---
name: trading-edge-hub-2
description: Continuation of trading-edge-hub (first file full). Ledger of tested RH runner ideas from round 6 on; read with [[trading-edge-hub]].
sources: [cowork]
aliases: [edge hub 2, trading hub continuation]
---
# Round 6 (3 Oct 2026): GMGN API + outside the box, rule-1 coins (109, 13 winners)
- [stated] Ben: new warnings count as minus points, dev keeping his coins as a plus; run as parallel shadow scores, merge later only if they win on forward data.
- [stated] Ben: always keep a tracker that checks each process and fixes it within an hour.
- [stated] Ben: do not buy Birdeye; GMGN (already paid in the stack) is the source.
# Round 7 (3 Oct 2026): chain tape, 50 ideas
- [stated] asked to log every trading insight here (standing): round 7 kept "nobody sold in first 12 s" (+), "sells >= 39% of buys before entry" (-), busy minute (+, weak); score-7 shadow live in rhfast; all tape exits killed (stale-price artifact: price exits at first print AFTER the signal). Full receipts: project doc claude/runner-ideas-round7.md.
- [stated] Ben expects holder count and buy vs sell volume tracked every 10 s; rhfast tick table does this since 3 Oct 18:15 BKK.
# Basket rotation, round 4 (3 Oct 2026 night)
- 2026-10-03 | cowork: hold vs rotate | Buy and hold of the CMC baskets 1 Jan 2025 to 30 Sep 2026: median -86% (BTC -12%); rule v4.2 rotation +18%/+51% per year | x50 bench | PROVEN (descriptive)
- 2026-10-03 | cowork: holding the strongest baskets every day (re-ranked daily, no BUY-day rule) | 2025 -30 to -70%: strong baskets are near tops and pull back hard | x50 round 2 | KILLED
- 2026-10-03 | cowork: BUY MORE (Turtle pyramid, tested baskets) | Add the same amount once at +15% above the buy before extended, then sell 75% at extended: book +1221% vs +980%, every year better, steps 10 to 15% all win; adding and selling only 50% loses | edge/basket-rule-v43-2026-10-03 | LIVE (rule v4.3)
- 2026-10-03 | cowork: 52 expert ideas on baskets | Faber 200d, Weinstein, Minervini, Clenow 100d/gap, Turtle skip-after-win (lower drop -18% but less return), BTC dominance, memes-after-large, equity curve and drawdown sizing, breadth-thrust leverage (2023-24 only), vol target, correlation cap, 4 slots: none beat v4.2 in both halves | x50_round4.txt | KILLED
# Round 8 (3 Oct 2026)
- [stated] asked to log every trading insight here (standing): round 8 kept "6+ buys in a row with no sell right before entry" as a minus (score-8 shadow live); copying past-winner wallets 5 s later loses (-4%); the Unicorn wallets never buy in the first minutes; second-chance entry at 2 min loses. Receipts: project doc claude/runner-ideas-round8.md.
- [stated] Ben requires the live system to have no downtime, instant alerts, instant fixes and a redundant checker (livewatch 5 s + livewatch2 20 s + super 5 min + Mac offbox, 3 Oct 2026).
# Community at the pullback (3 Oct 2026 evening, Bottom Lab session)
- [stated] Ben: main goal = pullbacks that lead to NEW ATHs (not bounces). Hypothesis: "trader coins" go quiet and come back when profit returns; "community coins" (cult forms in hours, days or weeks) stay active through 50 to 80% pullbacks, and an active community there is a strong sign.
- 2026-10-03 | cowork: COMEBACK + crowd still active | If a COMEBACK signal fires and real people (bots removed, contract + $TICKER + replies) still post at 50%+ of their ATH-day rate, then new ATH before -50% stop 46% vs 28% when quieter (77 vs 136 signals; early half 42 vs 29, late 48 vs 28; RH 55 vs 38, SOL 40 vs 25) | xsafe windows, x/cbx.pkl | TESTING (strong candidate filter)
- 2026-10-03 | cowork: crowd active at any dip, by depth | All first crosses of 50/65/80% under ATH (965 dips, ~280 coins): 50-65% down + crowd active 38% new ATH vs 21-29% quieter (both halves); 65-80% 27% vs 19-26%; 80%+ REVERSES: active 12% vs quiet 24% (still loud at -80% = hopium). Crowd size at the top does not help by itself; human-led top (low bot share) 28% vs 23% | dip_raw.jsonl, dipt2.py | TESTING
- 2026-10-03 | cowork: COMEBACK crowd mood | Mood read at the signal changes nothing (depression 29%, optimism 36%, euphoric 44%); activity level matters, mood does not | [[buy-framework]] v5 | TESTING
# Deep dive window (3 Oct 2026 night, shortlist pipeline session)
- [stated] Ben: two windows: (1) first run, seconds, trenches process; (2) a slower window for a full deep dive (website, tech, holder growth, more); asked which window is best (30 min, 1h, 12h, 1d, 3d).
- 2026-10-03 | cowork: window 2 timing | If a coin first tops $1M, drops 50%, then makes no new low for 12 to 24h (base), then deep dive it there: base cap median ~$1M, reached a median 1.5 to 3 days after the first $1M; from the base 5x+ 15%, 10x+ 7%, new ATH 32 to 35%; waiting for a 72h base cuts 5x+ to 10% (MOO entry $9.2M vs $2.9M); ~8 such bases a day (RH 3.4, SOL 4.7). Clock windows (1h to 7d since launch) are worse: half the $20M runners already peaked by 6h. Only 16 of 82 $20M+ runners gave a 5x+ entry after a 50% dip; the rest ran straight up (lane 1 only) | runtrack export 836 coins with launch history, 14d future required; universe survivor biased (filled 2 Oct from top-now lists), so hit rates are high side; examples MOO, OTC, CLANKER, APE, BTC (sol), GME, FATCOIN | TESTING
# SI scoring (3 Oct 2026 20:50 BKK, cowork)
- 2026-10-03 | cowork: SI floor-break SELL scored | The 2 Oct SELL at ~$35M (floor-break rule) was followed by the buy-back trigger (2 4h closes above $36M) at the 2 Oct 15:00 and 19:00 BKK closes ($44.6M, $43.9M); price 3 Oct 20:44 BKK $47M. Sell-then-rebuy cost ~25% vs holding; the 2 Oct 11:00 wick to $29M (below the $32.3M pocket) closed back at $33.9M, so the golden-pocket hold read was right and the floor-break sell was wrong on n=1 | GeckoTerminal 4h pool AG1EhPjsBrkxQWkViuLXF6pQnHwfr9f9sdNneeoykcEJ | TESTING (floor-break rule now 0 for 1)
- 2026-10-03 | cowork: SI structure at 20:44 BKK | higher lows $29.0M > $33.1M > $35.0M > $38.6M > $41.5M > $43.9M vs lower highs $68.9M > $60.2M > $53.1M > $52.1M > $50.4M (squeeze); buyscore v3.2 WAIT (33% under top); real people on contract 8 in 24h, mood calm/skeptical, no euphoria; 24h txns 3,400 buys vs 6,575 sells; volume per 4h falling $2.05M to $343k | n=1 | TESTING
# Fast new ATH, 50+ ideas (3 Oct 2026 21:00 BKK, Bottom Lab session; doc claude/fast-ath-ideas.md)
- [stated] Ben: new ATH within 30 days is not good enough; wants pullbacks that make a new ATH within a few days; doubts ATH under $3M matters; fewer alerts fine; NO alerts while building (Bottom Lab muted via /opt/bottomlab/MUTE, lab.py md5 55716a71cc1b5935a93b50abdf91bc33).
