--- name: trading-edge-hub description: THE shared trading brain for every chat and Cowork session: live risk on/off and rotation dashboard, index of every edge file, and the insight ledger. Read before any buy, sell or chart call. sources: [chat] aliases: [edge hub, trading insights, insight ledger, shared trading brain, risk on off, all coworks] --- # Why this file exists - [stated] 1 Oct 2026: Ben runs many Cowork sessions in parallel, each testing a different approach; across them he has analysed ~1000 coins across chains. He wants every trading insight from every session, old and new, available to every other session, so they build on each other and can judge any new chart. - [stated] He wants every call to use the risk on/off and rotation dashboard (rotation between chain baskets, weekly and monthly closes, risk on/off, topping and bottoming warnings for BTC, majors and indices) together with the 1000-coin insights. # Rules for every session (read, use, write back) - [stated] READ: before any buy, sell, hold or chart verdict, read this file, then pull the live dashboard JSON (below). The verdict must say which ledger rules fired. - [stated] WRITE: any new trading insight goes into the ledger below in the same turn, as one line: date, session/approach name, the rule (if X then Y), the receipt (sample size, hit rate, period), status TESTING / PROVEN / KILLED. Long detail goes in its own /areas/ file, linked here. - [stated] CONFLICT: a new result never silently overwrites an old rule. Add the new line, mark the loser KILLED with the receipt that killed it. # Live data any session can fetch (no login) - Risk on/off + rotation dashboard: http://159.223.88.168/rotation.html - Same data as JSON (word = basket price word from chains.json v3; btc_r7, sol_btc_r30, pad_r14, share_7d, stables_r30): http://159.223.88.168/rotation.json - Handovers: http://159.223.88.168/hx-c9ois/rotation/handover.html and http://159.223.88.168/hx-c9ois/rotation/dashboard-handover.html - Paper-trade logs: http://159.223.88.168/hx-c9ois/rh_paper.csv.gz , http://159.223.88.168/hx-c9ois/sol_paper.csv.gz - Fetch with a browser User-Agent. # Where each edge already lives (read the file before re-deriving) - Rotation and regime: [[rotation-watch]] (BTC 7d > +10% lifts launchpad volume, not meme prices; ledger 2 Oct). [[regime-tracker]] and [[rhx-index]] are RETIRED, history only. - RS: [[strong-coins]] (dip RS KILLED) - Second legs and bases: [[runner-lab]], [[pullback-comeback]], [[move-durations]]. - Flow and new launches: [[flow-model]], [[trench-model]] (parts 2 to 6), [[techscan]], [[casig-gmgn-overlay]]. - Calibration of past calls: [[base-rates]], [[screen-log]] (parts 2 to 9), [[screen-log-outcomes]] (parts 2 to 5), [[model-blind-tests]]. - Trader frameworks: [[okay-lets-ride]], [[uberm-framework]], [[unipcs-worldview]], ansem-worldview skill. - Sector maps: [[long-lane-map]], [[rh-launchpad-sector]], [[rh-value-accrual]], [[rh-holder-reward-coins]], [[stock-paired-venues]]. - Method and DD: [[token-screens]], [[crypto-trading]]. # Insight ledger (one line each, newest at bottom) - 2026-10-01 | chat: personal trading pattern review | Revenge trading: if the book has lost money every month this period, then stop opening new trades for 30 days; more trades under a losing method means more losses | receipt: Ben's own book, losing every month through Sep 2026 while still adding money; n = 1 trader, no hit rate | TESTING - 2026-10-01 | chat: exit discipline | If a buy has no written sell ladder and "I was wrong" price before entry, then no buy; Ben holds with no stops or exit levels and stops looking at red charts | receipt: Ben's own drawdown from his cycle peak this cycle, by his account; n = 1, no hit rate | TESTING - 2026-10-01 | chat: panic filter | If Ben has just checked his balance and feels panic, then no trade for at least 1 hour | receipt: none yet, rule set 1 Oct 2026 | TESTING - 2026-10-01 | chat: sleep filter | If Ben has had less than two full nights of sleep, then no new trades; big ideas from 2 to 4 hour nights are not trusted | receipt: none yet, rule set 1 Oct 2026 after two short nights | TESTING - 2026-10-01 | chat: edge fit | If the play is a trench flip on fomo, then remember that most fomo traders lose money and the profitable ones often have followers whose buys pump price right after theirs (a buffer Ben does not have); Ben's best past results came from large holds through bull runs (2024/25 cycle) | receipt: Ben's own observation of fomo traders plus his own cycle history; no measured sample | TESTING - 2026-10-01 | chat: basket divergence | If the Robinhood, Solana and Base baskets disagree (one topping while another rises), then judge each basket separately rather than calling the whole market; Ben is starting to track this on a dashboard ([[rotation-watch]]) | receipt: Ben's own observation, no measured sample | TESTING - 2026-10-01 | chat: early bottom bias | If a bear market setup looks 50/50 near a possible bottom, then assume Ben's read leans too bullish and wait for confirmation before buying; he tends to start buying too early and then sits through hard chop | receipt: Ben's own account of the last bear market turns; n = 1 trader, no hit rate | TESTING - 10-03 cowork: sell breakouts day 8-13; X loud+angry=still falling, quiet=floor; big coins roll first=top [[macro-direction]] TESTING - 2026-10-01 | cowork: strong coins vs BTC ([[strong-coins]]) | If a coin beats BTC in BTC up, down AND flat 4h windows (trimmed means, wins at least half the windows in each, last 7 days, 3 day half life recency weight), then rank it strongest; full pass on 1 Oct: HARMONIC 0xdEe52F2ab639b6942B0d0F0565400b93b7a0fbe5 (RH); 3 of 4: PUMP pumpCmXqMfrsAkQ5r49WcJnRayYRqmXz6ae8H7H9Dfn, SANTACOIN 0x38bb87cf3e7344a82be7cd671c191f015b97eee6, 0xZAPS 0xDd90bFa4adC7F4401E611AbaC692D939F9F4CB07, NOTE 0xc4f730335Fb9e439ca5552f7b52B8E638c4245B0, DATFRONG 0x352a746f58fd970358593b8f60d1283a46f90528, OPENAI PreweJYECqtQwBtpxHL171nL2K6umo692gTm7Q3rpgF, RVH 0x96765066f6a040a21EB027167D2315B707c82633 | receipt: 214 RH+Solana coins scored, 24 Sep 13:00 