---
name: trading-edge-hub-pro-trading
description: Pro Trading project's continuation of the shared trading insight ledger (main hub /areas/trading-edge-hub.md is full); fomo top 50 copy test lines. Read with the main hub.
sources: [chat]
aliases: [edge hub project copy, trading insights pro trading, insight ledger continuation]
---
Continuation of the main [[trading-edge-hub]] ledger, which hit its size cap. Same format: date | approach | rule | receipt | status.

# Insight ledger (continued)
- [stated] 2026-10-02 | chat: fomo top 50 copy test, Solana | If you copy a top fomo trader's Solana buy 1 minute later and hold, then expect to lose: the middle trade is minus 66 to 88% at 14 days for every trader with 10+ buys | receipt: 252 first buys, 20 traders, 1 to 16 Sep 2026, hourly prices to 336h (GeckoTerminal), no fees | PROVEN
- [stated] 2026-10-02 | chat: copy entry gap | If you copy a buy 1 minute later, then you pay 2 to 17% more than the trader did (median jump, notanicecat69 +37% on 6 buys) | receipt: same 252 buys | PROVEN
- [stated] 2026-10-02 | chat: copy exit timing | If copying anyway, then sell within the first hours; only Carver___________ was green at 1 hour (+23%, n=15); every trader with 10+ buys was red at the median by day 7; only 1 in 4 or 5 copied trades green at 14 days, averages carried by rare runners (dtrainz +86% avg, n=26, 27% green) | receipt: same set | TESTING
- [stated] 2026-10-02 | chat: fomo copy test, Robinhood Chain | Not finished: hourly candles for 278 coins nearly done, +1 minute entry still to price (plan: read from chain swap events, faster than GeckoTerminal) | receipt: 494 sampled buys, 12 per trader, 1 to 16 Sep 2026 | TESTING
- [stated] 2026-10-02 | chat: fomo copy test raw data | Container of this chat (temporary, not a public URL): /tmp/f50/sol_out.json, /tmp/f50/cache_solana.json, /tmp/f50/cache_robinhood.json, /tmp/f50/rh_in.json, /tmp/f50/sol_in.json, /tmp/f50/anal.py, /tmp/f50/pricer2.py, /tmp/f50/ids.txt
- [stated] 2026-10-02 | cowork: INU support retest (0x63Ee32Ac3077d1fbd8a77eBBA2a6ed4b8e9c1e18) | If a coin down 75% from ATH bounces off a double low and retests old support while wallet sellers outnumber buyers, then it is not a buy yet: wait for a 4h close back above the bounce high (~$6M) or buy small at support with a stop on a 4h close under the double low (~$3.8M) | receipt: n=1, 2 Oct 18:58 BKK, mcap $4.70M, 24h >$200 trades: 91 buyer wallets $305k vs 143 seller wallets $261k, net selling since 10:00 | TESTING
- [stated] 2026-10-02 | cowork: SHROOM sellerwatch W1 fire (0xab093def657f15df31b33922a95e047add645b29) | If sellerwatch fires BUY after a bounce has already happened (entry +33% off the 72h low, higher low held) and the coin's revenue has collapsed, then take it as a small buy only, using the alert's 30% stop (sits just under the real low) | receipt: n=1, alert 2 Oct 19:00 BKK at $12.59M, 7d high $31.56M (25 Sep), low $9.46M (1 Oct 22:00), flush only $8,766; DefiLlama holder revenue 7d $19,288 vs about $21k/day mid Sep (down about 90%); W1 backtest n=224 +14.2%/trade vs random +5.2% at 30/100 | TESTING
- [stated] 2026-10-02 | cowork: SELLER WATCH v2 strong-buy gate (Ben: "fix the alert so that it only gets strong buys") | Telegram now sends only STRONG BUY = W1c BUY plus strict quiet (big-seller selling since flush <= 10% of it) plus flush absorbed (pool buys >= pool sells in the flush hour) plus liquid (>= $100k pool volume 24h); plain BUY only in the 09:00 daily book | receipt: replay 2 Oct: SHROOM 19:00 alert becomes plain BUY (buys $9,512 vs sells $9,935, no alert); MOO 04:00 bottom (+50% after) also becomes plain BUY (since $14,813 = 22% of $68,299 flush; buys $143,945 vs $146,083), so the gate would have MISSED it; strict and liquid filters come from the 2,427-coin Dune test (W1 strict liquid +15.2%/trade vs random +3.8% at 30/100), the absorbed filter is untested beyond W6 n=28. Deployed /opt/sellerwatch/sellerwatch.py md5 41b0389f7f38c19be2d9b38cefb2ea49, backup sellerwatch.py.v1bak | TESTING
