---
name: trading-edge-hub-5
description: Trading edge hub ledger part 5 (continues trading-edge-hub-4, which is near its cap): sell system v1 and the liquidity-pull finding, from 5 Oct 2026 13:15 BKK. Read with [[trading-edge-hub]] before any buy/sell call.
sources: [cowork]
---
# Trading edge hub, part 5 (follows [[trading-edge-hub-4]])

## 5 Oct 2026 ~13:15 BKK: SELL SYSTEM v1 (sell session; doc claude/sell-system-v1.md; droplet /opt/sellsys; status http://159.223.88.168/hx-c9ois/sellx/sell_status.txt)
- [stated] Ben (sell brief): this session owns only WHEN we sell coins bought in the first 30 s; goal most money per trade over all buys; paper only.
- 2026-10-05 | LIQUIDITY PULLS (ModifyLiquidity events, 4,407 launches 21 Sep to 3 Oct, walk-forward 30 s model picks) | top 5% picks: 87% have their pool liquidity removed within 24 h, median 7 min after our buy (p25 1.9, p75 18 min); 299 of 316 coins that printed 10x were pulled later; half the pulls happen within 2% of the peak, median 3 s after the last trade (no warning on the tape); no other pool for the token in 3,121 of 3,440 pulls (not a migration) | engine.py + liq.py (container /root/sell) | FACT
- 2026-10-05 | every paper engine (sim.py, aipick5 books, reconcile 24 h exit, qualrun "after 10x 40% go 30x") values a pulled coin at its last trade price = fake profit | honest re-score of live paper picks: 12 booked profits were pulled pools; after 10x only 15% reach 30x before the pull (not 40%) | liveeval.py | FACT (hand-off: pickers and reconcile should check liquidity before booking a 24 h exit)
- 2026-10-05 | baselines on top 5% picks with honest liquidity (n 1,141) | live rule (10x, stop -40%, 24 h) +1.44 per trade (+$288 on $200); hold 24 h -0.59; never sell -0.77; all out at 2x +0.34, 3x +0.64, 5x +1.15 | grid of 1,530 price rules: nothing beats 10x by more than noise; partial take profits, moon bags, trails, per launch type targets, creator-based targets all lose or tie | grid.py, pertype.py | FACT: sell everything at 10x, never hold for more
- 2026-10-05 | v1 = 10x OR nobody trades for 60 s OR 30 min after buy | top 5% +1.74 vs +1.44 (early +1.61 vs +1.16, late +1.85 vs +1.67), better on 10 to 13 of 13 days for top 2/5/10%, same with a 6 s sell delay; beat random-best rules out of sample in 3 of 4 splits | cand.py, grid2.py | LEAD (history); first 6 h live paper per coin: v1 -$52 vs live -$30 on 48 coins (tie within noise) | LIVE PAPER as sellsys (v1, v1s = v1 + stop, live rule side by side)
- 2026-10-05 | old kills re-tested with liquidity: "5 min no buyer" is +0.10 vs live (was killed on fake data); 60 s no trade is the best dead-coin cut; time stops on price (not 2x by N min) do not help | ideas.py | CORRECTED
- 2026-10-05 | locked-liquidity launch types (adders other than 0x58daec3116aae6d93017baaea7749052e8a04fa7 and 0xd0fd42bd8374ac659d3d81242d032211c2481055 never pulled) | the pickers almost never buy them (53 of 2,275 top 10%) and they lose under every rule (-0.16 to -0.38) | locked.py | FACT: no runner bag case
- 2026-10-05 | built sellsys.service (mirrors every aipaper5/aipaper30/aipaper pick from 6 h before first start, Swap + ModifyLiquidity logs every ~3 s, sell priced 2 s after the signal with impact, fee and real liquidity, miss diary 6 h watch, sellstat every 5 min) + sellguard (20 s timer + cron: heartbeat, every open position checked within 45 s, every buy mirrored, no v1 position past 35 min; restart, roll back to lkg, alertgate only after 30 min) + supervisor checks; freeze test: caught and restarted | LIVE (paper)

## 5 Oct 2026 13:20 BKK: prelaunch watch (cowork; doc claude/prelaunch-watch-v1.md; droplet /opt/prewatch)
- 2026-10-05 | prelaunch announcement vs outcome (weighted sample: all 507 coins >= $500k + 1,499 random = 15,111 clean launches 19 Sep..3 Oct; only evidence dated before the first pool) | 10x within days: all 2.1%, own new X account teased first 0.0% (~994), any prelaunch sign 1.3% (~3,147), project GitHub before launch 0.0% (~215); $500k noise test p 0.88 to 0.99 for every group | bt_ident.py | KILLED (no edge, slightly negative)
- 2026-10-05 | X tease net (6,601 RH launch tweets 14 Sep..4 Oct): first coin launched with a teased ticker hits $500k 1.5% vs 3.4% all (p 0.97); copycats grab teased tickers first | bt_xnet_an.py | KILLED as a buy signal; FACT: ticker match alone is a trap
- 2026-10-05 | GitHub name match (9,130 RH repos since 1 Jun vs 25,092 launches) | $500k 3.2% vs 3.4%, 10x 1.7% vs 2.4%, p 0.55 | bt_gh.py | KILLED
- 2026-10-05 | deployer with an earlier runner (22,391 creators known) | $500k 8.4% vs 3.2% but 10x 2.1% vs 3.7%; serial deployer with no runner 10x 0.2% | FACT: bigger launches, not 10x
- 2026-10-05 | DATA TRAP: mcap.db has 19 old tokens re-pooled in the window (PONS, Delta, A Meme Coin) that look like giant new runners; catbase t0 is not always the first pool (453 tokens had an earlier pool); mcap sell caps cover a median 6.5 min | /opt/prewatch/bt/first.json = true first launch | FACT
- 2026-10-05 | built prewatch.service (GitHub collector, X tease net OFF, watch list, matcher ~7 s after launch, COPY marking), paper books pw_watch / pw_strong via reconcile.py pw_books (note: sell session found reconcile 24 h exits ignore liquidity pulls, so judge these books on honest exits), pwguard (30 s timer + cron) + super; alerts OFF (/opt/prewatch/alerts_off) | LIVE (paper forward test, verdict ~12 Oct)

## 5 Oct 2026 ~13:30 BKK: second-0 signals inside the pond (launch session; pondsig.py)
- [stated] Ben: are there other signs to score 0.01% fee / 10 ETH launches, or do we blind buy them?
- 2026-10-05 | pond = 1,449 coins 21 Sep..3 Oct, outcome sellable 10x from the 60 s buy 8.1% | ALL pond tokens share ONE token code shape (one launch tool), and every pond creator wallet is new (creator history useless here) | FACT
- 2026-10-05 | second-0 tilts: name copied from a coin launched in the last 24 h 9.7% vs new name 5.2% (both halves 8.0/11.2 vs 4.8/5.8); native ETH pool 14.4% (late half only, n160) vs WETH 7.4%; name longer than 12 letters 11.3% (unstable halves); launch hour no clear effect | pondsig.py | LEAD (copy-name STRONGEST)
- 2026-10-05 | CORRECTION of my 11:10 BKK qualrun claim ("after 10x 40% go to 30x, sell half at 10x and trail the rest"): it valued pulled pools at their last trade; with liquidity pulls only 15% reach 30x and the sell session found sell-all at 10x beats every hold rule | FACT (sell session receipts above)

