the proven rule). A1 v33 adds stop value to mc5 vectors; PP v45 aligns bot after vectors. PP v49 speed fix live 13:42 (fm_mc5 ef15ce505783f939cc06384a566ba331: 0.5 s poll while open, sell-path RPCs parallel, STOP_PATH log). Own signer/pre-sign held until STOP_PATH shows gmgn-cli is the bulk.
- [stated] FW 7 Oct 13:55 BKK: PP v50 v41 rebuild done: 145 of 147 old-born rows in PP's v3mc5 paper book were not at their true first swap; only 1 was a paper BUY (pool 0xd27657e890fc13ac8a3b9f2e197bbc4ef3afe742, 0.929x), voided. Paper book trade results otherwise unchanged. No new live trade since 13:31.
- [stated] FW 7 Oct 14:25 BKK: Consigliere v5.2 to v5.5 adopted: merge of FW and SW teams cancelled (D-016 v2), one shared live engine, wallet 1 = FW, wallet 2 = SW, $200 total cap, one kill switch; switch rule D-003 v3 = 5 matched shadow buys in a row or 8 Oct 12:40 (shadow at 1 of 5: pool 0x86303aff9563213a7c5f877bbe15eacde98d7c50 shadow 0.49x vs live 0.33x). Rings retired (D-022): nudge armed doorbells with update_trigger; FW Supervisor doorbell now trig_01Qn4DRR3Uwv4PXoCJd6AdJg. FW key check items F3/F4 (firstmin authorized_keys) held for Ben's OK.
- [stated] 2026-10-07 14:24 BKK (FW Analyst 1 v32/v33, http://159.223.88.168/hx-c9ois/fw/lab/mc5-vectors.csv + mc5-vectors-readme.md) | mc5 REFERENCE VECTORS from lab.py: every v3 WETH pool (any fee) with TRUE first swap 4 Oct 14:20 to 7 Oct 14:20 BKK: 831 pools, 38 BUY, 16 dust guard, 777 band; all 4 live trades BUY with same exit; nightly rebuild research-box 01:50 BKK, main pull 09:55 | STOP BASIS: lab.py stop = constant-product quote x 0.99 - gas (fee-in); raw-quote rule fires later; half B fee-in vs raw $10 2.26x vs 2.32x, $1k 2.09x both = immaterial, copy lab.py | pool 0x86303aff9563213a7c5f877bbe15eacde98d7c50: lab stop block 82262763 (0.4952), raw rule 44 blocks later (0.4323), bot sold ~2 min later 0.341x = mostly bot lag | LADDER lab.py half B: $10 2.26x +$164/d, $50 2.32x +$857, $100 2.31x +$1,699, $200 2.29x +$3,332, $300 2.26x +$4,896, $1k 2.09x +$14,160 | TRAP: lab.py now forces a size LADDER sweep (Harness Engineer header) and reads argv[1] as JSON, so wrappers must pass the rule path exactly | TOOL, FACT
- [stated] FW 7 Oct 14:27 BKK: A1 mc5 reference vectors live (fw/lab/mc5-vectors.csv: 831 pools 4 to 7 Oct, 38 BUY; all 4 live trades match). Stop basis: fee-in (lab.py) vs raw barely matters on the book ($10 2.26x vs 2.32x; $1k 2.09x both): copy lab.py; on pool 0x86303aff9563213a7c5f877bbe15eacde98d7c50 the 0.34x was mostly bot lag. Ladder on lab.py half B (n 149): $10 2.26x +$164/day, $50 +$857, $100 +$1,699, $200 +$3,332, $300 +$4,896, $1k 2.09x +$14,160; no depth cost through $300, 7 to 10% at $1k. A1 v35: latency cost curve (value of 1 s stop speed).
- [stated] FW 7 Oct 14:35 BKK: 6th live mc5 buy 14:28:39, pool 0x4d2f7469c43a7d0e0dd4f6bbba8289e4624ca5f2, $10, cap $404k, bot fill/sim 1.000 (A3 chain audit pending). PP v51: bot stop now copies lab.py (fee-in, per-ticket; tick-walk confirm off), vector replay 30/30 PASS incl. pool 0x86303aff9563213a7c5f877bbe15eacde98d7c50 at block 82262763 0.4952; fm_mc5 md5 e3ad18065c262e097b69813145411a58 installs at first 0-open moment.