to 1 Oct 13:00 Bangkok, BTC had 12 up, 12 down, 18 flat 4h windows; 1 full pass, 8 pass 3 of 4; no forward outcome yet | TESTING - 2026-10-01 | cowork: strong coins, luck test | If you rank RH/Solana memecoins by how they behave in BTC up/down/flat 4h windows, then read the result as "beats BTC whatever BTC does" (own trend), not as a BTC linked dynamic; shuffling BTC windows at random gave the same pass counts | receipt: 8 random shuffles over the same 214 coins and week: 0 to 3 full passes (real 1), 2 to 9 passing 3 of 4 (real 8) | TESTING - 2026-10-01 | cowork: strong coins, spike trap | If a coin ranks top on raw average excess return vs BTC, then check trimmed means, win rate per window and distance from 7d high before calling it strong; one spike bar can carry the whole score | receipt: CINEMA 0x9936403A9cc72CF306063E3496EA387428A3aeCd ranked #1 on raw means (+175% vs BTC in 7d) while 43% under its 7d high and down 21% in the last 6h on 1 Oct | TESTING - 2026-10-01 | cowork: strong coins, universe trap | If a top volume pool list is used as a coin universe, then drop tokenized stocks (GOOGL, MU, NVDA, SPCXx), wrapped coins (wNEAR) and stables first; their low volatility makes them look "strong" (win rate 0.52 to 0.64 vs BTC) | receipt: 5 such tickers landed in the raw top 30 of 233 on 1 Oct | PROVEN - 2026-10-01 | cowork: coin list decay | If a coin from the old memory analysis is reused, then expect most to be gone; of 473 RH+Solana tokens in memory, 332 still had any DEX pair and 218 were alive (liquidity at least $10k, 24h volume at least $2k) | receipt: DexScreener sweep 1 Oct 2026, 46% of the old list dead or near dead | PROVEN - 2026-10-01 | cowork: data sources for price history (container) | If hourly history is needed from the cloud container, then use Kraken OHLC for BTC (720 hourly bars, works) and GeckoTerminal pool ohlcv for RH and Solana (works for Uniswap v4 pool ids, about 30 calls per minute, 304 pools took about 35 min); Binance is geo blocked, CoinGecko market_chart throttles; DexScreener tokens/v1 must match EVM addresses lowercased (194 found case sensitive vs 332 lowercased) | receipt: live calls 1 Oct 2026; GeckoTerminal 24h change matched DexScreener within 1 to 4 points on 10 of 10 top coins | PROVEN - 2026-09-30 | cowork: chain baskets rule chain-1 (all baskets, bottomed + 30 day high break + BTC risk on, hold 14 days) | If any chain basket breaks out after a bottom, then buy and hold 14 days | receipt: 9 point in time CoinMarketCap baskets (Solana, Base, TON), Jan 2024 to Sep 2026, one trade at a time: 48 trades, +1.1% avg, $1.14 won per $1 lost | KILLED - 2026-09-30 | cowork: equal weight chain baskets bleed | If you hold an equal weight basket of chain coins (memes, AI, DeFi), then expect it to lose; dead coins included they fell median -7.3% a month | receipt: 9 baskets, ~6,700 basket days, Jan 2024 to Sep 2026 | PROVEN - 2026-09-30 | cowork: chain-2 rule, Solana + Base memes only | If a meme basket (Pump.fun, Bonk.fun, Solana older memes, Base memes) made a 90 day low in the last 60 days, is back above a rising 20 day trend and closes above its 10 day high while BTC is >1% above its 100 day average, then buy; sell on a close below the 20 day trend, -12% from entry, basket running hot (z>=2), or day 14 | receipt: 34 trades, +7.1% avg, +7.4% median, 59% wins, $2.55 per $1 lost after 1% costs, Jan 2024 to Sep 2026; first half $3.80, second half $2.04; all 27 parameter combos profitable | TESTING (live scorecard from 29 Sep 2026 close) - 2026-10-01 | cowork: chain-2 out of sample | If the same chain-2 rule is run on meme baskets of other chains (Ethereum, BNB, Tron, TON, Avalanche, Sui, Robinhood) or before 2024, then it loses: the Solana/Base result may be luck | receipt: 58 out of sample trades Sep 2022 to Sep 2026, -2.0% avg, 36% wins, $0.75 per $1 lost; all 91 trades +1.6% avg, per week about break even (46 weeks, -0.2%) | KILLED (as a general rule) - 2026-09-30 | cowork: sell rule matters most | If a basket breakout trade has no sell rule (just hold 14 days), then most of the edge disappears | receipt: same 4 meme baskets, hold 14 days: 29 trades +2.4% avg $1.30 vs +7.1% $2.55 with the sell rule | TESTING - 2026-09-30 | cowork: entry speed | If you buy a basket breakout one daily close late, then the average roughly halves | receipt: chain-2, next close entry: 33 trades +3.3% avg, $1.69 per $1 lost (vs +7.1%) | PROVEN (in sample) - 2026-09-30 | cowork: BTC risk gate for meme buys | If BTC is below its 100 day average (risk off), then skip meme basket breakouts | receipt: chain-2 signals with risk off: 23 trades, -0.7% avg, 26% wins, $0.87 per $1 lost | TESTING - 2026-09-30 | cowork: non meme baskets | If the basket is DeFi, AI, apps or TON coins, then a bottom + breakout signal has no edge | receipt: same rule on 5 non meme baskets: 44 trades, +0.1% avg, 39% wins, $1.03 per $1 lost; also failed in both halves (walk forward) | KILLED - 2026-09-30 | cowork: earlier entry | If you buy when a basket is merely turning up (above a rising trend, no breakout yet), then you lose | receipt: 120 first day signals, -3.3% avg over 14 days, 33% wins (2024 to 2026) | KILLED - 2026-09-30 | cowork: basket filters tried | If you add a stale price filter, a $100k volume floor, buy only the 3 strongest coins, require rising volume, or require SOL/ETH/TON above their 50 day average, then results do not improve | receipt: stale filter $2.64 vs $2.55 (marginal); $100k floor 36 trades +3.8% $1.67; top 3 coins median fell to ~0%; rising volume -1.5% vs +9.1%; chain gate passed 72 of 80 anyway | KILLED - 2026-09-30 | cowork: basket momentum rotation | If you hold the basket with the best 14 day return each week, then you still lose with the market | receipt: weekly rotation over 8 baskets Dec 2024 to Sep 2026: x0.20 vs x0.11 equal weight (both losing) | KILLED - 2026-09-30 | cowork: survivor bias in alt breadth | If alt breadth is measured on today's surviving coins, then it reads too bullish; measure on the top 100 coins as they were at the start (dead ones count) | receipt: 29 Sep 2026: fair 90 day 