## 5 Oct 2026 ~14:20 BKK: big runners round 2, earliest signs (cowork; http://159.223.88.168/hx-c9ois/runners/runner_patterns_v2.md; doc claude/runner-patterns-v2-2026-10-05.md)
- [stated] Ben: 5 more iterations on first-minute identifiers and the next earliest signs of big runners; group $20M+; double-check the score; explain "outside the launch pages"; run the out-of-the-box stack; Price Predict covers the 30 s 10x game, this is about $20M runners.
- 2026-10-05 | CORRECTION of 12:00 runner study: runtrack lists only coins alive ~2 Oct (missed 98% of coins that hit a sellable $500k in 6 h) and counted coins that LAUNCHED big (MOW start ~$105M to $164M, ZYNOREK $50M) as runners; old milestone STRONG rule on clean data: 1 of 55 reached $20M (1.8%), not 17% | an7.py | CORRECTED
- 2026-10-05 | DATA GAP: 84,067 of 118,904 day-DB launch rows (71%) have NULL supply, 24,116 with swaps; mcap.py-style code drops them; share of runners in the gap: $1M 18%, $2M 24%, $5M 42%; Multicall3 0xcA11bde05977b3631167028862bE2a173976CA11 backfills 80k tokens in ~4 min | supfetch.py | FACT (proposed supply_backfill.py, side DB)
- 2026-10-05 | clean minute ladder, 44,575 small-start v4 launches 15 Sep..4 Oct | sellable in 6 h: $500k 695, $1M 225, $2M 95, $5M 19, $20M 1 (si/acc) | build.py | FACT
- 2026-10-05 | HOOK CODE FAMILIES at second 0 (fingerprint = eth_getCode length + md5 of the TRADING pool's hook, not mint_to) | CAMPAIGN_WETH 5059_7adb2ec7 (WETH, fee 3000, ~3/day): 86% $500k, 48% $2M; 99.6 to 99.9% of swaps via one private router per coin; public routers bought but never sold in the main pool (PANDAKEY 61 buys/0 sells in 24 h; CHIPINU, PPORT, NFLOAT, RBPERP, STKDOG 0 sells); CMC-vote bots + "DexScreener Paid" posts on X | BUNDLE_ETH 341_6de2bb44 (native ETH, start ~$2.4k, ~7/day): launch tx buys through ~40 wallets (~$27k in 2 s), public buys and sells work, from 30 s entry median 2.5x (q75 4.9x), peak ~14 min | family_check.py | FACT (CAMPAIGN = AVOID, BUNDLE = fast flip only)
- 2026-10-05 | Price Predict paper books PANDAKEY (campaign family) at +8.8x (aipaper, aipaper5); its `pad` reads the mint_to pool hook, which misses this family | ppcheck | OPEN (hand-off: re-check that win, add trading-pool hook family, block campaign family)
- 2026-10-05 | organic launches, walk-forward by day | $1M later: AUC 0.83 to 0.91 at 30 s..60 min, top-3/day hit 14% (30 s) to 28% (60 min); $2M+: AUC 0.45 to 0.70 (no signal, ~1/day); 100+ buyers in 5 min 7x lift for $1M; alumni wallets AUC 0.78 to 0.80 for $1M but 0.45 for $2M; $2M+ coins have fewer alumni (median 0 vs 20) | an4.py, vets.py | FACT
- 2026-10-05 | X around sellable $500k crossing (695 coins, twitterapi CA search) | 96% of organic crossers have zero CA posts before; AUC 0.52 to 0.55; campaign family has X posts before in 45% (paid promo marker) | xq2.py | FACT (X is not a fast early signal)
- 2026-10-05 | 7-day follow-up of 640 sellable-$500k crossers (Gecko hourly, main pool) | organic: $1M 18%, $2M 4%, $5M 0.6%, $20M 0.2%; every $5M+ fast coin peaked within hours (7 d peak = 6 h peak, 11 of 11) | an7.py | FACT
- 2026-10-05 | where $20M+ comes from (48 runtrack coins with 2nd-highest hourly close >= $20M) | 27 already >= $2M at first candle, 11 started $200k to $2M, 10 ran from < $200k taking 15 to 799 h (median ~4.6 days); in-window big ones on venues the v4 tape misses: Uniswap v3 (SC, HI, SIF, SI), Pons V2 dex (AGE), bridged TIBBIR; late ignition on a new pool (Agrippa launched 17 Sep, ran 24 Sep on bankr) | rt + gecko | FACT: $20M is a multi-day, multi-venue game; launch minute is the 10x game
- 2026-10-05 | proposed (not deployed): family_check.py (tested: PANDAKEY flagged), supply_backfill.py, "ignition watch" = all venues, any token crossing sellable $250k, hourly outsider buys/sells per router, holders, alumni, X, collect ~4 weeks unbiased before scoring | PROPOSED (Ben hands over)

## 5 Oct 2026 16:45 BKK: phase 2 scoping + Price Predict data fix brief (100x session; brief http://159.223.88.168/hx-c9ois/hx/pp-datafix-brief.md, project doc claude/pricepredict-datafix-brief-2026-10-05.md)
- [stated] Ben: 100x and "$2M runner" are both the wrong frame; wants "second phase runners" scoped (not launch seconds; maybe from $300k or $500k), where it starts and ends, plus input for the Price Predict session to fix the missing data.
- 2026-10-05 | CORRECTION of 100x session v2 numbers: "27% of first $500k closes reach sellable $2M" and "41% from $0.5M" used runtrack as a complete universe; runtrack is not complete (runner patterns session: it misses most coins that hit a sellable $500k), so both are overstated; clean organic 7-day follow-up says $500k to $2M about 4% | CORRECTED
- 2026-10-05 | CORRECTION: "190 coins/day hit $300k in hour 1" counted price prints without money; prints in thin pools fake levels (prints say 26/day close above $500k vs 5.6/day with $40k hourly volume in runtrack) | CORRECTED (every level needs a money test)
- 2026-10-05 | minute data gaps (receipts in the brief): 23 of 57 runtrack $2M+ coins launched since 15 Sep have no swaps in the day files (mostly plain v4 pools and unknown pads, e.g. COUPONS 0xf086c53f2780e601f0d005fabb3ee5a220336062 pool with 0 swaps); multi-pool tokens; ~28% non-ETH quotes without USD; swaps only 0-6 h, bars start 6 to 325 h after launch (median 35 h); transfers only 30 min; ETH price table stale since 3 Oct 20:16 BKK | FACT
- 2026-10-05 | proposed phase 2 scope: start = first sellable $250k after hour 1, all venues (same as the proposed ignition watch); watch 7 days; win = sellable $2M, big win $5M, jackpot $20M; split fast ignition (peaks in hours) from slow build (days, new pool or venue) | PROPOSED