- 2026-10-07 14:40 BKK (SW Analyst 1 A1.29, http://159.223.88.168/hx-c9ois/sw/sw-scorecard.md md5 76b4328b9f9d1cd529a5a81b0712e9eb, cron main 18:00 BKK daily, /opt/a1score/sc29.py) | D-020 SW SCORECARD v1 first issue: (a) SW twin $ a day on coins FW did NOT take +$10,318/day since 1 Oct (22 trades) = ALL replay 1 to 6 Oct before FW traded; FORWARD (engine since 6 Oct 13:04) = $0, 0 trades: both engine-era twin trades (CHONK 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468, CrawlScan 0xa6caf95634d4bcffdfb535eec889c9c928ff13d5) were coins FW also took | (b) $2M+ fair runners caught after T0+60 s and before halfway: 12 of 33 (36%) since 1 Oct; engine era 1 of 3 (CHONK caught; CLAUS 0x206211d456a3a9dbe45e13507aac5131ace9c716 missed army 9.46%, CORES 0x2cdf271007c7ebb6013d55492e156031c535094f missed receiver 25%); 7 of 30 replay-era runners hit halfway inside minute 1 | (c) live/twin 0.852 (1 trade) | TRAP: w2.db swaps table starts late for some pools (CLAUS stored from 208 min after T0, CORES 159 min); scorecard fills from chain | to pass D-020 by 14 Oct SW needs forward wins on coins FW skips | FACT, OPEN
- 2026-10-07 14:58 BKK (SW Supervisor; Exec X7/X8/X9 http://159.223.88.168/hx-c9ois/sw/sw-exec-x5-x8.md, treasury http://159.223.88.168/hx-c9ois/treasury.html) | GMGN wallets trade ONLY through GMGN (server-side signing, about 10 s path; gmgn-cli quote 1.6 to 3.0 s from firstmin); current API key binds ONE EVM wallet 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286, wallet 2 not bound | D-021 built as: SW desk = engine wallet 0x04a057D15bd07581DC88b0a37E41EB69Ceb8C2D3 on fastexit, FW desk = GMGN wallet 1 | CORRECTION (mine, same turn): I first called the 0.33x vs 0.49x FW stop gap 'GMGN path, a third of each stop'; FW's audit shows it was mostly the raw-quote stop firing 44 blocks after lab.py's stop plus about 4.7 s in a cascade (PP v49/v51 fix); own-signer for FW waits on FW's STOP_PATH log | D-020 TRAP: SW scorecard +$10,318/day on coins FW skipped is all 1 to 6 Oct replay; forward = 0 trades, so SW has no forward D-020 proof yet; hunts queued A2.39 missed runners (21 of 33 $2M+ missed), A3.14 early exit for losers, A1.30 v4 pools (D-024) | X8: $200 combined cap PASS 7 paper cases, 4 gaps (FW open unmarked, no D-018 drop, ladder missing, kill does not stop FW) | twitterapi.io $13.69 left, about $18/day, dry about 08:00 BKK 8 Oct, Needs Ben | FACT, CORRECTION, OPEN
- [stated] FW 7 Oct 14:42 BKK: 6th live trade closed: pool 0x4d2f7469c43a7d0e0dd4f6bbba8289e4624ca5f2 stop 0.5x, bot x0.45, got 0.001738 ETH vs own quote 0.001624 (+7%), STOP_PATH cross->sent 1.51 s (cross->seen 1.24, gmgn-cli 1.13) vs ~4.7 s on the 13:31 sell: PP v49 speed fix works. New stop-rule build installed at 14:42:18 (SELFTEST_OK). Live book 6 trades: ~$60 in, ~$320 out.
- [stated] FW 7 Oct 15:35 BKK: 7th live trade: pool 0x644eac7020466c554369e0592637f24dd95fbb13 bought 15:16:57 $10 cap $703k (fill/sim 0.946), sold 15:35:06 stop 0.5x x0.43, 0.001654 ETH vs own quote 0.001873 (11.7% under quote). Live book 7 trades ~$70 in, ~$325 out; 4 of the last 4 since 05:53 are stops (0.57x, 0.34x, 0.45x, 0.43x). Kill rule not hit (13:31 and 15:35 >10% under, not consecutive).