51% vs survivor 59%, fair 30 day 75% vs 94%; no fair washout (<=5%) since Aug 2025 | PROVEN - 2026-09-30 | cowork: roster view trap | If a young DEX basket's history is rebuilt from today's top coins, then its past looks far better than it was | receipt: Bankr basket +620% for Sep 2026 and +326% over 30 days on roster view; Arc top coins +503% | PROVEN - 2026-10-01 | cowork: weekly close labels | If a major's weekly candle closes as engulfing, rejection, new high etc, then do not read it as a forecast for next week | receipt: 10 majors on OKX, all history, 3,187 weekly closes: next week up 45 to 57% for every label vs 50% base | KILLED (as a predictor) - 2026-10-01 | cowork: monthly close labels | If a major closes the month at a new 12 month closing high, then next month leans up; after a new low close or bearish engulfing it leans down | receipt: 10 majors, 705 monthly closes: new high close 130 cases, next month up 58%, median +4.6%; new low close 90 cases, up 42%; bearish engulfing 39 cases, up 38%; base 52% | TESTING - 2026-10-01 | cowork: Sep 2026 monthly close read | If you need the market state after Sep 2026, then: all 10 majors closed bullish (HYPE new high close, LINK +26.9%, ADA +24.6%), majors basket new high close +12.8%, 12 of 15 chain baskets bullish, Base memes new low close -33.1% | receipt: closes.json 1 Oct 2026 15:47 Bangkok; score next month against the monthly label rule above | TESTING - 2026-09-30 | cowork: data source timing | If a basket uses CoinMarketCap historical daily listings, then the file for day D (00:00 UTC prices) can arrive up to a day late; use live listing prices at ~00:35 UTC as a provisional close | receipt: 30 Sep and 1 Oct 2026 snapshots still missing 13+ hours after 00:00 UTC | PROVEN - 2026-09-30 | cowork: chains coverage | If you want a Kraken chain basket, then it is Ink (Arc is Circle's chain, mainnet 16 Sep 2026); Ink has only a thin meme market (pools $20k to $135k: OTO, ANITA, HKT, BEAST, OCTO, KRAKMASK, PURPLE, BERT, INKCAT, ICAT) plus tokenized stocks; Arc has a live launchpad scene since 24 Sep 2026 (Argus, Faze, Long, Foci, Peach) | receipt: GeckoTerminal pool scans 30 Sep 2026 | PROVEN - 2026-10-01 | cowork: chain baskets raw data | Public: http://159.223.88.168/chains.json (stats, live ledger, baskets, contracts), http://159.223.88.168/closes.json (closes), http://159.223.88.168/hx-c9ois/rotation/dashboard-handover.html (all trades, tables, full code). Droplet: /opt/rotation/chains/state.json (full basket lines, live trade ledger), /opt/rotation/chains/top/ (daily top 1,000). Raw bundle, public: http://159.223.88.168/hx-c9ois/edge/chain-baskets-2026-10-01/ (all.tgz, bt_wide.json = all 91 trades, evidence2/3, closes_hist, fairclock, rotdb_daily, trench_test.py, scripts) - 2026-10-01 | cowork: strong coins raw data | ranking.csv (all 214 scored coins with every metric), coins.csv (597 tokens extracted from memory with full addresses), universe.json (304 pools scored), ohlcv.tgz (hourly candles), btck.json (BTC hourly), scripts step1.py, step2.py, step2b.py, step3.py, all in /mnt/user-data/outputs/strong-coins-2026-10-01/ of Cowork session 01PuyEF2n5UMXZWNTsj7DCTM (lives with that session, not a public URL; scratch copy in that container is temporary) - [stated] 2026-10-01 | cowork: GP/PUMPCADE/STONK screen | Fake twin check: if two pools share a name and ticker and one shows a big market cap with about $0 pool depth, then ignore it and screen only the deepest real pool | GP twin 3SwCFPHSAkpCC9UCACgEzAa9C5raa5gHEMooyFoM8pBM printed $45.7M with $0 pool vs real HTmQz7My6MehV7bjhJ6jde8nDND1yvsz68d24LP7YgUQ at $6M with $270k pool; n=1, 1 Oct 2026 | TESTING - [stated] 2026-10-01 | cowork: GP/PUMPCADE/STONK screen | Distribution trap: if buy count beats sell count while price falls (1h under minus 3% or 6h under minus 8%), then WAIT, someone is selling into the buyers | GP 20:28 BKK 1 Oct: 1h buyers 132 vs 127 while price fell 9.6% in the hour; n=1 | TESTING - [stated] 2026-10-01 | cowork: GP buy scan (gpscan.py, hourly scheduled task "GP hourly update") | GP entry: BUY if mcap $2.1M to $3.5M with 3 hourly lows above $2.1M plus more buyers than sellers 1h plus net money in; BUY if 3 hourly closes above $5M plus more buyers than sellers 1h and 6h plus net money in plus 6h not down; STRONG BUY if 3 closes above $9M plus 1h volume 2x the 24h hourly average plus net money in; SELL if 3 closes under $2.1M (27 Sep low) | levels from the GP chart: ATH about $34.4M 23 Sep, low about $2.1M 27 Sep, $5.7M at setup; n=0 outcomes yet, score from the hourly runs | TESTING - [stated] 2026-10-01 | cowork: GP/PUMPCADE/STONK screen | Casino over tables: if a single coin and its launchpad token are both candidates, then prefer the launchpad token that earns from every coin (UM casino vs tables) | STONK held a $0.22 to $0.40 range for 3 weeks while its own coin GP fell 84% from ATH in 8 days; n=1 pair, Sep 2026 | TESTING - [stated] 2026-10-01 | cowork: STONK screen | Buyback vs market cap: if a token's verified open-market buyback, annualised, is close to or above its market cap and revenue has held 4+ weeks, then buy near the range low; kill if daily revenue stays under $500k for a week or price closes under $0.20 | STONK 6GmAFSYs4gk3FDao5FzzySQpPZaWsa4rUJHacpMpUNgx: DefiLlama 30d revenue $27.4M, 24h $0.94M; stonkfun API lifetime buyback $17.9M (188M tokens), about $550k a day vs $187M mcap; revenue was only $52k a day on 1 Sep, so 4 weeks of history only; n=1 | TESTING - [stated] 2026-10-01 | cowork: PUMPCADE screen | Stale catalyst: if a call's catalyst has been "soon" for 2+ months and daily volume has decayed below about a quarter of pool depth with lower highs, then no buy until the catalyst actually ships | PUMPCADE Eg2ymQ2aQqjMcibnmTt8erC6Tvk9PVpJZCxvVPJz2agu: @_0xghost_ called it 29 Jul at about $14.8M on "real money betting soon", $10.0M on 1 Oct (minus 32%), volume about $96k a day vs $444k pool, highs falling from about $50M in April; n=1 | TESTING - [stated] 2026-10-01 | cowork: GP screen | Reward-tax coins: if a stonk.fun reward coin charges a 3% transfer tax, then a trade