- [stated] 2026-10-02 | cowork: correction | "Close above the bounce high = strong buy" (said for SHROOM at $14.5M) contradicts the ledger: break-of-bounce-high W entries tested at or below random (KILLED). Do not use it as a strong-buy trigger | receipt: bottom-timing v3 W test n=118/78/40 | KILLED
- [stated] 2026-10-03 | cowork: BOTTOM LAB (Ben: keep fine-tuning the strong buy at bottom rule hourly on past and current coins until no further optimizations) | Droplet service bottomlab (/opt/bottomlab/lab.py, md5 of first deploy 31b7bd7c740f84030701b0704092223d): replays every RH coin in runtrack.tokens (522, dead ones included) from launch with rhledger's user-net method + per-wallet big-seller selling (prior 72h, $2k/$5k/$20k), own GeckoTerminal hourly candles (not GMGN: runtrack's limiter answered "bucket held"); hourly: live STRONG BUY check (takes over from sellerwatch via /opt/bottomlab/LIVE once the watchlist coins are built), random+local search over 17 rule settings, pick on coin half A, gate on half B (beat champion by 1.5pp and random by 5pp, both time halves, bootstrap floor), newest 20% of time held out; 24 runs without promotion = converged message, then every 6h; daily 09:00 summary; watchdog cron stale.py | receipt: OYSTER rebuild matched rhledger exactly (buy $1,487,236, sell $1,052,302, holders 309, 0 mismatched hours); synthetic test 262 configs in 3.2s | TESTING
- [stated] 2026-10-03 | cowork: droplet disk | 159.223.88.168 disk 98% full (1.9 GB free of 77 GB); /opt/price-predict-rh holds 41 GB (forward.db 30 GB stopped writing 30 Sep at its cap, exports 11 GB). BOTTOM LAB pauses builds under 1 GB free and messages once | receipt: df 2 Oct 18:04 UTC | TESTING
- [stated] 2026-10-03 | cowork: BOTTOM LAB v3, one replay | If two services replay the same RH logs, both RPC endpoints (rpc.mainnet.chain.robinhood.com, robinhood-rpc.publicnode.com) answer the droplet 429 and nothing finishes; so rhledger is the only replay and now also writes hourly per-wallet user nets (runtrack.db table rh_wnet, backup /opt/runtrack/rhledger.py.bak-wnet-1790979957); BOTTOM LAB reads runtrack.db read-only (writing its shared tables gave "database is locked" on every build) and computes big-seller features from rh_wnet; big-seller columns valid only 72h after a coin's wallet data starts (the 62 coins rhledger built before the patch become usable from about 6 Oct) | receipt: lab's own replay managed 0 finished coins in 4 hours (PONS 13M of 78M blocks in 15 min, memory peak 1.3 GB); batch eth_getCode returned 429 on both endpoints | TESTING
- [stated] 2026-10-03 | cowork: BOTTOM LAB cadence (Ben: "we want the metrics every 10 mins right?") | Live STRONG BUY check now every 10 minutes on the open hour (rhledger's finished 10-minute buckets + GeckoTerminal price snapshots of dumped coins), so alerts fire up to 50 min sooner; the rule SEARCH stays hourly, then every 6h once converged, because searching more often on the same past data only adds false winners | receipt: first 10-min pass 03 Oct 10:59 BKK, 91 coins updated in 7s; lab.py md5 8d9051aa5677aaf1042b3bc8d1b70512 | TESTING
- 2026-10-03 | chat: BOT / Dot.com screen ([[dotcom-bot]], Solana stonk.fun, SPCXx reward coin, mint 4WPn1oGVE7rza6PjFJwYpEXZn6qtjwxEKkCV9EtAxhUU) | If a reward coin is up 48 percent in 24h while holders fall (minus 3.7 percent) and the launch bundle (48 percent at ATH) has fully sold, then WAIT; buy zone $250k to $300k only with 6h holders turning positive, wrong under $200k 4h close; narrative hook verified (dot.com redirects to x.ai/bot) but no site, no X, dev has 19,346 mints | receipt: stonk.fun API, Jupiter datapi, twitterapi.io by mint, 3 Oct 11:30 Bangkok; n=1 | TESTING