- [stated] FW 7 Oct 15:53 BKK (FW Analyst 1 v35, http://159.223.88.168/hx-c9ois/fw/lab/latency-curve.md md5 a682f5cd1aa54809c91a7ecd97790e59, ledger FW-017) | LATENCY COST CURVE on lab.py half B (n 149): stop sell delayed 0 to 10 s after trigger = flat ($200: $3,334 to $3,350/day; $1k $14,125 to $14,195): 1 s of stop speed at $200 = $0/day | buy delay (fill_delay_s): live 6 s is best point ($3,334 at $200); 0 s $3,023, 3 s $3,287, 5 s $3,256, 7 s $3,154, 10 s $2,967; jagged, few coins decide | RULE: no engineering time on mc5 latency in lab terms; caveat: lab sells at next block end and cannot see live cascade gap (13:31 sell 0.34x), so the PP stop speed fix still matters for live vs paper fidelity | NO EDGE (speed), FACT
- [stated] FW 7 Oct 15:55 BKK (FW Analyst 1 v35 task 2, http://159.223.88.168/hx-c9ois/fw/lab/vector-replay-diff.md, ledger FW-018) | PP rp46 replay rerun independently: bot stop code (fm_mc5 md5 e3ad18065c262e097b69813145411a58) vs mc5-vectors.csv (md5 b9b25cc764c59c96655cd3ed40395239) PASS 30 of 30 stop rows, 0 mismatch | NOT YET PROVEN: entry buy/skip on 831 rows, 6 time exits, 2 pull exits (pools 0x351722e71c93290847c1e3b5a585369b11c42236, 0xad4cfc3433988aa9495d94b4163f84e2422bb1d2); ask PP to extend rp46 | FACT, OPEN
- [stated] 7 Oct 16:00 BKK, FW sup: KILLED latency as an mc5 edge (A1 FW-017): stop sell 0 to 10 s late changes half B by ~$0 a day at $200; live 6 s buy delay is the best point (0 s = minus $311/day). No engineering on speed; PP own signer stays off. Bot stop code matches lab.py on 30/30 vector rows (FW-018); buy/skip and time/pull exits being replayed next (PP v48). New A1.31: calibrate the kill rule's 10% gap for normal stop crash slippage.
- [stated] 7 Oct 16:15 BKK, FW sup: live fm_mc5 (md5 e3ad18065c262e097b69813145411a58) now EQUALS lab.py on all 869 vector checks (54 entry, 777 band, 30 stop, 6 time, 2 pull; PP v52). Ben asked why buys are above $500k: mc5 buys at +300 s by design. Opened A1.32 entry time ladder (+10 s to +5 min) on uncut v3 (831 pools) + v4 (PP builds): no look ahead (select only on data at t, band fit on half A only), honest tick walk fills, $10 and $1k, vs random p95. ETA v3 18:30, full 19:30 BKK.
- [stated] 7 Oct 16:25 BKK, FW sup: kill rule trap. Normal mc5 stop crashes land >10% under trigger quote 14 to 18% of the time (A1 FW-019); both strikes today (13:31, 15:35) were market dumps, not bugs. 15:35 sell (token 0x3796191a1289ac970286d7066dd04824c2e1e77d): a 0.8998 WETH sell landed AHEAD of ours in the same block 82335818, so fill/model vs end of previous block read 0.856. Rule: measure own slippage vs pool state immediately before our tx in log order (intra block). Proposed kill v2: stop/pull sells strike on intra-block fill/model < 0.90; time sells and buys keep 10% under paper. Sent to Consigliere.
- [stated] 7 Oct 16:30 BKK, FW: KILL RULE v2 approved (Consigliere v5.12). Stop/pull sell strike = own slip under 0.90 vs pool state right before our tx (intra block) OR trigger to sent over 2 s; time sells and buys keep 10% under paper; kill = two strikes in a row. A3 back fill: 0 own slip strikes on 7 sells + 7 buys (all 1.0000). But 15:35 send time 2.90 s (gmgn-cli 2.00 s) is a strike under the 2 s guard: speed is worth ~$0 a day yet gmgn-cli variance can trip the kill switch; PP asked for the gmgn-cli split.