## 5 Oct 2026 14:35 BKK: Mac sequencer feed coverage + self-heal (launch session)
- 2026-10-05 | feed vs launch ledger since feed start (~12:40 BKK) | pond 0.01% launches caught 1 of 1 (100%); other launches 163 of 208 (78%); the misses are dynamic-fee hook pools (fee 8388608) and fee-0 pools created inside launchpad contracts (initialize is an internal call, not in tx calldata) | feedcmp.py coverage lines + feedcov.json | FACT
- 2026-10-05 | pond launch rate collapsed today: 16 to 29 an hour on 4 Oct afternoon/evening, 0 to 3 an hour since ~06:00 BKK 5 Oct (ledger); rule_fee001_first booked 1 of 1 pond launches since aipick5 restart 12:19 BKK | reconcile.db launches | FACT (few pond picks = few pond launches, not a feed bug)
- 2026-10-05 | feed self-heal layers: feedwatch reconnect loop; relay.sh restarts it if dead or heartbeat >120 s; launchd com.kookwatch.feedrelay KeepAlive (survives kill/reboot); launchd com.kookwatch.feedguard every 120 s kickstarts relay if missing or hb >240 s; sentinel C8 alerts Ben only if hb >15 min (all Mac layers failed) or pond coverage <80% on 5+ launches | kill tests: feedwatch back in 10 s, relay back in <15 s | LIVE

## 5 Oct 2026 ~10:30 BKK: basket rotation, coin results of the live book (rotation session; dashboard http://159.223.88.168/rotation.html, doc claude/systems-inside-baskets-2026-10-04.md)
- 2026-10-05 | replay of the rotation book 19 Aug to 4 Oct (11 trades, 3 slots), coins = picks v6 (5 biggest liquid) bought at the BUY close, coin exits (half at 5x, after 2x sell on a 07:00 close 40% under top), CMC prices | book +35% (first 3 coins +52%) vs basket line book +2% vs BTC +25% (69.3k to 86.5k) | per trade: BNB AI 11 Sep to 4 Oct +45% (best +95%, AKE), Base infra 3 to 9 Sep +22%, Solana infra 20 to 26 Aug +14%, Base AI +8%, ETH AI +3%, Base DeFi 21 Aug to 11 Sep +1%, Solana AI 19 Aug to 1 Sep +6%, Solana memes top 15 26 Aug to 10 Sep -20%; open: TON since 9 Sep +6%, Bonk.fun since 11 Sep +11% (best +65%, 1 coin) | /tmp/claude-0/replay.py | FACT (small sample; one trade made most of the gain; trims not included)
- [stated] Ben (5 Oct): skeptical the baskets print money; nearly fully in Robinhood Chain coins and not willing to sell them near what looks like a bottom.
- 2026-10-05 | RH bottom read 15:10 BKK (chain.json, gauge.json, chains.json rh baskets) | coin index 78% under 3 Sep top after 32 days (wave 1: -64% over 33 days, 16 Jul to 18 Aug); 29% under its 20d avg (needs ~+41% to reclaim = HEATING buy); flat since 2 Oct; buyers vs sellers 24h 1.24; active wallets 7d 353k vs 60d low 244k (not washed out); fees 0.56x of 28d; gauge WAIT, bottom 73, trigger 1/3; AI basket at a new low today, others 4 to 25% above their lows; Ben's 2X rule (pump 18 Aug to 3 Sep = 16 days) puts the low by ~5 Oct; launchpad base rate puts the activity bottom day 55 to 90 from the 6 Sep peak = 31 Oct to 5 Dec | READ: price low close or in, not confirmed; if one more leg, about 15 to 30% more by late Oct; confirm = index above its 20d avg with BTC above 50d | OPEN (score 9 Oct and 31 Oct)
- [stated] Ben (5 Oct): asked how far the RH bottom is in % and days, or whether it is already in.
- 2026-10-05 | Ben's "low within 0.8x the pump time (most), within 2x (~90%)" rule, tested on CMC daily coins (cap >= $5M, vol >= $0.5M), zigzag: pump = swing low to high +100%, then 40%+ drop, low = next swing low confirmed by a 50% bounce | 2023-26 (4,202 swings): low within 0.8x 33%, 1x 42%, 2x 62%, 3x 71%, 5x 82%, median 1.4x; 2019-22 (10,685): 46% / 56% / 73% / 81% / 89%, median 1.0x | by depth: 40-60% drop 2x covers 80-85% (median 0.5-0.8x); 60-80% 61-71% (1.0-1.5x); 80%+ 37-51% (median 2.0-3.3x) | short pumps (<= 7 days) take longer: median 2-3x, 2x covers 45-56%; long pumps (60+ days) 2x covers 89-97% | lows that never got a 50% bounce are not counted, so true coverage is lower | /home/claude/swing/zz.py, zz2.py | FACT: 2x is about a 2 in 3 rule, not 90%; scale the multiple by depth; RH wave 1 (9 day pump, 33 day fall, -64%) was 3.7x; wave 2 (16 day pump, -78%, day 32 now) fits a 2 to 5x window = 5 Oct to 22 Nov
- [stated] Ben (5 Oct): asked whether the low typically comes within 2x the pump time across chains and coins, and what the best rule is.