needs about 6% just to break even and the holder yield does not hold the price up | GP paid $3.55M in GLDx to 15,217 holders (stonkfun /api/rewards, 1 Oct) yet sat 84% under ATH; same pattern as AGI in [[stonkfun-holder-apr]]; n=2 | TESTING - [stated] 2026-10-01 | cowork: GP/PUMPCADE/STONK raw data | notes.json (every number and full address used above) and gpscan.py (the hourly scan) in /mnt/user-data/outputs/gp_session_20261001/ of this Cowork session (lives with that session, not a public URL). Public copy planned at http://159.223.88.168/hx-c9ois/edge/gp-20261001/ once the Mac is online; not written yet - 2026-10-01 | cowork: 10x then higher base study | If a coin did ~10x, dumped 60 to 90% and holds above its old base, that alone is NOT a runner signal; a 2 to 5x first run comes back just as often | receipt: 527 screened coins, 4h candles to 1 Oct; clean set (tops after Ben screened the coin) new ATH 36% after 7 to 15x vs 34% after 2 to 5x, n=39/86, entry at first touch of minus 60% | KILLED - 2026-10-01 | cowork: dump depth | If bought at first touch of X% below the top, then odds of a new ATH fall with depth: minus 40/50/60/70/80/90% gives 41/35/29/22/13/3%; at minus 60% about 1 in 10 does 10x from entry | receipt: clean n=105/94/85/74/62/31, Aug to Sep 2026, mostly RH chain; full history same slope, higher level | PROVEN - 2026-10-01 | cowork: base speed | If a dumped coin has not restarted within about 2 days of touching minus 60%, odds collapse; still 60 to 90% down after 3 to 14 days gives 2% new ATH, after 14+ days 0 of 8 | receipt: winners break ATH median 2.0 days after the minus 60% touch (n=178 all), 0.7 days (n=25 clean); clean base length n=47/40/8 | PROVEN - 2026-10-01 | cowork: first dump vs repeat | If it is the first dump after the first run, then 40% new ATH and 33% reach 1.5x the old ATH; second or later dumps 24% and 13% | receipt: clean n=30 vs 55; full history 54% vs 48% | TESTING - 2026-10-01 | cowork: base market cap | If the base is $10M+, then skip (0 of 10 clean, 11% full); $0.5M to $2M band gives 31% new ATH, 16% big run, 9% 10x at minus 60% | receipt: clean n=10 and n=32 | TESTING - 2026-10-01 | cowork: height above old base | If bases are ranked by how far above the old base they sit (2 to 4x, 4 to 8x, 8x+), there is no signal (29/35/26% new ATH) | receipt: clean n=24/23/38 | KILLED - 2026-10-01 | cowork: survivorship method | If a study uses coins screened because they were hot, then count only moves AFTER the screen date, or hit rates come out about 2x too high | receipt: minus 60% entry new ATH 50% on full history vs 29% clean, same coins | PROVEN - 2026-10-01 | cowork: coin-specific MA | If an MA held 2 or 3 times on a coin, it holds no better next time than that coin's other lines or a fake copy shifted 3 to 10%; shifting any MA gives a smooth curve with no bump at the real line | receipt: 495 coins, 386k touches, 23 lines (SMA/EMA 9 to 200, 3 VWAPs), 15m/1h/4h; best line per coin 74 to 90% in sample, 46 to 67% after vs other lines 45 to 68%; top-quarter "respecter" coins +9 pts first half, -1 to -2 pts second half | KILLED - 2026-10-01 | cowork: MA held then ATH retest | If a line held and the coin then made a new ATH, the retest still does not beat a fake line | receipt: 1h 72% vs 75% fake, 296 coins | KILLED - 2026-10-01 | cowork: crowded line | If a line has been touched 6+ times without breaking, it fails MORE than a fake line with the same history (stops cluster there), so bid 3 to 5% under obvious lines | receipt: real minus fake hold rate -13 pts 15m (238 coins), -5 pts 1h and 4h | TESTING - 2026-10-01 | cowork: touch exhaustion | If any support has been touched 5+ times, cut size; hold rate is flat for touches 1 to 4 then falls, but fake lines fall the same, so it is trend age, not the line | receipt: 4h coin-balanced 70/72/72/71/68/62% for touches 1 to 6+, 178 coins | PROVEN - 2026-10-01 | cowork: flat support | If an old swing low has already held 3+ times, touches 4 to 6 hold about 5 pts better than fake levels (it gets stronger, not exhausted) | receipt: 4h 72 to 75% vs 66 to 70% fake; 15m 53 to 56% vs 44 to 53%; no edge on 1h | TESTING - 2026-10-01 | cowork: diagonal trendline | If buying the first touch of a fresh rising trendline, odds are below a coin flip (34 to 46% hold); small edge only after many touches, thin samples | receipt: 15m/1h/4h, 440 to 491 coins | KILLED - 2026-10-01 | cowork: trend filter | If price is above 1.5x its 4h 200 MA, a dip touch holds 74%; below the 200 MA 58%; within 15% of ATH 77%; fake lines move the same, so trade the trend, not the line | receipt: 4h, 89 to 167 coins per bucket | PROVEN - 2026-10-01 | cowork: confirmation and volume | If waiting for the touch candle to close back above the line, or for a volume spike on the touch, there is no gain once win/loss is measured from the entry | receipt: 4h close-above 61% vs close-below 64%; volume terciles 68 to 71% | KILLED - 2026-10-01 | cowork: method, fake-line test | If testing any support or indicator, compare against the same line shifted 3 to 10% and score coin-balanced and walk-forward; pooled numbers flipped sign once coins were balanced | receipt: 15m pooled edge +3 pts vs coin-balanced -3.6 pts | PROVEN - 2026-10-01 | cowork: holdings call | Live calls to score later: SELL INU 0x63Ee32Ac3077d1fbd8a77eBBA2a6ed4b8e9c1e18 ($4.5M, -76% from top, 12 days) and SCHIFFY 0x42aFA2124ca5a2B83898E46B2dA9a190995b1E18 ($3.75M, -71%, 11 days); HOLD HOOKR 0x18E674231A58c239Dc7DaeDcffE15Ec3A24cff5c (stop under $11.9M to $12.6M) and DELTA 0xe8ffd7e24187F72afB08d75B1bb13088A989a791 (stop under $12.3M); SI (Solana) DEW9dSN6QpWyNthphCpMmAbZP1Q4cEKR9xQXAri98WDP add at $36M to $38M (floor held 3x) and about $28M, exit if no new high within 2 days of the minus 60% touch | receipt: prices 1 Oct 17:00 Bangkok, rotation.json word BUY TRENCHES | TESTING - 2026-10-01 | cowork: ticker check | If a ticker exists on two chains, read the chain from the fomo position link before any analysis | receipt: SI was first analysed on Robinhood Chain (0xd36b68E9f384bD2b94bAF49c5e084b3B88fa1E18) while Ben holds