- 2026-10-07 16:35 BKK (FW Analyst 2 A2.46, http://159.223.88.168/hx-c9ois/fw/fw-analyst2-status.md) | KILL RULE v2 regrade of the 7 live sells on 7 Oct: own slip 1.0000 on all 7; strikes = 13:31 pool 0x86303aff9563213a7c5f877bbe15eacde98d7c50 (sent 2.7 s) and 15:35 pool 0x644eac7020466c554369e0592637f24dd95fbb13 (cross to sent 2.90 s); 14:42 pool 0x4d2f7469c43a7d0e0dd4f6bbba8289e4624ca5f2 (1.51 s) broke the streak, so NO KILL; streak now 1 | 00:19 and 02:41 stop sells had no send log (before v32), so they cannot be graded on the 2 s guard | time sell 03:40 30.18x vs $10 twin 30.39x (0.7% under): no strike | RISK: gmgn-cli alone takes 1.13 to 2.00 s of the 2 s budget (STOP_PATH), and 2 of 4 logged stop sells breached, so the 2 s guard mostly measures gmgn-cli and can trip the kill with no edge cost (A1 FW-017: speed worth about $0/day) | ramp grader: B3 = kill rule v2 from A3 per-leg table, B4 = TIME sells only vs same-size twin, buy gap bar 10%; TRAP: A3 changed its twin text to "paper twin $10 (live size): ...", which silently zeroed my B4 and the shadow grader's A3 buy delay until the regexes were fixed | FACT, TOOL, RISK
- 2026-10-07 16:45 BKK (SW Analyst 1 A1.30a, http://159.223.88.168/hx-c9ois/sw/a1-30-v4-count.md md5 f5dd864e0a79020e113f1a7e84888ffe) | V4 RUNNER POND SIZE since 1 Oct (PoolManager 0x8366a39cc670b4001a1121b8f6a443a643e40951, ETH/WETH quoted): 23,016 pools, 18,363 traded, fair $1M+ 119, fair $2M+ 75 pools (71 tokens, about 12 a day vs about 5 a day on v3), $10M+ 12; 66 of 75 hooked; 42 from one hook family (fee 3000, one new hook contract per coin, address ends a80) | v4 swap sign: swapper ETH delta < 0 = buy (99.3% move price up) | UPPER BOUND: no pull cut, no wash check (top row 0x0b7c6c138dc75622f4200d157b13fb93e6bd94fa $164M looks fake) | live gate sees none of these; money test waits for FW A3 uncut universe | scorecard sc29 v1 freeze time corrected to 14:34; (c) now reads Exec sw-live-trades.csv | OPEN
- [stated] 7 Oct 17:06 BKK, FW: send-time cause found and fixed (PP v54). Each stop sell spawned a new gmgn-cli process (node start 0.55 to 1.8 s + new TLS); gmgn-cli has no quote/confirm step. Fix: warm daemon gmgnd (same signed swap request, kept-alive, unix socket 0600, never double sends; falls back to CLI if down), round trip 0.25 to 0.35 s; feed poll 0.25 s while open. fm_mc5 md5 c69090a40e504d40245cf58bc3d76834 live 16:41, replay 869/869 PASS. Expected stop trigger to sent under 1 s. Consigliere v6.0 credit cut: seats wake on doorbell only; FW Supervisor stood down with handover on fw/fw-supervisor.md.
- [stated] FW 7 Oct 17:25 BKK (FW Analyst 1 A1.32 / J-001, http://159.223.88.168/hx-c9ois/fw/lab/entry-ladder.md FW-020, http://159.223.88.168/hx-c9ois/fw/lab/j001-crew.md FW-021) | EARLY BUY ON v3 KILLED: uncut universe 6,085 v3 WETH pools (13 Sep to 6 Oct), no look-ahead, band fit on half A: +20/+30 s half B $10 +$181/+$213 vs +5 min +$164 but WITHOUT TOP 3 $42/$53 vs $71, $1k no gain ($14,077 vs $14,160); coins +5 min skips lose without top 3 (+30 s B minus $57); even hindsight (same coins bought earlier) is below +5 min | crew flags at creation (funder roots for all 6,085 pools, earlier-launch record known before T0): 0 of 12 arms pass; skip puller crews cuts runners; hot crew only +$163 A / +$4 B (crew record does not carry) | TRAP: in the uncut universe mc5 at $10 does NOT beat random same-count p95 ($164 vs $210, cheap coins' 100x tail); at $1k it does ($14,160 vs $9,239); use uncut random baseline | open: PP score arm, v4 pools | KILL x2, TRAP
- [stated] 7 Oct 17:25 BKK, FW A1 J-001: EARLY ENTRY KILLED on v3 (6,085 pools). Buying at +10 s to +3 min does not beat the +5 min mc5 rule: +20/+30 s only match it via the top 3 coins (without them $42/$53 vs $71 a day at $10), no gain at $1k; even hindsight early buys of the mc5 set are below +5 min. Crew flags at creation (skip puller crews, hot crew): KILL, 0 of 12 pass (skipping pullers cuts runners). Open: PP indicator score arm and the v4 universe (PP v55: research-box /opt/sw/v4/v4vec.db ready).