- 2026-10-03 | cowork: FAST FLUSH | If a coin falls 50%+ under its ATH within 2 days of the ATH, RSI 4h is still >= 40 and 24h volume is >= 72% of the ATH day, then new ATH (4h close) within 7 days 31% vs 16% base (3d 19 vs 9, 14d 37 vs 21); early 28 / late 32, RH 31 / SOL 29, now-dead coins 31 vs 10, older coins 27 vs 9; median 3 days to the ATH | 263 events, 150 coins, ideas.pkl / combo.py | TESTING (best candidate; survivor-risk universe)
- 2026-10-03 | cowork: fast-ATH killers | Slow drop (5+ days from ATH) 4% new ATH in 7d vs 19%; coin 30+ days old 9% vs 19% | same set | TESTING
- 2026-10-03 | cowork: no edge for fast ATH | crowd >= 50% or louder than top day (+2 to +3, CI crosses 0), KOLs posting, returning posters, website live, app/product link, docs, GitHub, team posting, X followers: none hold in all halves; ATH under $3M helps inside FAST FLUSH (36% vs 24%) but small-cap effects failed off-list elsewhere, so optional | test50.py | KILLED as standalone signals
# Round 9 (3 Oct 2026)
- [stated] asked to log every trading insight here (standing): strong wallets make +14%/trade at their price but price jumps ~13% within 3 s, so copying at 3 s loses (-7%), at 1 s +7%; psych points (all 3 socials -2, JPEG logo, 04-07 BKK, lowercase name, minute :00 minus; US hours, past-winner logo, no socials plus) split 10x rate 1.7% vs 6.8% out of sample; logo colors, copied pictures and one-syllable names show no edge. Score-9 shadow live. Receipts: project doc claude/runner-ideas-round9.md.
# Basket rotation, v4.4 (3 Oct 2026 night; doc claude/basket-rotation-study.md)
- [stated] Ben: 2025 was a bear market, check past bull markets (DeFi summer, 2021, 2024-25); expects pullbacks to keep happening and wants to rotate into strength instead of sitting through 50 to 66% pullbacks; extension = euphoria; wants breakout day, extended and sell-on-turn defined and tested; labels SELL 25% / SELL 75% and "double your investment".
- 2026-10-03 | cowork: out of sample bull test | Rules on Oct 2019 to Sep 2022 point in time baskets (never used to design): v4.1 +1188%, v4.4 +2168% vs BTC +125% (BTC worst -72%, rules -24 to -27%) | bull/bk_old.json | PROVEN
- 2026-10-03 | cowork: run length | Breakout to top: 2025-26 median 15 days (+17%), 2023-24 27 days (+37%), 2020-21 bull 50+ days (+76%); next 30 days after the top median -23 to -30% in every period | runlen.py | PROVEN (descriptive)
- 2026-10-03 | cowork: extended and turn down | Sell only 25% at the first HOT close (75% if doubled) and arm the 20% trailing stop only after +30%: beats v4.3 in all 4 halves of both eras (18 of 27 neighbour settings); Fear and Greed as an extension trigger flips with the threshold; wait-for-bounce exits deepen 2023-26 drops | it20*.py | LIVE (rule v4.4)
- 2026-10-03 | cowork: money chasing brand-new coins | Volume share of coins listed under 30 days (top 1000) in its top fifth: next 4 weeks worse in all 4 periods 2019-2026 (new coins -7 to -20%, alts -2 to -12%); [CORRECTED 4 Oct: the Solana launchpad share claim was wrong, it only covered Dec 2024; full 2024 high launchpad share was followed by BETTER Solana memes (+23 vs +2% next 14d), 2025 mixed, 2026 better]; no meme buys when the share is in its top 20% of the year: 3 of 4 halves better | age.py, trench_share.py | LIVE (meme mania gate)
# Foundation audit (3 Oct 2026 21:00 BKK)
- [stated] Ben asked for a 5-iteration audit of every number before building further. Corrections (receipts in project doc claude/audit-2026-10-03.md): stops really fill at median -56% not -40% (rule-1 EV +36% -> +18%); "win" label was result>0, only 230 of 1,364 broad "wins" are true 10x; rule-1 score EVs (score-4..9) sit inside the noise of 250 random signs, so unproven until forward trades; best take-profit = all out at 10x, no initials, no moonbag.
# Basket rotation, v4.5 (4 Oct 2026)
- [stated] Ben: wants each run and correction counted per bull market (on-chain baskets, not BTC), the trench mania effect checked chain-wide, 50 more breakout ideas, 50 bottom-confirmation ideas (structure, MAs, sentiment) in 3 rounds, and the extended/sell rules re-checked on the baskets.
- 2026-10-04 | cowork: run catalog | Per basket runs (30%+ up, 20%+ correction): median length DeFi summer 2020 158d, 2021 bull 41d (98 runs), 2023-24 28d (22), Nov 2024-Jan 2025 14d (36), 2025-26 chop 26d (81); correction to the low median 15 to 35 days (-33 to -46%); back to a new high median 31 to 54 days, many never | cycles/runs.py | PROVEN (descriptive)
- 2026-10-04 | cowork: all-chain new coin share is market wide | Top fifth of weeks: 2025-26 next 28d SOL coin -11 vs 0, majors -10 vs 0, memes -21 vs -10, apps -15 vs -3; 2023-24 mixed | chainwide.py | TESTING (gate live)
- 2026-10-04 | cowork: 50 breakout ideas | Break of structure, MACD, RSI 60/65/70, golden crosses, Bollinger, volume/fear filters, ATH-only, chain-first, base length, weekly breakouts: none beat the 30d-high close (20d in hot markets) in both eras | r50.py | KILLED
- 2026-10-04 | cowork: BUY (BOTTOM) | After a 40%+ fall from the 90d high: higher low, then close over the last swing high (6 day swings), BTC above 50d: 2019-22 +3370% vs +2168%, 2023-26 +1901% vs +1914%, worst drop -24% both; standalone bottom systems far weaker than breakouts (best +384% / +1773%) | r50b.py, r50c.py | LIVE (rule v4.5)
- 2026-10-04 | cowork: exits re-check | 10 exit and 4 extended variants per basket type: live rule best overall; every exit keeps about 78 to 83% of the best close on winners (give-back 17 to 22%); rule profitable on memes, apps/AI and majors in both eras | exits10.py | PROVEN
# Basket rotation, round J (3 Oct 2026 late, Bangkok; raw http://159.223.88.168/hx-c9ois/edge/basket-ideas-roundJ-2026-10-03/)
- [stated] Ben: whole-market warning banner when money chases brand-new coins; extended label "Take % Profit. (25% if not add, 75% if you doubled)"; two exits = trailing stop 20% and close under the 15 day low; 3 iterations on new coin money; 30 ideas each on breakout day, bottoms, extended, profit taking.