the Solana SI | PROVEN - 2026-10-01 | cowork: data sources | GMGN token_kline (droplet key /opt/runwatch/gmgn_key, needs timestamp + client_id) is the fast candle source, about 15 coins/min, capped at 1000 rows (15m = 10 days, 1h = 41 days); GeckoTerminal about 5 coins/min per IP; DexScreener has dropped many dead RH pairs, so use GT /tokens/{ca}?include=top_pools; Dune out of credits 1 Oct | receipt: this session's fetch of 691 coins | PROVEN - 2026-10-01 | cowork: pattern + MA study raw data | /mnt/user-data/outputs/pattern-ma-study-2026-10-01/ of this Cowork session: ohlcv4h.tgz (4h candles, 527 coins, GeckoTerminal), kl_15m_1h.tgz (15m + 1h candles, 691 coins, GMGN), scripts.tgz (coins_all.json coin list with chains, dates.json screen dates, study.py, ma_study.py, horiz.py, diag.py, boot.py, t4b.py, profile.py and fetchers); lives with that session, not a public URL. Copies also on droplet 159.223.88.168 /root/kl and /root/gtlane*/out, and Mac ~/kl and ~/gtlane*/out - 2026-09-27 | chat: RH macro, stablecoins | If Robinhood Chain stablecoin supply is high, then do NOT read it as cash waiting to buy memes; about $855M of $1.07B sits in Morpho (Robinhood Earn USDG savings + Ethena USDe loops), less than $1M borrowed against stocks or memes | receipt: Morpho API + DefiLlama, 27 Sep 2026, n=1 snapshot | PROVEN - 2026-09-27 | chat: RH macro, bridge exits | If ETH leaves the canonical bridge escrow (Ethereum 0xDf8755334ce7A73cCF6b581C02eA649AE3E864b3), then check who withdrew before calling it an exodus; 9,708 ETH out in 31h was 99.9% two wallets (0x3723dc5372e4be3b1312fe6a5e93ce3eacfd2166 whale 6,090 ETH; 0x65c340eb0688d8f8b64e9c6d580f312a97b615d8 OKX 3,617 ETH) | receipt: L1 BridgeCallTriggered logs via ethereum-rpc.publicnode.com, 26 to 27 Sep 2026, n=1 | PROVEN (method) - 2026-09-27 | chat: RH macro, absorption | If chain activity falls hard while a frozen top-25 basket (picked at the peak) holds, most coins make higher lows and up/down volume is 50/50, then sellers are exhausted at that level; it is NOT proof of a price bottom | receipt: RH 7 to 26 Sep 2026: Pons fees -80%, frozen top 25 at 93 vs 15 Sep low 80, 33 of 51 coins above mid-Sep low; price impact per $1M up 2.6x (thin books) | TESTING - 2026-09-27 | chat: launchpad crash timing | If a launchpad's 7 day average fees fall 75%+ from peak, then expect the activity bottom around day 55 to 90 (median about 69) and a total fall of 85 to 99% | receipt: 20 episodes pump.fun, four.meme, Clanker, letsbonk (DefiLlama fees), 2024 to 2026 | PROVEN (base rate) - 2026-09-27 | chat: young chain first peak | If a new chain's DEX volume falls hard after its launch peak, then a reclaim of that peak, if it comes, takes 2 to 9 months, never weeks; about half of chains never reclaim (Blast, Berachain, Sonic, Plasma, MegaETH) | receipt: 17 chains, DefiLlama DEX volume; survivors reclaimed in 66 to 287 days (Base 198) | PROVEN (base rate) - 2026-09-27 | chat: price vs activity bottom | If meme activity is crashing but prices hold, then do not assume prices bottomed first; in Solana's 2025 crash meme prices bottomed about 2 weeks AFTER pump.fun fees, -79% | receipt: Binance basket WIF, BONK, PNUT, BOME, PENGU, FLOKI, PEPE vs pump.fun fees, ep1 Jan to Apr 2025; ep2 ambiguous; n=2 | TESTING - 2026-09-27 | chat: RH vs BTC | If judging Robinhood Chain coin prices, then they follow BTC (daily correlation 0.5, both +13% 17 to 22 Sep) but chain activity does not (Pons fees halved during BTC's rally) | receipt: frozen top 25 vs BTC, 7 to 26 Sep 2026, 19 days | TESTING - 2026-09-28 | chat: runner framework, volume drying | If a coin that peaked $1.5M+ is down 70%+ and this week's volume is under 0.6x last week's, then 10x odds within 30 days rise | receipt: RH 133 dips 17% vs 4% (Jul to Sep 2026); Solana 1,275 dips 1.5% vs 0.9%; 26 winners total | TESTING - 2026-09-28 | chat: runner framework, deep drop | If the dip is 85%+ off the peak (vs 70 to 85%), then 10x odds about double | receipt: RH 14% vs 6%; Solana 1.3% vs 0.7%; same samples; note the cowork dump-depth line above measures odds of a new ATH from a minus X% entry, a different question | TESTING - 2026-09-28 | chat: runner framework, quiet beats loud | If the project's X likes per $1M market cap (7 days before the dip) are in the top third, then 10x odds are about half | receipt: RH 5.3% vs 10 to 13%; Solana weighted sample 0.3% vs 0.7 to 1.1%; X read as of the dip date via twitterapi.io | TESTING - 2026-09-28 | chat: runner framework, no X account | If DexScreener lists no X account for the coin (or it is gone), then skip it | receipt: RH 0 of 9 dips did 10x; Solana 2.0% vs 0.8% base (no edge there); small n | TESTING - 2026-09-28 | chat: runner framework, Loco Motif bounce | If a dumped coin bounced 3x+ off its floor and is back at that floor, then it is a 10x setup | receipt: RH 15% vs 4.5% (19 vs 114 dips) but Solana 0.3% vs 1.2%; did not replicate | KILLED (as general rule; RH-only bonus at most) - 2026-09-28 | chat: runner framework, curl / higher lows | If price is 30%+ above its 14 day low or 3+ of the last 5 days closed higher, then pick it | receipt: no edge on either chain (RH 8.7% vs 10.9%; Solana flat) | KILLED (use for entry timing only) - 2026-09-28 | chat: runner framework, posting cadence | If the project's X posts often (4+ a week), then better 10x odds | receipt: RH 10.1% vs 20% for 0 posts (n=10); Solana 0.4% vs 0.7 to 2.6%; no edge | KILLED (dead-project filter only) - 2026-09-28 | chat: runner framework, contract shill count | If many 200+ follower accounts posted the contract address in the week before, then worse odds | receipt: RH 5+ accounts 3.3% vs 0 accounts 19.2%; Solana unclear; live counts on 28 Sep were about 30 to 38 for almost every coin (bot farms) | KILLED (unusable live) - 2026-09-28 | chat: runner framework, size | If the dip sits under $600k market cap, then better odds | receipt: RH better, Solana worse; no consistent edge | KILLED - 2026-09-28 | chat: Sparsity "ranges before runs" | If a coin has gone sideways 3+ weeks inside a 2x band, then it is NOT more likely to 10x; most coins range