- [stated] 2026-10-03 | cowork: BOTTOM LAB coverage + build order | rhledger built youngest coins first, so at 11:20 BKK 0 of the 25 watchlist coins (MOO, HOOKR, DELTA, SCHIFFY, INU, AI ...) had wallet data; patched (backup /opt/runtrack/rhledger.py.bak-prio-*) to build ONE priority coin per 10-minute cycle (watchlist first, then biggest ATH), capped at 240 s of the 480 s budget, rest stays youngest first; first priority = MOO. The only faster route is a keyed RPC (public RH RPC is rate limited, nodeflare public allows getLogs only with a free key); needs Ben to sign up | receipt: 92 coins with wallet data, 45 with full history, 22 already usable in the lab (2,960 valid hours); rhledger about 12 builds/hour, 526 left | TESTING
- [stated] 2026-10-03 | cowork: keyed RPCs for Robinhood Chain | Ben made Alchemy and dRPC keys, stored in /opt/common/rpc_keys.json (chmod 600, keys "alchemy", "drpc"). dRPC free: Robinhood Chain works (0.09 s per call vs about 1 to 1.7 s public), but eth_getLogs ranges over 10,000 blocks are refused and Solana is paid only; now first in rhledger's EPS rotation for eth_getCode and block anchors (backup /opt/runtrack/rhledger.py.bak-drpc-*), getLogs stays on the public RPC. Alchemy: key valid but ROBINHOOD_MAINNET not enabled on app fh9msiz5tezkkv5m as of 3 Oct 11:45 BKK (403) | receipt: live tests from the droplet 3 Oct | TESTING
- [stated] 2026-10-03 | cowork: BOTTOM LAB audit (Ben: "do 5 10 iterations to make sure everything is correct") | /opt/bottomlab/audit.py (invariants) and e2e.py (dry run on a copy of lab.db) now exist; run both after every change. Findings fixed: hourly pass took 589 s because GeckoTerminal 429s, now capped at 150 s so the 10-minute check is not delayed; a coin rebuilt from launch now gets its features recomputed (wallet history start tracked per coin); priority coin budget raised 240 to 360 s. Passing: wallet nets per hour equal buys minus sells (0 of 254 coin-hours off, first partial hour excluded); big-seller and buy-USD features recompute identically (0 of 117 hours); exit math; 24h alert dedupe; alert, daily and search paths run. Expected, not caused by these patches: rhledger forward gate failures on TROLL, DUCAT, Suited (supply drift 0.1 to 0.8%, then rebuild) | receipt: audit 3 Oct 11:40 BKK, 57 coins loaded, 0 champion trades yet (data too short) | TESTING
- [stated] 2026-10-03 | cowork: BOTTOM LAB upgrade round (Ben: 10 ideas, execute all) | Live in lab.py md5 0dbe6f50365c81d1f7f150b56b21b9cd: (1) SELL follow-up for every STRONG BUY when target, stop or max days hits (table closed); (2) chart link + promise of SELL in the alert; (3) rule search gained 3 wide-stop exits (50/100, 40/200, 25/40) and two 3-day exits (results testable after 3 days, not 7); (4) new condition: more buyer wallets than seller wallets since the dump; (5) promotion must also beat random on the newest 20% of time when it has 10+ trades; (6) daily report adds a short list of dumped coins missing exactly one condition (named), marked NOT buys; (7) watchdog adds data-freshness (newest hour > 2h old) and disk < 3 GB alerts; (8) daily 09:05 BKK self-audit cron, Telegram only on a failed check; (9) sellerwatch skips a STRONG BUY the lab already sent in 24h. Tried and REVERTED: dRPC eth_getLogs for rhledger's forward pass, its free plan refuses ranges of even 500 blocks (only about 100 works), so no gain | receipt: e2e dry run 3 Oct 11:56 BKK (10 alerts stored, SELL closed once and only once, short list printed), audit all PASS except known rhledger gate rebuilds, price sources differ >1.5x on 22 of 3,188 shared hours | TESTING
- [stated] 2026-10-03 | cowork: PRICE PREDICT first analysis (Ben: "run an analysis ... see if there are strong patterns, then 50 more ideas, execute the best") | RH launches, 51,322 coins recorded 20 Sep to 3 Oct (gap 25 Sep to 2 Oct, disk guard), 69 reached $500k (second-highest print or GMGN ATH), 17 reached $1M. Strongest early sign: bonding curve full by minute 5 (27 of 39 eligible winners; 7.5% of full-curve coins hit $500k vs 0.01% of the rest). Peak-cap stats overstate edge: from a minute-5 entry, best rules double before halving only 25 to 32% (break-even is 34%) | receipt: Mac ~/ppan an2-an7, time split at 22 Sep 13:00 UTC | PROVEN