- 2026-10-03 | cowork: new coin money is real and market wide | Weekly top-1000 volume share of coins listed under 30 days in its top 20% of the prior year (point in time): all 2019-26, ON minus OFF next 4 weeks top 10 alts -9% (p 0.01), top 100 alts -11% (p 0.00), memes -18% (p 0.02); BTC -4% (p 0.09). Same with under 60 days, top 500/2000; under 7 days and coin counts show nothing | nc/it1.py, circular shift test | PROVEN
- 2026-10-03 | cowork: what the new coins are | Big new launches and airdrops (UNI, SUSHI, CRV, APE, APT, ARB, SEI, TRUMP), not small memes; new coin $ volume +310% in 4 weeks while older coins only -9%; gap grows to -19 points by 8 to 12 weeks; strongest after a pump or flat month, nothing after a fall | nc/it2.py | PROVEN (descriptive)
- 2026-10-03 | cowork: new coin money in the trading rule | Meme block beats none (+1901 vs +1503%, +3370 vs +3248%); blocking ALL buys worse (+1811%, +2643%); trim or tighter exits when ON mixed | it_nc3.py | KILLED (beyond the live meme block)
- 2026-10-03 | cowork: 120 ideas (breakout day, bottoms, extended, profit taking) | 7 beat v4.5 in both eras; sizing wins are only more money in (plain 1.5x beats every conditional 1.5x); selling less at extended wins in bulls and loses in 2025-26 chop with no regime filter that fixes it (20/70: +2059% / +4011% but 2025-26 half worse); nothing shipped | it_j*.py | KILLED
- 2026-10-03 | cowork: DE-RISK MEMES when money chases new launches | While new coin volume share is in its top 20%, sell open meme baskets on a close 10% under the best close: 2023-26 +2256% vs +1901%, 2019-22 +4427% vs +3370%, worst drop -24% both; 7 to 15% trails, 3 to 5 day lows and top 10/20/30% thresholds all better in both eras; only 7 trades change in 2023-26 (6 better), so small sample | it_k*.py, edge/basket-rule-v47-2026-10-03 | LIVE (rule v4.7)
- 2026-10-03 | cowork: Nov 2024 to Jan 2025 runs | 36 runs across 10 baskets (about 3 to 4 each, median 14 days); 16 of those 36 tops were never beaten again through Sep 2026 | cycles/runs.py | PROVEN (descriptive)
# Sentiment before new-ATH breakouts (3 Oct 2026 23:50 BKK, cowork; doc claude/breakout-sentiment-study.md)
- [stated] Ben: social analysis so far looks best for tops and maybe bottoms, not breakouts; asked to test the hourly ticker crowd in the 10h before real vs fake new-ATH breakouts on every coin in the database.
- 2026-10-03 | cowork: crowd before breakouts | If the ticker crowd ($TICKER or name, xsafe) in the 10h before a breakout is bigger, growing, rushing in the final 3h, full of new voices or big accounts, then real vs fake breakout odds do NOT change: all within -4 to +5 pts of the 40% base, CIs cross 0, signs flip between halves | 789 breakouts (317 real = +30% before -15% within 72h, 472 fake), 383 coins, runtrack hourly, /opt/bostudy | KILLED
- 2026-10-03 | cowork: mood before breakouts | Mood read blind (500 events): BELIEF in half of all breakouts, real or fake; euphoria 36% vs 42% (-6, CI crosses 0), flat 53% (n=15), cooling crowd +5, heating -5; none stable | /tmp/bo mood reads | KILLED as a filter (hint only: flat/cooling slightly better, euphoric slightly worse)
- 2026-10-03 | cowork: price context at breakouts | Fresher old ATH (fewer hours since) -9 pts for the stale third [-16,-2], deeper pullback -7 [-14,-1], bigger cap -8 [-16,-2], older coin -7; volume ramp and last-10h return add nothing | same events | TESTING
# Basket rotation, cycles and single coins (3 Oct 2026 late, cowork; raw http://159.223.88.168/hx-c9ois/edge/coins-vs-baskets-2026-10-03/)
- [stated] Ben: cycles start when BTC bottoms and last about 2 years; the Nov 2024 to Jan 2025 "bull" framing was wrong; he trades single coins, not baskets, and asked whether the basket rule applies to coins.
- 2026-10-03 | cowork: runs by BTC cycle (CORRECTS the period split of the 4 Oct run catalog) | Bull 1 (BTC low Mar 2020 to top Nov 2021): 8.7 runs per basket, median 43 days, +87%, pullback after a run median -36%, 19 of 122 tops never beaten; Bull 2 (BTC low 21 Nov 2022 to top 6 Oct 2025): 8.5 runs per basket, median 18 days, +60%, pullback -42%, 39 of 110 never beaten; bear phases about 3 runs per basket | cycles/cycles2.py | PROVEN (descriptive)
- 2026-10-03 | cowork: coins pull back deeper than baskets | Bull 2 coins (point in time basket members): pullback after a run median -47%, 44% deeper than -50%; ones that later made a new high median -37%, 81% gave back 61.8%+ of the run; baskets -42% | coins/study.py | PROVEN (descriptive)
- 2026-10-03 | cowork: basket rule on single coins | Coins on their OWN signals (rule v4.7, book of 5): 2023-26 -47% (worst -80%) vs baskets +1257%; 2020-22 +1721%; wider trails (30 to 50%) and longer lows do not fix 2023-26. Basket signal, coin execution (buy the basket's coins on basket BUY, sell on basket SELL): 2023-26 all coins equal +1270% (-35%), 3 strongest by 7d +1342% (-44%), single strongest +1295% (-81%) | coins/study2.py | PROVEN (time coins by the basket, spread over 3+ coins)
# Pullback to new ATH on a FAIR list (3 Oct 2026 23:15 BKK, cowork; doc claude/pullback-ath-study.md section 4)
- [stated] Ben: evaluate 50 more ideas for buying pullbacks that go back to new ATHs, test on the tokens we looked at, implement the best.