- 2026-10-07 17:30 BKK (Third Window DC Analyst 1 Q5c INTERIM, http://159.223.88.168/hx-c9ois/dc/dc-a1-7-q5c.md md5 a5a9b3c2ba528c5d4a8be0491236ad4b) | Q5c V1 HOLD LONGER (outside holder growth >= +528 at entry +15 m / +60 m, then 50% trail or 48 h) on the launch-based universe (Engine q5c_paths, every coin with a first sellable $300k, 2,096 with logs, weighted to the true 5x / non-5x mix): $ added A +$13,559 / B +$12,686 ($500, 150 fires) but beats random only 86% / 60% (bar 98%); without the 323 partial paths B +$3,448, 39% | MECHANISM: the money is the trail, not the signal: 50% trail on ANY open position adds +$108 / +$145 a position at the 15 min read; holder growth picks no better holds | v4: 14 of 40 B fires pulled within 48 h, so a v4 HOLD signal is mostly moot | V2 (LP gate) and V3 (70% trail armed at 2x) still unscored | FAILS (interim), TRAP
- 2026-10-07 17:35 BKK (SW Analyst 2 A2.39, http://159.223.88.168/hx-c9ois/sw/sw-a2-39-missed-runners.md md5 bff7e8f36af5552c9097f2d45de14d3d, lab ledger SW-A2.39) | HOLDER FILTER DOOR KILLED: buying $1M gate coins (age <=3 h) that the live top 0.08 / army 0.015 filters block, when the blocking holders sold back <=10% before the gate (D1, picked on half A): +$70 / -$92 a day vs live, worst day B -$6,480 (bar -$3,951), beats random 2.5% / 6.5% | the 75 blocked coins: median 0.28x, 5% reach about 5x vs 34% of the 70 live coins = THE HOLDER FILTERS EARN THEIR KEEP; a crew that has not sold yet still dumps after the gate (0xb6039ac45d5d355dd98f318dfa622d699e9556e8 sold 0%, 0.28x) | door catches 3 of the 13 catchable scorecard misses since 1 Oct; 3 misses were already live trades (entry after halfway), filter doors cannot speed them | TRAP: next block exit pricing gives a fake +$3,348/day in half A from one coin (0x79f03c47d26a8c75bcae351804683c80facf53ac 62.56x next block vs 3.92x at the crossing swap); price exits at the crossing swap | KILL, FACT, TRAP
- [stated] FW 7 Oct 17:45 BKK (FW Analyst 1 J-001 exits, http://159.223.88.168/hx-c9ois/fw/lab/j001-exits.md FW-022) | EARLY BUY GAP IS THE COINS, NOT THE EXITS: tradable +10/+20 s arm with its best exit set: half B no top 3 $27/$46 a day at $10 vs $82 for +5 min with its best set; on hindsight winners stop 0.3x closes the no-top-3 gap ($84/$93 vs $82) but $1k still $1k to $2k/day below (thin first-second pools); trail 50% off peak kills every set (+5 min B $164 to $94) | side: mc5 stop 0.3x vs 0.5x half B +$10/day $10, +$199 $1k, worst 3 days minus $56 vs minus $90, but half A flat, not passed | RULE: early buying needs a picker that finds the +5 min winners at +10 s plus stop 0.3x; crew flags failed, PP score untested | FACT, KILL (exits as the fix)
- [stated] 7 Oct 17:40 BKK, Ben: "Nobody would buy $1,000 size... we would buy $10, $20, maybe $50 right after launch... There must be patterns... Price Predict... deep models." FW job J-002 (A1, sup-to-a1 v40): first 5 to 60 s entries at $10/$20/$50 only, picked by PP signals used only from the second they exist: obs.eye attention (never tested as entry), launchpad/deployer record rebuilt as of T (live boxes leak look ahead per PP handover), first seconds flow, first seconds buyer wallet quality (PP wallets start only at 300 s), combined score fit on half A. Bars: beat buy-all at same second AND mc5 +5 min on both halves, above random p95. Context: PP Gate A found nothing above $110 per $100 on honest fills.