most of the time, the post studied only winners | receipt: 10x within 60 days: Solana 1.4% in range vs 2.2% not (3,023 samples); RH 0% vs 8.8% (310) | KILLED - 2026-09-28 | chat: dip under the range | If price dips below a 3 week range then closes back inside it, then 10x odds are slightly higher | receipt: Solana 1.8% vs 1.2% (169 vs 966 samples, 3 winners) | TESTING (likely noise) - 2026-09-28 | chat: range breakdown | If a coin breaks below its range, then it usually does not collapse; only 16% fell another 50% within 30 days | receipt: Solana 424 breakdowns | TESTING - 2026-09-28 | chat: raw data | Bundle /mnt/user-data/outputs/rh-macro-runner-data-2026-09-28.tgz of that chat (not public): RH 798 + Solana 411 coin daily OHLCV, dip episodes, X-as-of-dip, range test, scripts - 10-03 cowork: X talk vs tops, 78 coins: talk/arrivals extension sell rules near top 19-30% vs random 15%, price extension 50-67%; talk peak = top week; social NOT sharper than price [[macro-direction]] v7 KILLED - 10-04 cowork: SOL stables +8%/14d -> crash odds 3x; doc hx-c9ois/research/real-money.html TESTING - 10-04 cowork: big-post views peak -> SELL SOME, 48% before top (price 9%); exposure.py v8 TESTING - 2026-10-01 | chat: Birdeye wallet screen ([[copytrade-targets]]) | If a wallet list comes from Birdeye "profitable traders", then rank on COUNT of $10k+ winners and top-name concentration, never on headline net PnL; the filter sorts by dollars so it puts one-trade wallets above repeatable ones | receipt: 8 Birdeye RH wallets, full ledger rebuilt on chain 1 Oct 2026; 5 of 8 had one name carrying 85 to 100% of net; the best wallet (0xF86a7c853B60C36f79F6E1dE0E7868d109E97004, ten $10k+ winners, nine exits above 3x, median entry about $550k mcap) ranked mid-pack by dollars | TESTING - 2026-10-01 | chat: wallet concentration test | If a wallet's top position net profit is larger than its total net, then every other position combined loses money and it is a lottery winner, not a process | receipt: 0x1d357f7492A5d01d1C7CA1cFbB7297ed06e34434 net +$104,165 with microduck alone +$123,236 over 104 positions, 33 wins; same shape on 0xDFA2e325f31D5c5CB4A093c9845555BE8764ED28 (BUDDY 94% of net, 56 sell legs in 0.9 days, token now 99% below exit) | PROVEN - 2026-10-01 | chat: dex.trades undercount trap | If a Robinhood Chain wallet ledger is built from Dune dex.trades, then buys go missing (launchpad bonding-curve fills and ERC-4337 relayed txs do not record the wallet as taker); rebuild per tx from erc20_robinhood.evt_Transfer plus robinhood.traces native value, then price the quote leg | receipt: dex.trades showed 0xDC60c2d2deBfA058F86005374EDaaCf453130688 selling 1,398% of what it bought and 0x0f02aFcB1550fbEAFAF001251b06bA73c68A5384 selling PONS with zero buys; rebuild fixed both, 8 wallets, 15,611 flow rows | PROVEN - 2026-10-01 | chat: entry market cap with no extra API | If entry quality is the question, then derive it: avg_entry = buy_usd / qty_bought, om = price_now / avg_entry, entry_mcap = mcap_now / om, exit_mcap = entry_mcap x (avg_exit / avg_entry); one batched DexScreener call prices the whole set | receipt: 271 tokens, 240 priced in 9 calls (tokens/v1/robinhood, 30 per call, browser User-Agent or it 403s every time), 1 Oct 2026 | PROVEN - 2026-10-01 | chat: transfer leakage test | If a position shows 0% sold but a zero current balance, then the tokens left by plain transfer and that wallet's PnL is not self-contained; drop it as a copy target | receipt: 0x0f02aFcB1550fbEAFAF001251b06bA73c68A5384 bought $39,934 NET and $120,180 HMM, sold neither, holds neither; its headline net was +$39,623 on $746,888 deployed | PROVEN - 2026-10-01 | chat: alert floor vs clip size | If a copy target is added to an alert bot, then set the floor from ITS OWN entry clips, not the fleet default; the best early wallets enter small and a blanket floor hides exactly the trades worth copying | receipt: 0xF86a7c853B60C36f79F6E1dE0E7868d109E97004 entered TWO at about $850 a clip (14.7x), CHIPS $734 (6.3x), NTF $1,943 (7.8x), BONER $2,488 (41.1x); fleet floor was $1,000 | TESTING - 2026-10-01 | chat: wallet selectivity filter | If a candidate wallet touches many tokens, then count distinct tokens, txs and active days before calling it smart money; a wallet doing 170 txs a day against 41 real swaps is running a script, and one good trade inside that is not a process | receipt: 0x736841752C5380187Ba8a520ab4b60347F311200, 9,368 txs over 55 days vs 41 DEX taker trades, 374 tokens, PONS 85% of its +$940,449 net | TESTING - 2026-10-01 | chat: late-entry whale tell | If a wallet's median entry market cap is in the tens of millions, then it is a size player buying majors late, not an early wallet, however large its net; check entry mcap before copying | receipt: 0x342744fcDfA9742bB17773F7caf60f75f83B8125 deployed $1,486,956 in 6 days at AI $105M, CASHCAT $217M, STONKBROKER $38M, DELTA $31M (minus $23,762); only BONER ($5.7M entry, 10.8x) was early, and $1.88M of its net is an open mark, not banked | TESTING - 2026-10-01 | chat: live call to score | Added 0xF86a7c853B60C36f79F6E1dE0E7868d109E97004 ("Birdeye Top Trader") to strongtrader.service on kookwatch, per-wallet min_usd_buy $400 and min_usd_sell $750 (tracker.py patched to read per-wallet thresholds, backup tracker.py.bak-20260901-064746). At add time: deployed $372,859, net +$351,605 in 29 days, $388,011 open across VLR, Rabbit, TWO, BONER, QUOTRON, PROLOGUE. Score its next 10 alerts on entry mcap and whether it holds past 3x | receipt: screened 1 Oct 2026, no forward outcome yet | TESTING - 2026-10-01 | chat: Birdeye wallet screen, raw data | Container files gone (ledger.json, flows.json, wal.json, px_now.json, degree.json, dest.json in /home/claude/wf/, not recoverable); screen note kept in the wallet's entry in /opt/strongtrader/config.json on 159.223.88.168; reproducible from the dex.trades-undercount and entry-mcap lines via Dune standing query 8562361. Dune gotchas: "from"/"to" reserved, addresses unquoted hex, traces.value uint256 so cast(0 as uint256), dex.trades covers blockchain='robinhood' (211M rows) but