- 2026-10-03 | cowork: FAIR LIST built | Every Pons V2 graduation from chain logs (9,986 coins, 4 Aug to 3 Oct, dead included) + GMGN ATH + 1h candles: 857 coins closed above $250k, 1,556 pullback events. Use this, not runtrack, for any "does it recover" test (fw/univ_pons.json.gz, fw/pons_grads.json) | grads.py, univ3.py | PROVEN (data)
- 2026-10-03 | cowork: 50 new ideas | None pass FDR on the fair list; every wait-for-confirmation entry (EMA reclaim, +30% bounce, 24h high break, base, ignition volume, RSI divergence, capitulation wick, double bottom) did the same or worse than buying the first touch; scale-in, 3d time stop, exit at 90% of ATH: no gain. Plain first touch: new ATH 14d 35/25/17/9% at 50/60/70/80% down, 2x trade +7 to +11% (CI touches 0) | t50.py, a50.py | KILLED
- 2026-10-03 | cowork: first-round winners re-run fair | young coin, small ATH cap, fast run, near $500k: all near zero on the fair list (were survivor bias) | univ100_res.pkl | KILLED
- 2026-10-03 | cowork: FAST FLUSH re-test on fair list (my approximation: 50% drop within 48h of ATH, RSI 4h >= 40, 24h vol >= 72% of ATH day, 1h data) | new ATH 7d 27% vs 25% others (3d 24 vs 20, 14d 29 vs 28); 906 of 1,144 Pons pullbacks qualify, so it barely filters there | ff.py | TESTING (not confirmed off-list)
- 2026-10-03 | cowork: two leads left | COMEBACK rule fair +4 pts (CI -2..+11), other lists +18, runtrack +1; BASE HELD (price gave back 75% or less of its whole run) fair 29 vs 9% at 50% down, 22 vs 9 at 60, 16 vs 6 at 70 (confirm half held, FDR 0.14, not seen on other lists); both at 50-60% down 31% new ATH, 2x +7% (97 events); neither 13%, -20% | buyscore v3.4 md5 58fa26f7601ea02202cedd5e5eb473b3 prints a NEW-ATH ODDS line | TESTING (paper only)
# Sentiment DURING and 12h AFTER new-ATH breakouts (4 Oct 2026 00:40 BKK, cowork; doc claude/breakout-sentiment-study.md part 2)
- [stated] Ben: add the 12h after the breakout; goal is to avoid failed ATH breakouts and find the big runners (fail, sideways, small, medium, big).
- 2026-10-04 | cowork: crowd after the breakout vs class | Inside price bands at T+12h: new people posting the ticker in 12h FAIL -16 [-22,-10], BIG 3x+ +12 [+6,+16]; big accounts (10k+) FAIL -16, BIG +10; all halves same sign. As an entry at T+12h (WIN +50% before -30%) counts add nothing | 402 breakouts (279 RH, 123 SOL), /opt/bostudy bo_anal3.py | TESTING (confirms a held breakout; not an entry signal)
- 2026-10-04 | cowork: don't buy the noisy first hours | Loud first 3h then buy at T+3h: -30% first +8 to +11 pts; loud first 6h then buy at T+6h: +11 to +16 pts (posts, new people, big accounts), CIs above 0 | bo_six.py | TESTING
- 2026-10-04 | cowork: hot then COOLING crowd | Mood read blind (286): hot first 6h then calmer hours 6 to 12: FAIL -14 [-22,-7], BIG +10 [+3,+19], entry at T+12h WIN +13 [+3,+24]; crowd HEATING into hours 6 to 12: WIN -11 [-21,-1]; new people still arriving hours 6 to 12: BIG +11 [+4,+18] | aftmood reads, single reader | TESTING (best candidate; needs forward test)
- 2026-10-04 | cowork: X credit drain | Shared twitterapi.io balance fell 3.66M to 0.72M on 3 Oct 21:45 to 00:30 BKK at 22k to 49k credits/min, mostly NOT this study (~4k/min); caller not visible on the droplet (likely another cloud session) | xstatus.json | open issue
- [stated] 3 Oct 2026 late, Ben: rotation is the name of the game; use the basket signal on coins; wants objective criteria for WHICH coins to buy inside a risk-on basket, from a much wider pool than the 15 basket coins.
- 2026-10-03 | cowork: coin picks inside a basket BUY | Pool = up to 50 coins of the basket's category ($100k+ volume); 20 features, 510 scores, both cycles. Pick the 3 coins up the LEAST over 90 days plus listed the LONGEST: 2019-22 halves +1686/+136% vs random 3 median +691/+113%; 2023-26 +612/+214% vs +277/+17%; worst drop -25 to -31% vs -44 to -52%. 3 hottest by 7d: great 2020-21, -51% in 2025-26. Hit rate of catching a top-20% runner only 23% vs 20%, so the edge is mostly fewer blow-ups | pick/feat.py, grid.py, halves.py, edge/coin-picks-2026-10-03 | LIVE (picks.py, shown on risk-on cards and BUY alerts)
# RH chain history rebuild, first daily read (3 Oct 2026 23:45 BKK, cowork; doc claude/ops-standard.md)
- [stated] Ben: analyze the whole rebuilt history every day at 08:00 BKK, only new days, send updated insights (hist-analysis.service, /opt/hist/analysis/coins.db).
- 2026-10-03 | cowork: RH base rate from chain (1 day, 2 Oct) | Buy 60 s after the first trade, top must be confirmed by the next trade: 1,657 traded coins, 2x+ 41.8%, 10x+ 1.63% (27), 50x+ 0 within ~1 day; median 46 min to the top | rh/2026-10-02.db | TESTING (1 day; 14d window not closed)
- 2026-10-03 | cowork: tape features on chain history (1 day) | 10x rate x5.2 with 50+ buyers in first 5 min (178 coins, 15 hits), x3.5 first sell after 12 s (421, 24), x3.1 sells <= 38.6% of 5 min trades (472, 24); first sell within 12 s x0.2 (1,134, 4); no sell at all in 5 min 0 of 102 | daily_analysis.py | TESTING (matches round 7 and score-7 on independent data; needs more days)
# ATH breakout buying on single coins (4 Oct 2026 00:00 BKK, cowork; doc claude/athbreak-study.md, scripts claude/athbreak-scripts.py.txt)
- [stated] Ben: basket rotation buys breakouts; asked if buying a coin's new-ATH breakout works too, how to tell a sticky big runner from a failed breakout (holders, X volume, sentiment 12h before / during / 12h after, new voices, 15h MA), fib extension targets, 50 ideas; his thesis: identify strengthening baskets first, rotate out of a euphoric basket, and breakouts in the same or a correlated basket right after its euphoria tend to fail.