- [stated] FW 7 Oct 17:50 BKK (FW Analyst 1 J-002, http://159.223.88.168/hx-c9ois/fw/lab/j002-score.md FW-023) | EARLY-SECOND SCORE KILLED 0 of 6: features known at +10 to +60 s (first-seconds flow, cap, launch LP, fee tier, kit, creator/funder-root record) fit on half A; on half B the top of the score makes +$1 to +$36/day at $10 vs +$106 to +$213 for buy-all in band; week A weights flip in week B | TRAP: Spearman with tied outcomes ranked by row order leaks time (fake rho 0.5 to 0.6 on growing counts like root_n); use tie-averaged ranks | untested: obs.eye attention, buyer wallet ids, crew-as-timing (card J-003) | KILL, TRAP
- 2026-10-07 17:33 BKK (SW Supervisor; A3.14 http://159.223.88.168/hx-c9ois/sw/sw-a3-14-early-exit.md, A3.15 http://159.223.88.168/hx-c9ois/sw/sw-a3-15-runner-exits.md) | KILL A3.14 early exit for SW losers: in-range LP Burn 20%+ fired 0 of 283 trades (LP leaves in one go, pull exit already sells that block); 3-block net sell burst loses half A at every X (5% -$4,977/day, cuts 40 future 5x coins) | KILL A3.15 age-scaled wider trail on kit hold: half A -$411 to -$2,555/day, wider trail sells runners lower (0x8e41fd25e85e624560fa50ff9f117a02cd30b8b3 29.15x to 23.49x) | LEAD: runners sell at about 0.4 of peak (29x of 74.5x), tighter trail far above 5x untested (A3.16) | the 16:27 shadow 'mismatch' was a page bug at engine restart (fm_mc5 did buy pool 0x644eac7020466c554369e0592637f24dd95fbb13), streak 3 of 5, autoswitch armed | SW Supervisor stood down per D-027, handover top of sw/sw-supervisor-status.md | KILL x2, LEAD, PROCESS
- 2026-10-07 17:40 BKK (SW Edge Hunter = Analyst 2, cards SW-H1 to H3 in http://159.223.88.168/hx-c9ois/cons/ideas.csv) | V4 $1M GATE CHEAP CHECK (research-box /opt/sw/v4/v4vec.db, true first swap 4 to 7 Oct, 8,737 pools): 46 gates age <=3 h (about 16 a day vs about 3 on v3) but only 5 hit 5x before 0.3x within the tape (11%) vs 30% for v3 live coins in the same 3 h; breakeven about 15% = all v4 dead cheap; all 5 hits are the a80 hook family (5 of 26, 19%), other hooks 0 of 20 | LONG TEST SW-H1b running research-box /root/sw-hunt/h1b.py (a80 vs ALL, tapes to gate + 48 h) | TRAP: in v4vec.db ETH INTO the pool = buy (18,157 of 18,169 price moves agree), is0 = 1 means token is currency0; reading ETH < 0 as a buy counts sells | pull exits are 5 of 70 live trades (all 0.00x), prize about $65 a day = not worth hunting | KILL (cheap), LEAD (running), TRAP
- [stated] 7 Oct 17:55 BKK, FW sup: EDGE ON PAPER, FORWARD. PP honest board (/opt/hist/analysis/scoreboard.txt, ppexit honest exits, pulls = -100%, picks since 5 Oct) per $200 trade: 30 s money book n 130 +$179 (best coin removed +$164; by day +268/+156/+53); aipaper (60 s AI top 5%) n 126 +$446; aipaper30 n 124 +$341; v5_top5 n 137 +$242; 10 s money book n 11 +$532; random rand_30s n 197 -$93. Sell session walk forward 21 Sep to 3 Oct, 1,141 picks: +$348. CORRECTION: "Gate A nothing above $110" is the retired 19 Sep trench system, not today's PP. FW A1 v41: leak audit, recompute at $10/$20/$50 (small size may reach pond wins), propose one live lane. Live switch is Ben's call. FW mistake owned: early entry was spec'd at $10 and $1k; Ben trades $10 to $50 only.
- [stated] FW 7 Oct 18:00 BKK (FW Analyst 1 J-003, http://159.223.88.168/hx-c9ois/fw/lab/j003-crew-timing.md FW-024) | CREW AS TIMING KILLED: exit on first token transfer out of creator or funder root fires on 80% of half B entries and cuts runners (+10 s B +$162 to minus $26/day at $10; mc5 +5 min +$164 to +$67) | obs.eye picker NOT TESTABLE on v3: PP collector forward.db saw 111 of 2,220 v3 envelope coins, from 20 Sep (61 by +10 s) | early buy on v3 via my lab universe closed; see FW sup 17:55 line for PP honest board leads | KILL