see undercount line - 2026-10-01 | chat: Arkham | If identity resolution is needed for a wallet, then say the gap out loud: there is still NO Arkham key anywhere in the stack (skills, droplet /opt configs and .env, Mac ~/kookwatch, ~/.config, ~/.zshrc all checked 1 Oct 2026). research-sources lists Arkham under keys Ben does not own | PROVEN - 2026-10-02 | chat: ray breakouts v2 | If price closes 2% above a ray (big high to next big high), then do NOT buy on that alone: 41% vs 47% random win, CI -8 to -4 pts, -0.38R; a ray moved 5% up does as well (45%), the line has no power | receipt: 3,256 breaks, 159 coin-timeframes, RH/Sol/Base, 1m to 1h, to 1 Oct | KILLED - 2026-10-02 | chat: ray, 1 leg no lower low | If the ray breaks after one lower high and no lower low, then skip: 22% vs 34% random, CI -15 to -10, all timeframes | receipt: n=1,422 | PROVEN - 2026-10-02 | chat: ray details | 2 legs + lower low ~random (n=116, v1 81% was a fluke) KILLED; corrected line worse each redraw (-7/-14/-18 pts) KILLED; low volume break -10 pts PROVEN; ATH break = random (70% vs 71%, n=322) KILLED; up channel 1h +7 (n=114) and retest-by-base-low TESTING | detail + raw data /mnt/user-data/outputs/breakouts-2026-10-01/ (chat, not public) - 2026-10-02 | chat: reversal zones ([[reversal-zones]]) | If buying a zone, then the exact line (order block, golden pocket, fib, range top, EMA) adds ~0 vs a random level at the same depth and time; OB 1h/4h -3 to -5 pts (n=4,130) KILLED; EMA20/50 touches -11 to -15 KILLED; GP 0.618-0.66 0 to +8, not special (conflicts with the SI GP line above, which also missed) | receipt: 128 coins 1h, limit -15%/+30% 7d, control = same coin, same time, level shifted 3 to 15% | KILLED - 2026-10-02 | chat: what does help a dip buy | Depth: 50 to 70% off ATH best (+0.30R, CI +0.17 to +0.44), 85%+ worst (-0.15R); RSI 1h<30 beats random closes nearby by +9 pts (n=1,378), +12 above EMA200; old high retest with RSI 1h<35: 45% win, +0.60R (n=489, 61 coins) | receipt: same study; absolute R survivor-biased up | TESTING - 2026-10-01 | chat: fomo top 50 copy test, chains | If copying a top fomo trader, then read their Robinhood Chain wallet first; most of the 30d top 50 trade there, not on their verified Solana wallet | receipt: FomoScan 30d board (1 Oct 00:04 Bangkok), 1 to 16 Sep: 2,702 real Robinhood Chain first buys across 49 traders vs 484 Solana first buys across 20; 21 of 50 had 0 Solana buys | PROVEN - 2026-10-01 | chat: fomo top 50, board shape | If a fomo board trader has few trades and a huge return (Visi235 43 trades +4,917%, ExactTallTakin 189 +737%, vladsbutler 74 +319%, Jols 111 +313%), then treat as picky and copyable on paper but possibly one lucky coin; many trades with small return (frankdegods 1,482 +9%, derek518 1,045 +11%, 0xsneakky 3,354 +17%, dreamloaderxo 5,842 +22%) means a grinder, hard to copy | receipt: 30d board, 50 traders; copy outcome (buy +1 min, hourly to 336h) not measured yet | TESTING - 2026-10-01 | chat: spam airdrop trap, Robinhood Chain | If counting a wallet's buys from incoming token transfers, then keep only transfers sent by the fomo router 0xb92fe925dc43a0ecde6c8b1a2709c170ec4fff4f or PoolManager 0x8366a39cc670b4001a1121b8f6a443a643e40951; spam contracts (e.g. 0x0048236efc26147a1f1f61512495f9ad6a9161db, 0xe3734208d8ccaf5ed60f777a344bbd6db168a2ed) push free tokens to 5 wallets per tx with no payment | receipt: 21,406 first arrivals, only 2,702 real buys, 49 wallets, 1 to 16 Sep 2026 | PROVEN - 2026-10-01 | chat: data sources, container | Robinhood Chain RPC getLogs caps 30,000 blocks (about 50 min) unless filtered by address; put all wallets in one OR topic list (49 wallets, 30 days = about 860 calls, 15 min). Solana old txs: api.mainnet-beta works, publicnode returns empty (1,919 of 6,884 empty there, 0 on mainnet-beta). GeckoTerminal from the container: 429s above about 10 to 12 calls a minute. FomoScan: board about 500 CU, handle 100 CU, batch PnL 30,000 CU per trader (never use) | receipt: live calls 1 Oct 2026 | PROVEN - 2026-10-01 | chat: fomo top 50 copy test, raw data | temp chat container /tmp/f50/ (ids, RH+Sol tapes, first buys, candles, scripts), gone; FomoScan boards at Mac /tmp/fomo50/boards.json; copy test unfinished - 10-04 cowork [[system-benchmark]]: 112 systems x 15m..1d, RH/SOL/Base, 3 mo, 2% cost: RH all lose (15m worst); SOL 1h-12h dip buy +2-5%; Base 4h-1d trend (SuperTrend 7,3, MACD) +25-30%, holds 1 bar late PROVEN; 72h losers rotation 8,800x = survivor fake KILLED - 2026-10-02 | chat: copy frankdegods? | If a KOL's PnL depends on copiers buying right after him, then expect it to vanish without them; @frankdegods (532k followers) traded a private wallet with no copiers around Apr 2026 and reportedly lost (seeexbt, 10 Apr 2026), then went public again and grinds: 30d fomo board 1,482 trades, +9% return, 24h flips, sold STONK and Niu Lai at local bottoms by his own account, copiers say his exits kill the chart; he posted "don't copy my wallet" 18 Aug 2026 | receipt: X via twitterapi.io 2 Oct 2026, no own-tape measurement yet; test = his entry-to-exit mcap vs price 1h after his exit, from his fomo wallet on the Mac | TESTING - 2026-10-01 | chat: PRIORS screen (0xeDBf91223639800BCd5756815CAf908Df3b890bE, RH, AI agent credit) | If a token's story is "fees buy and lock the token behind loans", then read the protocol contracts from the site's config.js and check balances by RPC before believing it; PRIORS at $8.9M had $660 USDG in its lending pool (0x0259889e6EBab1a18CeE7e62Bc5B9648FB6C44e5) and 0 PRIORS in its seat vault (0xb1c3a04496238D62E3c93118C297163855e22192), "14.9K loans repaid" are $5 lines | receipt: RPC eth_call 1 Oct 22:30 Bangkok; comps ORBIO $84M, WALLET $23M, HARMONIC $6.2M, NAVEN $1.9M; call: no chase at $8.9M after +133% 24h, buy zone $6.0M to $6.5M, wrong under $4.9M 4h close; Blockscout was Cloudflare 403 from the container, GeckoTerminal /tokens/{ca}/info gave holders 2,130 and top 10 at 34% | TESTING - 2026-10-01 | chat: MOO 0xD9dB30BB0D2b8d2eae3826A1372117E058791e18 bottom call | If a flat floor that held 3 times breaks on about 10x normal 