- 2026-10-04 | cowork: plain ATH breakout on coins | If you buy the first hourly close above the old ATH (ATH 12h+ old, 20%+ dip, $500k+), then no edge: Pons fair list avg -18% (stop on a close 20% under old ATH), 2x/-50% -7%; 3 in 4 close 15%+ back under the old ATH within 48h; mix 36% fail / 25% 1.3-2x / 23% 2-4x / 14% 4x+; chart features cannot pick runners (GBM AUC 0.47-0.56 out of sample) | 824 events, 357 coins (250 Pons fair + 574 runtrack/kl/gtlane) | PROVEN (tokens are not baskets)
- 2026-10-04 | cowork: vertical breakouts | If the breakout comes after +100% in 24h, 50%+ above the 24h EMA, a +20% candle, a 10% wick or bearish RSI divergence, then worse (e.g. 24h +100%: -18% vs +3%), all 4 splits, CI below 0 | run_ideas.py | PROVEN (avoid)
- 2026-10-04 | cowork: retest entry | If you place a limit at old ATH +3% instead of buying the breakout close, then 97% fill within 72h (median 1h) and 2x/-50% +4.0% vs -0.6% (fair list -5.6% vs -9.4%) | ev5 | TESTING (better, still not positive on the fair list)
- 2026-10-04 | cowork: basket context (1-day lag) | Chain meme basket 7d +20%: 2x 53% vs 33%, fail 24% vs 38%, trade +7% vs -9% (CI +3..+33), but 129 of 148 events are 25-30 Aug; chain basket 1-10d after its euphoric top and 10%+ off: trade -16% vs -5%; RH Pure Memes post-top: 4x 7% vs 16%, fail 44% vs 33% | post_euph.py | TESTING (supports Ben's rotation thesis in direction; one episode)
- 2026-10-04 | cowork: X around breakouts (contract + ticker, 3+ likes, 268 events; plus bostudy pull 420) | Busy or accelerating chatter into the break = worse trade in both pulls (busy 12h: -18% vs -6% and -21% vs -12%; last 3h half of 10h: -22% vs -13%, CI below 0); caller/shill-led tone 40%+: -28% vs -6% (LLM read, one pull); big runners' chatter cools in the 12h before then climbs after, failed ones stay loud then flat (chart); buying 12h later on rising chatter: unstable (+7 vs -2, -13 vs -7) | xtest.py, tone2 reads | TESTING
- 2026-10-04 | cowork: fib extensions | Tops after a breakout do not cluster at fib levels of the pullback leg (26.5% vs 26.1% random); reach rates 1.272 71%, 1.618 54%, 2.0 42%, 2.618 32%, 4.236 21%; from 1.618, 60% reach 2.618 | fib test | KILLED (fibs as sell levels), PROVEN (descriptive)
- 2026-10-04 | cowork: killed for breakouts | base length, earlier ATH touches, higher lows, coin age, time of day, weekend, majors basket, Pons launch mania, silence before (mixed), new voices before, big accounts before, every exit variant (none turns the plain breakout positive on the fair list) | run_ideas.py, exits.py | KILLED
# X mentions vs moving averages + odds check (4 Oct 2026 00:30 BKK, cowork; doc claude/pullback-ath-study.md section 5)
- [stated] Ben: cut the fat (shut down what does not work), 5 iterations on the NEW-ATH odds, 10 iterations on X mention / new-account moving-average patterns at bottoms, ATH retests and tops; his idea: parabolic new voices far above their MA = everyone already bought = top.
- 2026-10-04 | cowork: odds check (5 passes) | Base held (gave back <= 75% of the run) is the real driver: 25 to 31% vs 12 to 15% new ATH in 14d, holds in discovery/confirm halves, early/late, 1h and 4h data, the live buyscore replay (99.8% same calls); COMEBACK adds nothing beyond it on the fair list (confirm half 26 vs 28%) | verify_odds.py | TESTING (best price lead)
- 2026-10-04 | cowork: crowd still active at the pullback | Inside base held, X 3+ like posts in the last 24h >= 50% of the ATH day: 40% [32,46] vs 23% [14,34] new ATH in 14d (both halves); confirms Bottom Lab's "activity matters, mood does not" on the fair list | 269 fair Pons coins, 18,790 comparable posts, xan.py / xan3.py | TESTING
- 2026-10-04 | cowork: NEW VOICES at an ATH retest | If price is back within 10% of the ATH after a 30%+ dip and 6+ accounts posted about the coin for the first time (3+ like posts, not seen in prior 7d) in 24h, the breakout (+10% over the old ATH in 3d) happens 49% vs 73% with 0 to 2 new voices; dose response 74/72/59/50/47%; holds with volume, momentum and coin-size controls (logit p < 0.001), first retest per coin, both halves; gone with X delayed 7 days (placebo) | 199 retests, 94 coins | TESTING (warning, live in buyscore v3.5)
- 2026-10-04 | cowork: NEW VOICES at a new top | 11+ new voices in 24h at a fresh ATH: only 63% [53,72] go 10% higher within 7d vs 82% (Ben's parabolic top idea confirmed at the extreme only; 3x MA version 71 vs 82); moving averages of raw mentions (3d, 7d, 14d) add little anywhere | 620 new ATHs, 119 coins | TESTING (TOP RISK warning live)
- 2026-10-04 | cowork: buyscore v3.5 md5 98da93e2348726ff91d1dbd0a2e0ba64 (droplet backup buyscore.py.bak-v34) prints NEW-ATH ODDS (base held + crowd active ~40%, quiet ~23%, base lost ~12% avoid, 80%+ ~6%) and the RETEST / TOP RISK warnings | 2 extra min_faves:3 searches per run | LIVE (paper)
# ATH breakouts v2: Solana last cycle, basket picks, exits (4 Oct 2026 01:30 BKK, cowork; doc claude/athbreak-study.md v2)
- [stated] Ben: use the basket new high as the BUY signal and pick WHICH coin; test on Solana last cycle with multiple baskets, balanced small/mid/large samples; wants the key insights repeated in every answer; prefers pre-breakout signals (waiting 3 to 6h can miss the move); wants a tested exit for breakouts that chop or fail (depth under ATH, drop from peak, hours, social).
- 2026-10-04 | cowork: coin pick inside a Solana basket BUY | Biggest (and oldest) coins beat the basket median 62 to 70% (both 2023-24 and 2025-26); the coin already at a new 30d high: no edge (48 to 54%); strongest 7d/30d: worse (40 to 46%); coin up 30%+ that day or 15% wick: beats only 30 to 40% | 66 basket BUYs, 6 Solana baskets, CMC point-in-time listings Nov 2022 to Oct 2026, picks_an.py | TESTING (agrees with live picks.py older-listing lean)
- 2026-10-04 | cowork: Solana ATH breakouts (496, 295 coins, small 113 / mid 247 / large 136, hourly) | basket 7d +10% or basket broke its 30d high in the last 3 days: +20% vs +7% (CI +2..+26, 4/4); no basket break in 30d: +3% vs +17%; 3rd+ breakout of a coin: 2x 10% vs 34%; daily RSI under 65: +4% vs +15%; large $100M+ fail less (8%) but 2x only 16% vs 33 to 40% | athh3.pkl, ideas50.py | TESTING
- 2026-10-04 | cowork: CORRECTION vertical breakouts | "Avoid vertical" holds only for small RH coins; on Solana CMC coins up 100%+ in 24h doubled 55% vs 26% (trade +19 vs +11, CI crosses 0) | athh_an.py | size-dependent
- 2026-10-04 | cowork: CORRECTION quiet X before the break | Solana: 0 real posts in the 12h before = worse (+3% vs +14%, CI below 0); contract posted in 20%+ of pre posts +42% vs +9% (CI +5..+61, 4/4); the RH "quiet then loud" runner signature is descriptive only | 496 X pulls, xan3.py | KILLED as a general filter (RH small coins only, TESTING)