## 5 Oct 2026 ~16:10 BKK: coins that reach $20M within a week (cowork; http://159.223.88.168/hx-c9ois/runners/r20/runners_20m_week.md; doc claude/runners-20m-week-2026-10-05.md)
- [stated] Ben: wants a detailed breakdown of $20M runners incl. ones needing up to a week (okay with multi-day), all 48 with contracts, patterns at 30 s, 15 min, 1 h, 4 h+ vs non-runners, exclude straight scams; idea: buy at ~20 s (existing), then a deep check confirms high odds and slows the selling (keep more for later).
- 2026-10-05 | CORRECTION of 14:20 line "27 of 48 already >= $2M at first candle": runtrack keeps only ~41 days of hourly candles, so older coins looked "launched big"; with GeckoTerminal first-pool candles and the clock at the first $5k-volume hour: 54 coins reached a confirmed $20M; 26 started under $1M and hit $20M within 7 d (22 under $50k), 11 started $1M+ or hit $20M in hour 1, 17 needed more than 7 d | paths.py | CORRECTED
- 2026-10-05 | timeline of the 26 small starters | median $1M 1.5 h, $5M 6.5 h, $20M 22 h; p75 to $20M 75 h, p90 131 h; 11 of 26 had a 50%+ dip on the way | paths.py | FACT
- 2026-10-05 | runners vs coins that stalled at $1M to $5M (same tracker) | 30 s nothing (AUC ~0.5); 30 min top-10 holder share 3.6% vs 14.8% (AUC 0.33, lower = runner), holders 482 vs 174; 15 min cap $388k vs $106k (0.67), volume $209k vs $62k (0.70); 1 h $749k vs $148k (0.71); 4 h $1.5M vs $355k (0.74); 8 h $3.8M vs $477k (0.89), still at its high (0.71), churn volume/cap 1.0 vs 2.1 (0.32); X authors h4-24 24 vs 2 (0.70); holders 958 at 4 h to 2,982 at 24 h vs stall ~840 (n9) | f30.py, hstage.py, xstage.py | FACT (tracker = survivor sample)
- 2026-10-05 | hour checkpoints (603 tracker coins with visible start, 20 runners) | 8 h at $3M+: 29% reach $20M in 7 d (12/20 caught); 8 h $2M+: 25% (16/20); 4 h $1M+: 10% (12/20); "at its high" or low churn add little; clean v4 tape: still $3M+ at 6 h 1 of 5 | hstage.py | LEAD (survivor-biased; build an unbiased stage sample)
- 2026-10-05 | venue of the small starters | pons_v2 12/25, unknown 6, o1_rwa 2; share of $1M+ coins reaching $20M by launchpad: unknown 16%, bankr 9%, v3 8%, pons_v2 6.5%, longxyz 2%, plain v4 0 of 32; USDG quote 8/25 vs 24% of stallers; native ETH 3/25 vs 29% | f30.py | FACT
- 2026-10-05 | scam check of the 54 (public-router sells first 2 h, 3 biggest pools) | ZYNOREK 0x6a49a7dbb4e9e401054c9d164f22f5c2e75d820a opened at $50M with 0 public trades; STANDARD 0x88ad8ddf1e3898412146a534538d418c6f8a9062 7,568 public buys vs 41 sells | probe2.py | FACT (both SUSPECT, excluded)
- 2026-10-05 | operator table (v4 launches 15 Sep..4 Oct, small start) | hook family 17916_7bdd2b2a 4/day, 10x from 60 s 12%, tradeable; 12193_d2d5b6ba 11/day, 12%, tradeable; factory family 25533_3080c1dd 331/day ~0 (junk); 1,081 of 1,238 serial creators (5+ launches) never had a $500k coin; bundle family: back near ~$6k by 6 h, liquidity removed within 8 h in 4 of 6 checked | probe.py | FACT
- 2026-10-05 | proposed (not deployed): stage watcher for every 30 s pick and every $500k coin on any venue at 15 min, 1 h, 4 h, 8 h, 24 h (cap and close/high, volume/cap, holders + top-10, X authors and 10k+ accounts, liquidity lock) -> SELL NORMALLY / KEEP SMALL BAG / KEEP BIGGER BAG; bags only where liquidity cannot be pulled | PROPOSED

## 5 Oct 2026 ~16:00 BKK: other venues live paper (venuepick) + data layer (launch / Price Predict session; detail in [[pricepredict-data]])
- [stated] Ben: get the launches outside the v4 stream too (other exchanges, bridged, coins that ran later on a new pool); this session owns the database and all data collection.
- 2026-10-05 | GeckoTerminal top 200 RH pools by 24 h volume: Uniswap v3 59 pools $337M vs v4 97 pools $341M; new pools per hour ~97 Pons V2 curve, 51 v4, 28 v3, 15 Bankr (v4), 9 v2 | gecko API | FACT (v3 is as big as v4 by volume, mostly established coins)
- 2026-10-05 | venuepick blind buy every non-v4 launch at 30 s, sell v1, real liquidity, first ~1.5 h: v3 59 closed, 55 at -100% (pool liquidity() 0, ~0 WETH left = pulled, checked on chain); v2 126 closed avg -13%, 1 hit 10x; Pons curve 86 closed avg -14% | venues.db | EARLY (tiny sample; v3 launches look like rug factories)
- 2026-10-05 | whole chain pool census since block 0: 2.21M pools, 1.67M tokens; one token often has many pools (Aria 55, WORM 1,517 incl. Pons) | tokens.db birth | FACT (per-pool data understates volume and misprices; build token-level bars)

## 5 Oct 2026 16:20 BKK: PHASE 2 process v1 (phase 2 session = owner; doc http://159.223.88.168/hx-c9ois/hx/phase2-process.md, project claude/phase2-process-v1.md)
- [stated] Ben: phase 2 owned by this session; gate = $250k AND really traded (buys and sells); two shapes, quick (a couple of hours) and slow (up to ~45 days, happy to wait); wins = sellable $2M, $5M, $20M; at $250k a process gives odds to reach $2M; needs help defining sell some at $2M / keep some on signs; not live yet, collect and analyze now.
- 2026-10-05 | gate on v4 tape (first $100+ sell at cap >= $250k, 15 Sep..4 Oct) | 261/day raw, 132/day new launches without campaign family; sellable ($20k buys at target, first 6 h of pool) $1M 5.1%, $2M 1.7%, $5M 0.19%, $20M 0.04% | trig.py, q2.py (container /tmp/p2) | FACT
- 2026-10-05 | family at the gate (hook code length) | BUNDLE (341) ~6/day: 30% sellable $2M, median 16 min; plain no-hook ~115/day 0.1%; other hooks 1.3%; campaign skipped | FACT
- 2026-10-05 | money test buy at gate +5% slip, sell at sellable $2M or -50%, 3% cost | bundle 1.31x learn / 1.30x test (n 91/42); plain 0.70x; other hooks 1.01/0.43 | TESTING (bundle liquidity pulls within 8 h in 4 of 6 checked by the 20M study, so sell at target, never hold)
- 2026-10-05 | non-bundle signs at the gate (target $1M) | how far past $250k before the first real sell AUC 0.72/0.68; low sell share before the gate better both halves; buyer counts flip | LEAD
- 2026-10-05 | slow shape | Gecko hourly follow-up of a random 376 gated coins (all venues, up to 20 days): sellable $2M 1.1%, all within hours, no slow wins in sample; 15 of 375 trade mainly on Pons V2 or pools.trade (outside the v4 tape) | FACT (slow wins rarer than 1 in 300; need collector + backfill)
- 2026-10-05 | built /opt/phase2/p2collect.py (GeckoTerminal new/top/trending pools every 5 min, gate = cap >= $250k + 3 buys + 3 sells + $5k h1 volume, follow every pool 5 min to 6 h, 30 min to day 3, 2 h to day 45; new-pool check for late ignition) + p2guard.py; systemd timers + cron + flock; alertgate only after 3 failed fixes; first run 99 seen, 26 gated | LIVE (data only, no alerts)
- 2026-10-05 | sell plan v1 (untested) | at sellable $2M sell 75%, keep 25% only if new buyers/hour, absorbed sells, holders and X real people keep rising; bag stop -40% from peak or 6 h with no buyers; at $5M sell half the bag | PROPOSED (first weekly scoring 12 Oct)

## 5 Oct 2026 ~16:40 BKK: phase 2 follow-up (phase 2 session)
- [stated] Ben: use the Price Predict database read-only, never edit it, prefer no own database; questioned bundle sellability and the "shoots past $250k before any sell" sign.
- 2026-10-05 | bundle family sellable? (128 gated bundle coins 15 Sep..4 Oct) | median 192 distinct sellers in the first 30 min, 91% had 10+ sellers, none had 0 to 1; gate reached at ~19 s, entry cap median $458k, so sellable $2M is ~4.4x, hit 30% in ~15 min; the +30% per trade is after the 70% that hit the -50% stop or fade | FACT (selling works; liquidity pull risk still unchecked)
- 2026-10-05 | CORRECTION: "blows through $250k before the first real sell" is not a usable non-bundle sign; median non-bundle gate is at 1.04x $250k and only 1% had their first real sell above $400k (the AUC came from a handful of coins) | KILLED
- 2026-10-05 | phase 2 collector p2collect/p2guard STOPPED (timers disabled, cron removed) and GeckoTerminal backfill workers stopped; source of truth = Price Predict data layer read-only (tokens.db birth, supply, chainswaps every swap/liquidity/sync log of every pool from 15 Sep, coming token minute bars); /opt/phase2/p2.db left as is, not used | DONE