4h volume, then do not trust the 1h bullish RSI divergences that follow; each one broke to a new low | receipt: MOO floor about $10.5M held 27 to 29 Sep, broke 30 Sep 03:00 Bangkok on 241k 4h volume vs 8k to 50k before; 1h divs at $7.65M (RSI 17), $5.81M (20), $5.30M (26) all failed, low $4.64M 1 Oct 19:00 Bangkok; ATH $44.4M 10 Sep, -89%; 4h order block (last red candle before the 31 Aug launch) $3.25M to $3.75M; n=1 | TESTING - 2026-10-02 | chat: fomo picky winners, own-tape check | If a fomo board trader shows few trades and a huge return, then expect ONE coin to carry the whole net before copying: Visi235 = AI (RH) $985k sold in 42 clips, nothing else above $10k; ExactTallTakin = SI (Solana) $43k in from 0.6h after pool open at ~$2.4M mcap, $869k out at ~$48M, 20x, 8 day hold, rest are $3 to $500 same-day flips; vladsbutler = PONS $267k in at ~$195M, $594k out, 2.2x, 18 day hold, late size player; Jols = PONS carry plus dust grinding, entries on $12M to $270M coins, 1.0 to 1.3x. Copier boost absent on all four: median price 1h after their buy clips x0.86 to x1.02. Only Exact's first-hour entry is an edge; watch AGdtsMmphhymH5wYW3dWKSkAxpLYRrNDCTFAfuq8ii9v for first-hour buys only | receipt: 35 day RH chain tape (fomo router 0xb92fe925dc43a0ecde6c8b1a2709c170ec4fff4f, USDG quote custodial via 0xf70da978 in and 0x4cd00e38 out) plus full Solana tape, GeckoTerminal hourly candles, 2 Oct 2026; Visi AI and Jols PONS buys predate the window, older scan unfinished | TESTING - 2026-10-01 | chat: floor break, volume test ([[positions-manager]]) | If a 4h floor that held 3+ times breaks on a close, then sell without waiting for volume; on MOO the 3x volume demand delayed the sell from $21.3M (12 Sep 23:00 Bangkok, 0.8x volume) to $7.8M (30 Sep); buy back only after 2 closes above the lost floor | receipt: MOO replay, n=1 coin, agrees with the KILLED volume-confirmation line | TESTING - [stated] 2026-10-02 | cowork: SI live call | SELL SI DEW9dSN6QpWyNthphCpMmAbZP1Q4cEKR9xQXAri98WDP at about $35M: the floor-break rule fired (the $35.7M to $36.6M 4h floor held 5 times 30 Sep to 1 Oct, then the 1 Oct 19:00 Bangkok 4h candle closed at $32.7M); buy back only after 2 4h closes above $36M; next level $27.6M (minus 60% of the $68.9M ATH, 30 Sep 19:00) | receipt: GeckoTerminal 4h, pool AG1EhPjsBrkxQWkViuLXF6pQnHwfr9f9sdNneeoykcEJ, 1 Oct 23:20 Bangkok; Ben's entry $53M; 1h 80 buys vs 201 sells, minus 13%; score the price 2 days later | TESTING - 2026-10-01 | chat: alts vs onchain rotation (Ben's idea) | If alt market cap pulls back, then onchain/trench coins turn up, and vice versa (they alternate) | receipt: Ben's observation 1 Oct, unmeasured; dashboard that day said BUY TRENCHES, launchpad volume +68% 14d, so "onchain pulling back now" was not in the data; test = TOTAL3 vs launchpad fees 2024 to 2026 | TESTING - 10-03 cowork: KingCobra full X+TG 2023-26: direction, targets, HTF calls all = random (BTC 90d 50% vs 51%, n=382) [[kingcobra-trades]] KILLED - [stated] 2026-10-02 | cowork: SI golden pocket (Ben's read) | If a coin pulls back into the 0.618 to 0.66 retracement of its pump and holds it several times with buyers returning, then a small buy with a stop just under 0.66 is allowed before the 2 close confirmation; fib extension targets from that low are 1.0 and 1.618 | SI DEW9dSN6QpWyNthphCpMmAbZP1Q4cEKR9xQXAri98WDP: pump $13.4M (29 Sep 19:00 Bangkok) to $68.9M (30 Sep 19:00); pocket $32.3M to $34.6M; low $32.0M to $32.3M held 4 times 22:45 to 00:00, bounced to $36.5M by 00:25 2 Oct, low about 29h after the top (inside Ben's 2X duration rule); targets about $87M (1.0) and $122M (1.618); fib not yet tested against fake shifted lines, n=1 | TESTING - [stated] 2026-10-02 | cowork: SI bottom check | If whale sell flow fades but holder count still falls, then call it a pause, not a bottom; wait for 6h holder change to turn positive (Jupiter datapi.jup.ag/v1/assets/search stats6h.holderChange, free, no key) | SI 2 Oct 05:17 Bangkok: net $1k+ flow -$239k 21:00 to 00:00, about flat since; holders 58,919, -3.5% 24h, -1.4% 6h | TESTING - [stated] 2026-10-02 12:32 Bangkok | cowork: SI golden pocket update | Pocket ($32.3M to $34.6M) failed: one 4h close above $36M (03:00, $38.5M) then back to $35.1M (07:00), new low $31.6M; holders 57,762, -1.8% 6h; $1k+ net selling 6 of 7 hours | score the golden pocket line as a miss if the 15:00 4h closes under $31.5M - 2026-10-02 | chat: dip buy v2 ([[reversal-zones]]) | If a coin is 50 to 70% off ATH and falls back to an old top it broke (by 15%+) with RSI 1h<35, then limit buy there, stop -15%, target +30%, max 7 days | receipt: 177 coins, after 2% costs +0.33R (CI +0.04 to +0.61), n=734, both halves; RH +0.64R, BSC +0.42R, Solana/Base no edge; only +6 pts vs shifted level; order blocks and EMA touches -0.22R | TESTING - 2026-10-02 | chat: Ben 61.8 dip buy ([[reversal-zones]]) | If a coin that pumped 3x+ first touches -61.8% from ATH, then limit buy, sell at -38.2% from ATH (+61.8%), max 14d; stop at -80% beats -75% | receipt: 275 coins 1h after 2% costs: stop -75% +10.2%/trade (CI +2.8 to +17.8, win 42%, n=131); stop -80% +16.2% (CI +8.9 to +23.3, win 53%); random entry same stops +5.3%; Base -4.9%; 98 new coins ~0% (n=24, too few); old-high+RSI plan held on new coins +6.4% (n=1,109, CI +2.4 to +9.5) | TESTING - 2026-10-02 | chat: Moneytaur OBs | If an UNTESTED 4h/7h/12h order block (down-candle BODY before a 2 ATR structure break) is first touched, then it bounces LESS than a random level 5 to 20% away; nested 1h/2h OBs inside make it worse | receipt: 275 coins, 1,297 4h touches: 24h bounce +5.1% vs +7.4%, +20% before -10% 18% vs 20%; nested 3+: 14% vs 18% | KILLED - 10-02 LONG eco bottom [[long-eco-bottom]] TESTING; ledger continues in [[trading-edge-hub-2]] - 10-02 cowork: sellers quiet>random; 1st dip after ATH, stop -80% from top: +19-32%; GP W30d high, BTC>100d avg, majors ON; half when hot; OFF: close<15d low; rotate 3 strongest 30d | 14 baskets 23-26: 118 tr +8.1% PF2.65, Sep24+ PF1.81 | TESTING; BUY TRENCHES KILLED [[rotation-watch]] - 10-04 MA stack: chop+4h apart lose [[ma-stack-study]]