- 2026-10-04 | cowork: EXIT after a breakout | Retest entry + sell if no close 30% above entry within 48h + trail 40% from the best close once +50%: Solana mean +23% (worst tenth -24%) vs hold 14d +27% (-60%); RH fair -11% vs -37%; best rank across both chains and all halves of 686 exit combos. Small coins add a stop on a close 20% under the old ATH (worst tenth -37% vs -81%); on Solana mid/large that stop lowers the mean (+16% vs +23%) for no tail gain | exits.py, exit_grid.pkl | TESTING (best candidate)
- 2026-10-04 | cowork: chop and drawdown odds | 48h after the break, chop within 10% of entry: 2x later 11% (Sol) / 26% (RH); up 30%+: 73% / 82%; after +30%, a 40% drop from the best close leaves new-high odds 25% (Sol) / 43% (RH) | exits state tables | PROVEN (descriptive)
- 2026-10-04 | cowork: combined rule (basket strong + 1st/2nd breakout + retest + 48h/+30% time stop + 40% trail) | Solana n=165 +34% (median +2%, worst tenth -24%, win 54%) vs rest +16%; 2023-24 +41 vs +22, 2025-26 +14 vs +11; RH runtrack +20 vs +6; RH Pons fair -4 vs -14 | combined test | TESTING (paper rule)
# RH chain history, honest-entry check (4 Oct 2026 00:50 BKK, cowork; evtest.py)
- 2026-10-04 | cowork: CORRECTION "50+ buyers in first 5 min" | It was scored with a 60 s buy but is only known at 5 min. Bought honestly at 5 min: 10x 1.6% (307 coins) vs 6.7% with the leak; ~2.7x the 5-min base (0.7%) but no better than buying everything at 60 s. Daily report now scores every 5-min feature from a 5-min entry | 3 days 30 Sep to 2 Oct, 4,583 coins | KILLED as a big lift (leak)
- 2026-10-04 | cowork: first sell after 12 s (honest, known at 12 s) | Buy at 60 s: 10x 4.8% (979 coins) vs 1.76% all | evtest.py | TESTING (holds on chain history)
- 2026-10-04 | cowork: break-even math | All out at 10x, stops fill median -47 to -65%: if every non-10x trade loses at the stop you need ~6% 10x hits to break even (5% loses, 8% ~+15 to +20%/trade). BUT on chain history ~85% of trades hit neither 10x nor the stop (coin goes quiet, last price median ~1.8x entry), so the real P&L depends on what $100 to $500 can actually be sold for in a thin pool. Paper EV of blind buys at 60 s shows +112%, which is not believable: sell impact, launch fees and fill at 60 s are not modelled yet | evtest.py | OPEN (needs sell-impact model + forward paper)
# Coin picks v2 + cross-check of the athbreak v2 session (4 Oct 2026 00:50 BKK, cowork; raw http://159.223.88.168/hx-c9ois/edge/coin-picks-v2-2026-10-04/)
- [stated] Ben: bet on strength, not laggards (the SOL / HYPE / AVAX idea); asked to double-check the athbreak v2 session.
- 2026-10-04 | cowork: picks v2 (strong but not stretched) | Inside a basket BUY pick the 3 coins closest to their 30d high (weekly points) with the smallest 90d gain: halves 2019-22 +1422/+366%, 2023-26 +297/+157% vs random 3 +691/+113%, +277/+17%; laggard v1 slightly better on numbers (+1686/+136, +612/+214) but replaced on Ben's call; pure 7d/30d strength 2025-26 half -51/-45% | halves2.py, halves3.py | LIVE (picks.py v2)
- 2026-10-04 | cowork: CROSS-CHECK of "biggest + oldest beat the basket 62 to 70%" | Daily CMC basket BUYs, pool of 50: biggest + oldest beat the basket 35% (Solana) / 37% (all, 2023-26) / 48% (2019-22) vs random 37 / 37 / 44%; biggest alone 31 / 37 / 43%; oldest alone 41 / 41 / 46% | xcheck.py | DISAGREES on "biggest" (oldest small plus agrees; hottest 7d/30d, 30%+ day and already-broke-out worse or no edge: AGREES)
- 2026-10-04 | cowork: CROSS-CHECK breakout facts on daily Solana CMC coins (5,378 breakouts over a 30d high, 2023-26) | 1st/2nd new all-time high doubled within 14d 11% vs 3rd+ 3% (agrees); large caps halve least 4% (agrees) and double least 4%; small vs mid order differs (6% vs 9% doubled); 2 day time stop at +30% + 40% trail once +50%: worst tenth -53% to -33% but 2023-24 average +6.5% to +1.0% (partly agrees: cuts losses, costs upside) | xcheck2.py | daily data, rates much lower than the hourly ATH study (different universe)
- [stated] 4 Oct 2026, Ben: asked why I switched to his strength pick when my laggard pick scored better, why this was not found earlier, and to test the pick rules on as many not-yet-analyzed baskets and coins in this cycle as possible; asked if the real-money (stablecoin inflow) study belongs on the dashboard.
- 2026-10-04 | cowork: coin picks OUT OF SAMPLE | 11 never-used baskets this cycle (Solana gaming, BNB four.meme, other-chain memes, TON games, Solana Binance Alpha + 6 overlapping), 85 basket BUYs (32 on fully new coins): no pick rule beat 3 random coins of the pool (laggard v1, strength v2, biggest, oldest, biggest+oldest, near 30d high); strongest 7d = lottery (beats basket 36%, drops 54 to 62%) | pick/extra_eval.py, edge/coin-picks-oos-2026-10-04 | KILLED as an edge (the edge is the basket timing; split evenly); dashboard overlay now says so
- 2026-10-04 | cowork: Solana stablecoin inflow flag in the rule | 7d avg supply +6/+8/+12% in 14/30d: blocking meme buys +2161 to +2169% vs +2256% live; 10% trail on Solana memes while flagged +2289 to +2300%, 2025-26 unchanged | pick/stables_test.py | KILLED for the dashboard (too small)
# Big-cap holds and the power law (4 Oct 2026 01:20 BKK, cowork; bigcap.py, bigcap2.py)
- [stated] Ben: if we can't find a real edge the runner work is wasted time; would rather hold to $1M / $5M / $10M market cap, even if rare, if a system can pick them with good accuracy; doubts that is realistic.
- 2026-10-04 | cowork: POWER LAW on the Pons fair list | Of 9,986 Pons graduations: ATH $1M+ 5.1% (1 in 19), $5M+ 0.97% (1 in 102), $10M+ 0.39% (1 in 256). Odds of reaching cap M once a coin reached cap m are almost exactly m/M at every step ($250k to $1M 24.4% vs 25%, to $5M 4.7% vs 5%; $1M to $10M 7.7% vs 10%). So holding to a target pays exactly the odds: EV ~0 before costs, and a filter's whole edge = how far it beats m/M | univ_pons.json.gz | PROVEN (descriptive)
- 2026-10-04 | cowork: buy after a coin crosses $250k, hold to 4x / 20x | Bought at the next hourly open (median entry $173k): reached 4x 22% / 15% (early / late half) vs 25% fair, 20x 4.6% / 1.6% vs 5%: lift 0.33 to 0.93, worse in September; crossing-hour volume $500k+, close near the high, launch-to-graduation speed, quote asset, hour: none beat the fair line in both halves | bigcap2.py | KILLED (simple filters do not pick big runners)
- 2026-10-04 | cowork: why paper trades are ~0 | rhfast runs since 3 Oct 13:25 BKK; rule-1 (curve 99%+ within 75 s and few holders per $10k) fired once in 12h (swordcat 0x475d568003e27f8dd746573f0f7c173f1efc76bf, -85%); shadow scores 7/8/9 only rate rule-1 coins, so they are 0 too; 4,734 launches evaluated | fast.db | FACT
# Pullback to fast ATH, UNBIASED Pons rebuild (4 Oct 2026 ~01:30 BKK, Bottom Lab session; doc claude/pullback-fast-ath-unbiased.md)
- [stated] Ben: re-test on a wide set of Pons launches, no filter, dead coins included, ~1,000 coins, 50 new ideas; fewer alerts with higher accuracy preferred; social analysis skipped this round (add later). This is "the pullback to fast all-time high unbiased study".