## 5 Oct 2026 ~18:30 BKK: Price Predict review, honest 10x and scams (launch-improvement session, read only; doc claude/pp-review-scams-2026-10-05.md, public http://159.223.88.168/hx-c9ois/brx/pp-review-scams.md)
- [stated] Ben: this session never writes a picker; it only reviews Price Predict and proposes improvements; goal = catch 10x from the point coins are buyable, and they must not be scams; report data needs and inconsistencies.
- 2026-10-05 | chain re-score of every paper pick (784 picks, 312 coins; 226 coins 6 h+ old; Swap + ModifyLiquidity from public RPC) | honest 10x before the pull 11% of coins (books 8 to 18%) vs booked 11 to 24%; 82% pulled before 10x or stop, median 7.5 min after buy; 25 of 29 booked "24h" exits (avg +202%) were pulled pools = 0; 371 open picks sit on pulled pools; honest avg per pick -0.4 to +0.8 vs booked +1.1 to +4.1 (closed only) | evalpp.py (container /root/pp) | FACT
- 2026-10-05 | rugs are the norm (14,158 launches 21 Sep..3 Oct) | 97% creator-held LP via position manager 0x58daec3116aae6d93017baaea7749052e8a04fa7; 87% pulled within 24 h, pond 100%; pulls take >=90% in one go, never re-added, trading stops; 97% of pulls by the launch wallet; no warning (median 1.8 s after last trade, no fee collect first); runners (500k + 10x from 5 s) all peaked before the pull, pull ~10 min after peak; locked-LP launches (3%) run 0.9% | hist1.py, pullchk.py, rmtx.py | FACT: "not a scam" must mean sellable via public routers and sold before the pull, not "no rug coins"
- 2026-10-05 | private-router honeypots booked as wins (fee 0.3% WETH family): 5 of 24 honest-10x coins had public buys 13 to 52 and 0 public sells, >99.5% of trades via one private router | FACT: gate = public buys but no public sells + mostly private router -> skip (flags 4 coins at 30 s, 3 honeypots)
- 2026-10-05 | clean 10x by type in PP picks | pond 22% (n50), fee 0.05% 11% (n98), fee 1% 0 of 29; pool liquidity >10 ETH at buy 19% vs 8% (both halves 24/13, 14/4), pull later (16 vs 8 min); book entry realistic (+0.8% median at buy +3 s) | evalpp | LEAD
- 2026-10-05 | data inconsistencies: aipaper.db and aipaper30.db have no dec_b and aipaper.db model is NULL (fills not auditable); 85 catch-up picks booked on pools already pulled before decision; public RPC has no historical state, Blockscout blocks the container (Cloudflare), so creator funder/operator clustering needs an archive RPC or Blockscout key | FACT (needs Ben: archive RPC or Blockscout API key)

## 5 Oct 2026 ~16:30 BKK: chain lifecycle phases (chain brain session; detail in [[chain-brain]])
- [stated] Ben: chain-wide heat/top, a lifecycle (bottom turn, slow wallet growth, parabolic = take profit, top turn), and a "more downside likely" read; Price Predict DB read-only.
- 2026-10-05 | phases on 12 chains (Artemis DAU/PRICE/FEES 2020-26) + Solana memes 2024-26, 25% drop before 30% rise in 60 d | FALLING (cheap, under 20d, fees <0.8x or users <0.9x of 28d) 43% / 51%; BASING 35 / 39%; CLIMBING 23% (10 of 12 chains up); HOT (fees 1.5x + 20% stretched) rose 30% first 49% / 66% | phase.py, t4.py (container /home/claude/cb/multi) | FACT: falling activity while cheap = more downside likely; HOT = trim, not exit
- 2026-10-05 | "parabolic wallets = top" | 12 chains: no edge (44% rise first); Sol memes new users 1.5x = +57% next month; RH wallets lagged price top by 12 days and spiked on an airdrop | KILLED (fees lead, wallets lag)
- 2026-10-05 | RH 5 Oct: FALLING since 18 Sep (index 0.77 then, 0.42 now); daily launches 12k to 14k = same as the Aug bottom | gauge.py v2 live (Telegram 08:47 BKK) | LIVE
- 2026-10-05 16:35 BKK | supervisor: first minute plan v1 (http://159.223.88.168/hx-c9ois/firstmin/first-minute-plan.md) | If a first-minute buy is to be kept as a runner bag, then only on a pool that cannot be pulled; pullable pools sell 100% on v1 | receipt: 87% of top picks pulled (median 7 min), moon bag -$125/trade vs sell-all at 10x, first 30 s AUC ~0.5 for $20M vs hour 8 AUC 0.89, 12 of 25 small-start $20M runners on pons_v2 vs 0 of 32 plain v4 | TESTING (steps: honest scoreboard, can-it-be-pulled flag, speed curve, venue coverage, hold engine, real money pilot needs Ben's yes)

## 5 Oct 2026 ~16:45 BKK: sell session, take profit framework v2 (study only)
- [stated] Ben: the sell session only studies and builds the take-profit system; nothing runs live or in paper until he says so. sellsys/sellguard/sellstat and the 7 Oct scheduled check were stopped and removed (13:30 BKK); code files only in /opt/sellsys.
- 2026-10-05 | LP OWNER AT BUY TIME (v4 PositionManager 0x58daec3116aae6d93017baaea7749052e8a04fa7 NFT, owner from its Transfer logs for the tokenId in the first ModifyLiquidity salt) | 60 of 60 pulled coins: LP owned by the creator wallet (EOA); 74 of 80 kept coins: LP in a contract (top 3: 0xeff166aaf189323c58dc27ed1206eb2c37faacdf, 0x267444d099b10fb5ed7c3cc7b7c767adca574952, 0x35c5c1037eef9eb07b2826d537e4fdae61878852) or burned. All 3,895 coins: CREATOR lane pulled 97% (early 96%, late 97%), LOCKED contract lane 8%, burned 0% | lpown2.py, it5.py | STRONG (known at second 0)
- 2026-10-05 | exits by lane (top 15% picks) | CREATOR (2,776 picks): 10x + no trade 60 s + 30 min +0.99 per trade vs live rule +0.64, both halves; bags after 10x lose (25% bag +6.45 vs all at 10x +7.87 on 10x coins). LOCKED (334 picks): about 0 for every rule, 60 s rule hurts locked 10x coins (+3.47 vs +7.88), bag vs all at 10x a tie (n 21) | it5.py | FACT (CREATOR), LEAD (LOCKED)
- 2026-10-05 | big runners vs our picks | v4 30 s picks almost never reach $2M before the pull (0.5% of top 5%, $5M 0); runtrack $500k crossers (499, mostly pons_v2) show 36% to $2M and hold beats sell, BUT 75% were first seen 11+ days after crossing (survivor sample) | it1.py to it3.py | CORRECTED: runtrack hold results not trusted; rerun on Price Predict chainswaps (all venues since 15 Sep, ~5 to 13% done 5 Oct) and the phase 2 collector
- Framework v2 doc: claude/sell-system-v1.md, public http://159.223.88.168/hx-c9ois/sellx/sell-system-v1.md