- 2026-10-04 | cowork: universe | Every Pons launch 20 Jul to 22 Sep from chain logs (~150k), swap-sampled every 10 min: 732 ever reached $50k, 213 $100k, 128 $250k; 4h bars from every swap (container /home/claude/bl/pons) | PROVEN (data, independent of the 9,986-graduation fair list)
- 2026-10-04 | cowork: FAST FLUSH re-test | New ATH within 7 days 16.2% vs 14.3% all dips ($100k set, same-depth lift -0.4); July blind 9.4%; the runtrack 31% was survivor bias | ideas2.py, oos.py | KILLED
- 2026-10-04 | cowork: fast-ATH drivers that held (all 4 halves) | fast drop <=2d +11 pts, clean ATH candle (small wick) +10 vs wick-spike top -9, 1+ prior comeback +9, market cold +7 vs hot -7; slow 5+ day drop 0/23; old 14d+ -11 | t50.py | TESTING
- 2026-10-04 | cowork: BASE HELD replicates | Gave back <= 75% of the run: new ATH 7d 24% vs 11% (Aug-Sep), July blind 29% vs 7% | same data | TESTING (best lead, agrees with fair-list session)
- 2026-10-04 | cowork: CLEAN DOUBLE TOP (R2) | ATH candle not a wick spike + price already came back to 90%+ of ATH once + not far below 7d average: 35.6% new ATH in 7d (73 dips, 34 coins) vs 14%; July 2/7; with base held 35% (43 dips) | combos.py (364 combos mined, prereg.json before July check) | TESTING (small n; "Asia hours" combos looked best but scored 0% in July: KILLED)
# Coin picks, far-and-wide check (4 Oct 2026 02:30 BKK, cowork athbreak session; doc claude/athbreak-study.md v3)
- [stated] Ben: double-check the coin-pick disagreement far and wide; the key question is how to pick future outperformers when the basket breaks out.
- 2026-10-04 | cowork: WIDE pick test | 14 point-in-time baskets on ETH, BNB, Solana, Base (memes, DeFi, AI, gaming) + L1 alts, 8,908 CMC coins, Oct 2019 to Oct 2026, 210 basket BUYs (30d high + BTC > 100d), vanished coins -50%: pick-3 averages +11 to +21% at 30d, equal-weight all members +15% with the smallest worst tenth (-25% vs -28 to -32%); biggest = steadier (beats basket 42% vs 36% random, median +4.8% vs +2.5%) but below random in the 2020-21 bull held to the sell; LIVE v2 (near 30d high + small 90d gain) -3 to -5 pts vs random; strongest 7d = lottery (beats basket 36%) | bk2/wide.py, wide_an.py | KILLED as an edge for every single pick rule (agrees with the OOS session); CORRECTS my v2 "buy the biggest" (held on Solana 2023-26 only, CI above 0 there)
- 2026-10-04 | cowork: up 30%+ on the basket BUY day | Held to the basket sell: beats the basket 19% vs 38%, median -22% vs -3%, worse in all 4 eras (2020-21, 2022, 2023-24, 2025-26) | wide.pkl | PROVEN (skip these coins)
- 2026-10-04 | cowork: exits vs hold in bull years | Both sessions: the 48h/+30% time stop + 40% trail cuts the worst tenth (-60% to -24% hourly Solana; -53% to -33% daily) but lowers the average in 2023-24 (+43% to +29 to +31%; +6.5% to +1.0%) and helps in 2025-26 chop | exits.py, xcheck2.py | AGREED (risk tool, not a return booster)
# Coin picks v3 + X chatter (4 Oct 2026 01:40 BKK, cowork; doc claude/coin-picks-study.md; raw http://159.223.88.168/hx-c9ois/edge/coin-picks-v3-2026-10-04/)
- [stated] Ben: agreed with the wide pick test; wants the coins in the "Coins to Buy" pop-up followed by key insight reminders (basket first, spread out leaning bigger, skip +30% day or long wick, don't chase the 7d leader, exit limits losses).
- 2026-10-04 | cowork: X chatter as a coin pick | 832 picks in basket BUYs, Solana and Base 2023-26: quiet coins (0 to 2 posts in 7d) beat the basket 49% vs 36% alive (halves 46/29, 51/40); fifths U shape (quietest 45, loudest 42, middle 36, 4th 27); happy middle worst 35 vs 39; surging 3x+ worst in 2025-26 (29%, median -10%) but mixed 2023-24; likes nothing | pick/xanal.py | TESTING (not shipped; reminder only)
- 2026-10-04 | cowork: picks v3 live | 5 biggest liquid coins of the category (vol $100k+, listed 8+ days), minus +30% day coins and the 7d leader, equal amounts; replaces v2 (v2 was -3 to -5 pts vs random) | picks.py md5 0d4eee58f708c44b451e49cfc17213b6 | LIVE
# ATH breakout deep test, 6 size bands (4 Oct 2026 ~01:40 BKK, cowork; doc claude/athbreak-deep-study.md)
- [stated] Ben: doubted buying 12h after a breakout (likely already ran); wants equal micro/small/medium/above medium/bigger/biggest samples, 50 ideas, a call at the first-hour retest on which breakouts run big; his idea: coin beating its basket / first in its narrative.
- 2026-10-04 | cowork: 12h criticism | Runners (2x in 7d) were a median +6% above the first-hour retest at T+12h, 5% of the move done (18% already +50%); entry at T+12h -0.7% vs -1.9% at the retest; never buy at T+12h if already +20% (-4.8%) | 1,928 breakouts, 678 coins, ab_12h.py | PROVEN (descriptive)
- 2026-10-04 | cowork: beats its basket / first in narrative | Top third 3d strength vs chain basket: trade -7% vs +4% (touches 2x more, crashes more); first in narrative -6.4% vs -5.3% | ab_anal.py | KILLED
- 2026-10-04 | cowork: HOT breakout at the first-hour retest | 2+ of: big 24h run, far ahead of basket, high launch multiple, no 12h base, young coin. Late half out of sample: drops 50%+ in 7d 75% (HOT) vs 48% (MID) vs 9% (CALM); trade -13.6 / -5.4 / +4.6%; 2x 27 / 32 / 15%; reversed halves hold (HOT -17%). Fair Pons list: CALM small coins still -18% (n=40) | ab_score5.py | PROVEN (avoid HOT); CALM = low crash, rarely runs
- 2026-10-04 | cowork: X at the breakout (1,164 events, all 6 bands) | Pre 24h, first 2h, new voices, big accounts, quiet, loud vs own baseline, acceleration: no money edge at the retest (all trade CIs cross 0); chatter rising after = hindsight, as a T+12h entry -5.8% vs 0.0% | ab_xtab.py | KILLED for breakout timing
- 2026-10-04 | cowork: basket strong at the break | +5% vs -7% overall but a time effect: inside each time half CI crosses 0 | ab_anal.py | KILLED as a gate (regime proxy only)
- 2026-10-04 | cowork: bowatch v1 live | /opt/bowatch/bowatch.py, cron :07 hourly, scans ~1,080 runtrack coins, publishes fw/breakouts.json with HOT/MID/CALM + odds; first run caught a kline vs snaps mcap seam that made 6 of 7 fake breakouts (fixed) | md5 aa71abf47107cc485d29a50a1bae77e9 | LIVE (paper)
# Second leg base, honest census (4 Oct 2026 02:00 BKK, shortlist pipeline session)
- [stated] Ben: test the base window honestly on many more coins, this cycle only, RH and Solana, meme vs project, with hourly X crowd via the shared spam filter; AI (Artificial Inu) as the model case.