## 5 Oct 2026 ~16:50 BKK: chain gauge v3, meme phases (chain brain session; detail in [[chain-brain]])
- 2026-10-05 | CORRECTION of my 16:30 "RH FALLING = more downside likely": that came from 12 chain coins (trend-following); on 5 meme markets (CMC meme caps ETH/BNB/Base/Solana 2021-26 + Solana panel) the FALLING phase had no edge (43% vs base 44%). Memes mean-revert | it5.py, it5b.py (container /home/claude/cb/multi) | CORRECTED
- 2026-10-05 | meme phases, 25% drop first vs 30% rise first in 60 d | STRETCHED 30%+ over 20d avg 59/40 (halves 55, 63) = take profit; DEEP STOPPED (60%+ under 90d high, under 20d, fees 7d >= 0.8x 28d) 30/66 (halves 39, 25) = buy, scale in; DEEP BLEEDING (fees < 0.8x) 40/59, Solana 73% drop first = get ready | it5b.py | FACT (5 series; summed caps inflate rises)
- 2026-10-05 | DATA TRAP: runtrack RH index before late Aug is survivor-biased (15 July crossings of 879 tracked) | tokens.csv | FACT (fix = Price Predict all-swap backfill)
- 2026-10-05 | RH now DEEP BLEEDING (fees 56% of normal); buy trigger = fees back to 80% of their 28 day average while still deep | gauge.py v3 live | LIVE

## 5 Oct 2026 ~17:30 BKK: HOLD OR SELL AT $250k (phase 2 session; http://159.223.88.168/hx-c9ois/hx/hold-or-sell-250k.md, project claude/hold-or-sell-250k-v1.md)
- [stated] Ben: real flow = buy at a couple of seconds, then at $250k decide hold or get out; build on DeepCheck / stage watcher; run phase 2 tests on PricePredict data (read only).
- 2026-10-05 | honest paths (every Swap + ModifyLiquidity of each gated pool from start to 5 Oct, sell priced with pool depth, $3k bag, 1% fee), 2,133 v4 coins gated 15 Sep..2 Oct, learn <= 24 Sep (564) / test (1,569) | plain no-hook (1,795): 95% LP pulled within 72 h, median 5 min after the gate (p10 71 s, p25 3 min, p75 9 min); sell at gate 0.94/0.88 of value, best hold rule 0.43/0.15, hold 72 h ~0 | sim.py, look.py (container /tmp/p3) | FACT: sell at $250k within 60 s (+60 s keeps 0.89, +5 min 0.57 test)
- 2026-10-05 | main launcher 0x58daec3116aae6d93017baaea7749052e8a04fa7 LP NFT at the gate | sent to 0x000000000000000000000000000000000000dead before gate 18 coins: 0 pulled; not sent 1,744: 1,726 pulled (99%); but locked coins still fade (24 coins: sell 0.93, hold 72 h 0.04, best trail 0.70, 1 of 24 reached sellable $1M) | nft2.py | FACT: locked is not a hold reason on v4
- 2026-10-05 | BUNDLE (121): hold to sellable $1.5M, out if cap < 70% of gate or liquidity -50%, max 6 h = 1.42 vs sell 0.91 (learn 1.58, test 0.99 to 1.08 vs 0.92); $1M/stop 50% hold minus sell +0.37 (90% CI +0.21..+0.53) | stage check alive (price >= gate and liquidity >= 80%) at 1/5/15 min: bundle hold beats sell both halves (test 1.28 vs 1.11, 1.72 vs 1.40, 2.36 vs 2.07), even at 1 h; plain and other hooks alive: selling wins every checkpoint | bgrid.py, stage.py | LEAD (edge shrinking, paper only)
- 2026-10-05 | process v1: hook family at second 0; non-bundle sell 100% at $250k within 60 s; bundle hold with target/stop/pull guard and 1/5/15 min alive checks; weekly re-test | the slow-runner hold question needs other venues (Pons V2, v3, pools.trade) via PP chainswaps (~10% done at 16:15 BKK) | PROPOSED
- 2026-10-05 | asks for PricePredict (PP builds): finish chainswaps; per-pool liquidity table (adds/removes, share removed, sender+salt, depth after); LP lock table (launcher NFT owner history, dead-address time; v2/v3 LP burns/lockers); hook family column; bundle seller wallets after 30 min | OPEN