- 2026-10-04 | cowork: DATA TRAP fake caps | pump.fun ath_market_cap and GeckoTerminal candles show $20M to $500M "ATHs" on dead Solana coins with under $1M total volume (fake liquidity spikes; 161 of 649 sampled "$20M" coins). Rule: trust an hour only if 24h volume >= 5% of its mcap, and an ATH only if volume traded up to it >= 20% of it. After that, 1.4% of pump.fun graduates reach a real $1M | random 10,560 of 116,828 graduations 15 Jul to 3 Oct (MigrateV2 txs), sol_pf.json, gt_sol.json | PROVEN
- 2026-10-04 | cowork: base alone vs power law | Pons fair list (all 9,986 graduations): 915 base events (first $1M, -50%, 12h no new low; re-arm after a 30% deeper low) on 332 coins: reached $5M 12.9% vs 15.8% expected from m/M; new ATH 13%; -50% hit before 2x 59%. A bare base is BELOW the power-law line | events_pons.json | PROVEN
- 2026-10-04 | cowork: AI case | Artificial Inu fired 4 bases after its first leg ($17.3M top): $9.6M, $4.9M, $3.7M (each hit -50% first), then $1.9M on 2 Aug which ran 207x; X real 3+ like posts stayed 5 to 20 a day through the -90% base (peak days ~30) | ai_hour.json, ai_x.json | n=1
# Picks v4 + quiet check (4 Oct 2026 02:10 BKK, cowork; raw http://159.223.88.168/hx-c9ois/edge/rotation-ideas-2026-10-04/)
- 2026-10-04 | cowork: CORRECTS picks v3 "lean biggest is fine" | 5 biggest eligible vs all eligible equal: 2019-22 +33.9 vs +38.2% (100 buys), 2023-26 +7.7 vs +10.6% (80); paired -4.3 [-9.8,+1.2] and -2.9 [-9.6,+3.7] pts; 10 spread big to small +39.2 / +9.3% | pick/v3wide.py | LIVE (picks v4: 10 coins spread)
- 2026-10-04 | cowork: quiet on X re-check | Quiet coins beat the basket more often inside every size third, but average return mixed (small third quiet +8.5 vs alive +21.2%); quiet tilt vs random 5: 2023-24 +26.2 vs +17.0, 2025-26 +4.6 vs +12.4 | pick/quiet2.py | KILLED as a pick rule (loses less often, not more money)
- 2026-10-04 | cowork: rotations per cycle (rule v4.7 on the bench) | Bull 2 (Jan 2023 to BTC top 6 Oct 2025, 14 baskets): 5.5 BUYs per basket (median), 95 total, ~3 a month, win 52%, 15 trades of +50% (~1 per basket); Bull 1 (Mar 2020 to Nov 2021, 15 baskets): 7 per basket, 110 total, ~5.5 a month, win 73%, 39 trades of +50% (~3 per basket); bear 2022: 1 per basket, win 13%; since the Oct 2025 top: 2 per basket, win 60% | it100/rotcount.py | PROVEN (descriptive)
# Paper vs real (4 Oct 2026 02:15 BKK, cowork; sim.py, realwal.py)
- [stated] Ben: daily report must show if we had traded yesterday / this week, money made or lost per approach, and whether we caught the runners; wants much better picking (cabal and organic runner patterns).
- 2026-10-04 | cowork: REAL early buyers | 2,842 real wallets that first bought 45-90 s in (600 coins, 2 Oct): median 0.24 ETH back per ETH, 28% in profit; per coin real ~1.00x vs our paper rule 1.55x on the same coins. Over 3 days, "no sell in 12 s" paper +$199k on $200 bets but real -$5k: paper NOT trusted; hookless fee-0.01% pools hold 51 of 62 10x coins but real ~0% | sim.db | OPEN (only a small live test settles it)
- 2026-10-04 | cowork: X crowd at the base vs power law | Pons census, 315 bases (210 coins, base $250k+), real 3+ like posts 24h before base vs peak day: crowd held reached $5M 26% vs 31% expected (lift 0.8); crowd left 8% vs 20% (0.4, AVOID); best gate (held + 5+ people + base $500k+) 33% vs 39% (0.9). No base feature beats m/M in both halves; negatives: crowd left 0.4, volume dried <10% of peak day 0.31, peak under $2M 0.47, depth 80%+ 0.64. Small RH sample "40% reach $5M" was cap confounding | x_pons.json | PROVEN (no buy edge at the base; avoid filters only). Solana random sample agrees: 3.2% vs 3.1%. Live: /opt/legwatch short list + AVOID/WATCH crowd check, alerts off ([[shortlist-pipeline]])
- 2026-10-05 | cowork HOOKR claim check | RH infra tokens price on ANNOUNCEMENT cadence, not usage: HOOKR went $7.9M cap (15 Sep) to $28M high (3 Oct) while DefiLlama revenue fell from ~$600/day to ~$140/day, launches went 115 to 116 in a month, and every burn in 20 days was team inventory (deployer 865,791 + wallet 0xb266ef3a1a76cfc245d308e30e99aa27d1be48b5 516,247), zero market buyback. Trade the partner-post wave, exit when posts slow; never value on fees | receipt: Llama hookers revenue series, getLogs Transfer to dead 20d, hookr.fun/api/protocol-stats | TESTING
- 2026-10-05 | cowork HOOKR | Hook pool share is won by being a LAUNCHPAD's engine, not by selling hooks: Doppler hook 0x4e3468951d49f2eea976ed0d6e75ffcb44a9a544 = ~30k of 42k hooked pools in 14d because long.xyz LongLaunchFactory 0x1eef016f22a943abc7dd11422edee9d235942104 makes ~65% of them (Doppler free SDK, takes 5% of trade fees). Hookr "95% outside hookr.fun" TRUE (4.4% lifetime, 0.25% last 30d via Hookr routers), but 87% of Hookr ETH volume was one churned pool (GOOSE). Method: Swap logs topic1=poolIds, count topic2 sender | TESTING