## 5 Oct 2026 ~17:10 BKK: first-minute plan executed (Price Predict session; no live test)
- [stated] Ben: execute the first minute plan except the real-money pilot; also review pp-review-scams and check every point on live data before changing anything.
- 2026-10-05 | step 1 honest board (/opt/hist/honestboard.py, on top of scoreboard.txt; exits by ppexit.service = copy of the sell session's v1 engine, because sellsys was stopped ~16:00 BKK) | live picks 07:35..15:55 BKK, distinct coins: v1 +$26 per $200 (78 coins, 7 at 10x, 16 pulled), old rule -$3 (55 of 74 pulled); paper books under old pricing claimed +$972 | FACT (tiny sample)
- 2026-10-05 | honest by book: model books +$200 to +$1,276 per trade (n 3 to 21, shared winners); rule_smart2_clean -$96 (n41); rule_fee001_first (pond blind buy) -$35, 0 of 16 at 10x, 11 pulled; random baseline book rand_30s added | FACT (n small)
- 2026-10-05 | where the honest profit comes from: fee 0.3% WETH coins 5, 4 at 10x, +$1,403 per trade; fee 0.05% 38 coins -$138; pond fee 0.01% 18 coins +$19 (1 at 10x); fee 1% 11 coins -$53 (1 at 10x) | revcheck.py | FACT (n small)
- 2026-10-05 | step 2 pull flag (chainswaps v4 ModifyLiquidity, pools with full cover) | adder = PositionManager 0x58daec3116aae6d93017baaea7749052e8a04fa7: 46% pulled within 10 h (17,912 pools, median 7.2 min); every hook/launchpad adder ~0% (0x4e3468951d49f2eea976ed0d6e75ffcb44a9a544 5,235 pools 0 pulled; 0x6f02324d20cc679d0e585290caa6b16bacbc0f77 997 0) | pullflag.py | FACT; live flag _pullable in aipaper5 scored feats + venues.db launches.pullable (pons 0, v2/v3 1, v4 by adder) | LIVE
- 2026-10-05 | step 3 speed curve (walk-forward 30 s model, 17 days, v1 exits, no pull data so the level is high, slope is the point) | top 1%: buy at 1 s +$1,048, 10 s +$1,006, 20 s +$956, 30 s +$868, 60 s +$614; top 5%: +$691 / +$654 / +$601 / +$537 / +$375 | speedcurve.py | FACT: 30 s -> 10 s is worth about +15%, after 30 s value falls fast; decision: keep 20 s stage, test a 10 s number model next (picks at 10 s carry less information, so the real gain is smaller than the curve)
- 2026-10-05 | step 8 kill list LIVE in aipick5 (gate5.py): campaign hook 5059_7adb2ec7, junk factory 25533_3080c1dd, serial creators (1,206 of 1,308 creators with 5+ launches in 20 days never had a real run = sellable $500k and 5x from entry), token ending 777; bundle family 341_6de2bb44 -> bundle_flip book only; every skip logged "KILL skip" with the books it would have bought | LIVE (measure what the kill list removes)
- 2026-10-05 | review point CHECK, private-router honeypots (fee 0.3% WETH) | all 5 example coins have SELLS through public, verified aggregators: RobinHoodSettler 0x6aa80dbbed9ae5ab45fbf61f9644fada3b29326e (40 to 70% of trades, buys and sells), UniversalRouter 0x204faca1764b154221e35c0d20abb3c525710498 (37 sells on 0x53d025fbaee3057622d7ea0817b0b81b6645af66), UniversalUniswapV4Adapter 0x36dc95f1f088e11c0066dc19173f745655002bc2; only the main UniversalRouter 0x8876789976decbfcbbbe364623c63652db8c0904 shows buys without sells (1 to 3 buys) | routers.py + Blockscout names | NOT CONFIRMED: no sellability gate added (it would remove the only profitable group)
- 2026-10-05 | review point CHECK, "pond first, 22% clean 10x" | live honest: pond blind book 0 of 16 at 10x, -$35; pond coins in model books 1 of 18 | NOT CONFIRMED live (small n); "skip fee 1%": live 1 of 11 at 10x, -$53 | WEAK, not applied yet
- 2026-10-05 | fomo trending: FomoScan /v2/leaderboard/tokens/trending is free (0 CU) but its snapshot did not change in 11 h (captured 05:41 BKK); fomo app gets trending live over its WebSocket (topic "trending", REST /proxy/trendingTokens needs the logged-in session) | built ~/fomotrend (Playwright, own profile, Ben logs in once, records frames, ships every minute) | WAITING for Ben's login
- 2026-10-05 | chain backfills: ETH in the L1 bridge 0xDf8755334ce7A73cCF6b581C02eA649AE3E864b3 per day since launch (Blockscout ETH balance history, 36,574 changes): net +124,860 ETH Jul, +87,540 Aug, +84,114 Sep, -9,186 Oct 1..5; 287,823 ETH locked now | chainday.db bridge_eth | FACT; stablecoins on RH (USDG 0x5fc5360d0400a0fd4f2af552add042d716f1d168, USDe 0x5d3a1ff2b6bab83b63cd9ad0787074081a52ef34, U 0xce24439f2d9c6a2289f741120fe202248b666666, rwaUSDi 0x9000123cb195804f35cee9553e2596978227910f) rebuilt from mint/burn, matches live supply exactly: $18M 15 May, $103M 15 Jun, $246M 7 Jul, $635M 15 Aug, $1.06B 15 Sep, $1.10B 5 Oct (DefiLlama within 3%) | chainday.db stables_day | FACT; daily swaps/traders 30 Apr..15 Sep: scans running (Mac) | OPEN
- 2026-10-05 17:15 BKK | supervisor: break-even hit rate | If a launch book's 10x hit rate is under ~5.5% (one 10x ~+$1,700 vs ~-$100 per v1 miss, $200 trade), then it loses by construction; money book = model top 1/top 2 only, wide rule books control only | receipt: honest board 5 Oct 07:35 to 17:07: rule_smart2_clean 1/41 -$96, fee001_first 0/16 -$35; aipaper 6/10 +$1,020, v5_top1 2/4 +$1,100, ev_top1 2/3 +$1,276; all books mixed +$26 | TESTING (tiny n, needs ~100 distinct top coins + random book; note http://159.223.88.168/hx-c9ois/firstmin/next-steps.md)

## 5 Oct 2026 ~19:30 BKK: rug-crew (operator) check with paid dRPC (launch-improvement session, read only; container /root/pp cr1..cr5.py, op1.py)
- 2026-10-05 | creator wallet of 14,122 launches 21 Sep..3 Oct (LP add tx.from) -> 9,297 creators; first funding block by archive balance search; funder = sender of that ETH; ~265k calls (~$1.60) | 84% direct funder, 16% funded inside a contract call | FACT
- 2026-10-05 | pond creators (0.01% fee, 10 ETH): 99% funded inside a cross-chain relay call in the same minute as launch (1,154 via 0x5b8d85ebabf17cf6b67bfa2fe6795623951cd70e, 267 via 0xf341c935599822f1bc1ec29e27606724fbbf6595), every creator fresh, no shared funder | FACT: the operator's money comes from another chain, invisible on RH
- 2026-10-05 | 10x before the pull (print max from 5 s price), non-pond base 12.0% | exchange-like funder (nonce >5,000, 30% of launches) 5.9%; operator 3+ earlier launches with 20%+ 10x 24.1% but unstable (early 15.0%, late 26.2%, n212); 3+ launches none ran 11.0%; creator funded <1 h before 20.7% vs 7 d+ 5.6%, mostly launch type (inside fee 1%: 11.7% vs 7.7%) | op1.py | KILLED as a main signal; small tilt only (fresh creator in fee 1% launches)
- 2026-10-05 | CORRECTION of my 18:30 "private-router honeypot" claim: in the main pool only one router (0xc955efd62944fcea581194ba722db08592865b8b, ~10k buys and ~9.8k sells) trades both ways, but Price Predict found public sells for all 5 example coins through other pools/aggregators (RobinHoodSettler 0x6aa80dbbed9ae5ab45fbf61f9644fada3b29326e, UniversalRouter 0x204faca1764b154221e35c0d20abb3c525710498) | NOT CONFIRMED: drop the sellability gate as proposed; my check looked at one pool only

## 5 Oct 2026 ~17:50 BKK: chain gauge v4, true-holdings labels (chain brain session; detail in [[chain-brain]])
- 2026-10-05 | CORRECTION of my 16:50 v3 lines: meme "stretched = take profit 59%" and "deep stopped = buy 30/66" were an artefact of summed CMC caps (coins leaving the volume filter look like crashes); with true holdings labels stretched 42% vs normal 41%, deep stopped 38% | memes/fixed.py, multi/it7.py, it10.py | CORRECTED (no reliable daily top signal for memes)
- 2026-10-05 | deep (60%+ under 90d high) + under 20d avg + fees 7d < 0.8x 28d | 52% another 25% drop first vs 41% normal, both halves, Solana 84%; matches 12 chain coins FALLING 43% vs 29% | it10.py | FACT: more downside likely while fees still shrink
- 2026-10-05 | RH bridge (Price Predict chainday.db, read only) | first net ETH outflows ever since 27 Sep (-10k to -13k ETH/wk); at the Aug low inflow slowed to 7.6k/wk but stayed positive; inflow peaks came 0 to 3 days after price tops | FACT (n=2, receipt only)

## 5 Oct 2026 ~17:45 BKK: next-steps executed (Price Predict session; no live test)
- [stated] Ben: barely profitable because wide paper books buy almost everything; model best picks win big; execute next-steps.md, live comes last.
- 2026-10-05 | money book = books ending _top1/_top2 only, distinct coins, v1 honest (honestboard.py MONEY BOOK block, per day, best coin removed) | 07:35..17:28 BKK: money n24, 4 at 10x (16.7%), +$349 per trade, best coin removed +$288; control (all other books) n100, 7 at 10x, +$30, best removed +$9 | FACT (n24, one day; trust at ~100 coins)
- 2026-10-05 | random baselines in aipick5: rand_30s (5% of scored launches, started 17:28) + rand_gated (5% of launches that pass the kill gate, started 17:35) | LIVE (result ~7 Oct)
- 2026-10-05 | fomo trending: real source = POST https://prod-api.fomo.family/proxy/trendingTokens (logged-in session, limit <= 100); returns ~53 tokens, ALL Solana (networkId 1399811149), no Robinhood coins, chain params ignored | fomotrend.py polls it every 60 s from the logged-in headless page (needs a normal Chrome user agent; headless UA gets "Failed to fetch"), ~/fomotrend/trend.jsonl shipped to /opt/pricepredict-data/fomotrend/ | LIVE; the "pop after trending" test can only run on Solana coins
- 2026-10-05 | liqlock.py (paid dRPC): whole-chain v4 Initialize + ModifyLiquidity + PositionManager NFT transfers -> /opt/pricepredict-data/liqlock.db tables liq (every add/pull, share removed, owner at that time), lock (NFT id, owner history, dead-address and burn time), poolfam (hook size+md5+family), bundle_sellers (wallet of every bundle-family sell after 30 min) | BUILDING (answers the hold-or-sell session's asks 2 to 5); chainlogs parse crash on look-alike events fixed, paused while liqlock runs

## 5 Oct 2026 ~18:30 BKK: RH bottom clock + index fix (chain brain session; detail in [[chain-brain]])
- 2026-10-05 | CORRECTION of "RH coins -97% under the 90d high" (and the 15:10 "-78%"): equal weight log index exaggerates; CMC cap weighted: 25 biggest coins at the 3 Sep top are -67% as a basket, median coin -65%, AI -47%, CASHCAT -44% | memes/rh_cmc.pkl | CORRECTED
- 2026-10-05 | chain-wide cycle clock: 14 crashes on 12 chains + pump.fun (fees 7d down to 25% of the 60d peak, 15 to 50 d after it, price already -35%+) | price low median 62 d after that day (p25 44, p75 115), median another -40% (random day -25%), low within 10 d in 1 of 14 | multi/cyc.py | FACT; RH crash day 29 Sep -> low window ~12 Nov to ~30 Nov (late case 22 Jan); live in gauge
- 2026-10-05 17:51 to 18:12 BKK | supervisor: GMGN speed on RH launches | If a launch is bought through GMGN (gmgn-cli swap --chain robinhood), then GMGN can route it within seconds of pool init (first 9: 1.1 to 5.2 s; full run 24 quoted: median 1.4 s, p90 3.3 s, max 6.3 s); FREE TIER IP LIMIT: polling ~4 tokens at once got RATE_LIMIT_BANNED after ~5 min (12 of 37 probes lost), so the live bot must call only on picks, never poll; free tier also needs $100 balance in GMGN wallets, checked on an hourly snapshot, so it is not the bottleneck for a 10 s / 30 s entry; first quote hit HTTP 401 40101600 in 3 of 9, next try worked; GMGN fee 1% per trade (docs) | receipt: ~/feedwatch/gmlat.py on Ben's Mac, 9 fresh launches (v4 + Pons V2), router 0x65050a9b7e5075a2ba5ced7b1b64ee66262c40dc | TESTING (swap-to-inclusion time unmeasured until first real $1 trade)

## 5 Oct 2026 ~19:00 BKK: RH catalyst signal (chain brain session; doc claude/out-of-the-box-thinking.md)
- 2026-10-05 | new app taking chain fee share (>=3% now, <=1% 28 d ago, DefiLlama RH breakdown) | Pons V2 fired 10 Aug, 8 d before the 18 Aug low, flat ~$1.6M/wk 10 days, then $10M/wk; Fables firing since 25 Sep at $1.5M/wk, flat 9 days | LEAD (n=1); ignition = weekly fees 2x and >= $1M, live in gauge
- 2026-10-05 | new chain survival, 19 chains: fees -80% from launch peak -> only 6 regained half the peak within 180 d | FACT: ~1 in 3 risk RH fades instead of a new wave

## 5 Oct 2026 ~18:45 BKK: sell session, Pons study (study only; doc claude/sell-system-v1.md v3)
- [stated] Ben: urgent to fix that the 30 s picker never buys the big runners; focus on Pons; still mainly the 10x game but let some big runners run; nothing live until he says so.
- 2026-10-05 | WHY PICKER MISSES BIG RUNNERS | aipick5 scores only tokens minted straight into a v4 pool; Pons tokens launch on a curve (factory 0xe33e9e479df8802cb0866d5d05258bec4cf62948) and graduate via 0x7ed598bcef8bd9edd8c97a195c6d13f40801ec7e into a native ETH v4 pool, hook 0xe5e702641ea86f4ae6cc3cdaed2b886f976be044; neither is scored (AGE never scored) | FACT
- 2026-10-05 | Pons: 90,810 launches 13 Sep to 5 Oct, 1.26% graduate (median 9 min, 25% within 52 s); graduated pools pulled 0 of 1,708; blind buy at graduation +30 s: 10x 8.3%, $2M 2.8%, EV -21% (all at 10x) | FACT (HOLD lane, needs a filter)
- 2026-10-05 | graduation picker top 10% (n 114, walk-forward) | all at 10x +13%/trade; stop -50% + half at 10x + trail 50% +62% (halves +71%/+48%) | LEAD (paper first)
- 2026-10-05 | Pons at $250k ($3k bag): sell now 0.77, trail/bag 0.81 to 0.82; 11% reach $2M | FACT: holding is free optionality on Pons, opposite of v4 plain coins
- 2026-10-05 | 30 s curve picker on tape counts only: full path 10x top 1% 7.5% vs base 0.8% | LEAD: needs wallet features

## 5 Oct 2026 ~21:00 BKK: POND CREWS (review session; http://159.223.88.168/hx-c9ois/brx/pond-crews.md, project claude/pond-crews-2026-10-05.md)
- 2026-10-05 | 1,421 pond-type creators funded via peel chains of throwaway wallets (nonce ~5) through send-to-many contracts 0x5b8d85ebabf17cf6b67bfa2fe6795623951cd70e and 0xf341c935599822f1bc1ec29e27606724fbbf6595 (no events, trace only); traced 20 hops (~95k calls) -> 11 crews | crews graded week 1, tested week 2: good 24.5% sellable $500k runners (n212) vs bad 6.2% (n340), all 13.0%; best 0xcb7277454416d90e7fe8799cd300744d5797de03 17%/26%, worst 0xae3d565892c79fa8fc9243b380283dd419e28ec9 3%/5% | hop2.py (container /root/pp) | STRONG (13 days; propose live crew index for Price Predict)