- [stated] 2026-10-03 | cowork: PRICE PREDICT rule-1, FROZEN | If at age <= 60 s the launch-board reading shows curve progress >= 0.99 AND holders per $10k of cap < 20 (big buyers), then paper-buy at that cap, sell 10x, stop -50%: n=100, EV +32% per trade after 3% costs, 90% range -7 to +74, P(EV>0) 91%, positive in both time halves, about 1 in 8 wins, carried by few big runners (AGE +870%). Live paper test: pp-paper.timer every 2 min, /opt/price-predict-rh/paper1.py md5 8711256002f1f5b640bd06609f5ac4ad, paper.db, daily scorecard 09:05 BKK, guard alerts if stale; replay reproduced all 92 backtest coins. Real money only after 30+ closed paper trades stay positive | receipt: an5/an7 bootstrap 5,000 | TESTING
- [stated] 2026-10-03 | cowork: PRICE PREDICT dip buy | If a coin filled its curve by minute 5 and you buy its first -50% (or -65%) drop within 24h, then expect to lose: EV -10 to -16% per trade at 2x/3x/5x targets with -50% stop, P(EV>0) 0 to 7% | receipt: n=244 to 266 | KILLED
- [stated] 2026-10-03 | cowork: PRICE PREDICT side signals | Copycat name of an earlier $500k winner: 1.36% hit $500k vs 0.13% base (n=516, 7 hits, mostly SI and DEED); creator with an earlier $100k coin: 1.93% hit $100k vs 0.96% first launches, serial creators with no hit 0.21%; alive coins (50+ holders at minute 5) launched 20:00 to 04:00 or 12:00 to 16:00 BKK hit $500k about 3.5 to 4% vs 1.3% other hours; among alive coins winners have fewer holders per $ (big tickets), higher avg swap ($141 vs $85), more net buy, more smart-degen tags (8.5 vs 1) | receipt: an3/an6 | TESTING
- [stated] 2026-10-03 | cowork: project named BOTTOM LAB; Ben: "only interested in significant reversals ... high-timeframe bottoms on coins that then go and make new all-time highs" | Lab v3 live (lab.py md5 5b6d19952abe4bb7de3c84cd11c87175, backup lab.py.bak-v2-*): every exit aims at a NEW ATH (or 3x), 30 days, wide stops 40/50/60/80%; entries only 50%+ under ATH (ATH from intraday highs since launch, history must start within 3 days of launch), ATH 1+ day old, one signal per coin per week; new rule settings: depth under ATH, days without a new low, distance off the low, prior 5x rocket, first-pump cap, wallet layer optional; learns on RH + Solana, alerts RH only; score = coin-balanced log growth AND goal rate, must beat random dips by 5 points in both, holdout too; NOTHING is sent until a rule is proven | receipt: e2e + audit PASS 3 Oct 12:40 BKK; first search 446 rules, 91 coins with a full 30-day window, start rule 13% new ATH vs 20% random dips, none passed | TESTING
- [stated] 2026-10-03 | cowork: LOOK-AHEAD TRAP in tracker universes | If you backtest coins from a tracker that lists coins because they LATER got big (runtrack = reached $500k), then bars before that moment are survivor-biased: base rate of "new ATH within 30 days after a 50%+ dip" read 44% using all bars vs 14% using only bars after the coin first crossed $500k (matches the 13.9% of the 5 Sep pullback study). Rule: only test hours after the coin was knowable (crossed_ts); BOTTOM LAB v3 enforces it (audit "peeking hours 0") | receipt: htf.py on runtrack kline + lab px, 123 vs 45 coins | PROVEN
- [stated] 2026-10-03 | cowork: high timeframe bottom features, first look | On the unbiased set (45 coins, 30-day forward), nothing clearly lifts the new-ATH rate over the 14% base: depth 50 to 98% flat (10 to 17%), days since ATH flat, volume trend flat; weak hints only: prior 5x rocket 17% vs 9%, base 3 to 7 days without new low 17%, first pump 10 to 30x 34% (n=8 coins). Too few coins; more history arrives daily | receipt: htf_an.py 3 Oct | TESTING
- [stated] 2026-10-03 | cowork: sellerwatch STRONG BUY now also needs the coin 50%+ under its all time high (runtrack ath_mc; unknown ATH = no alert), message shows the % under ATH; stays the live alert until BOTTOM LAB proves a new-ATH rule (then /opt/bottomlab/LIVE hands over) | receipt: patch 3 Oct, backup sellerwatch.py.bak-ath-* | TESTING
- [stated] 2026-10-03 | cowork: PRICE PREDICT rule-2 (round 2 of 50 ideas) | If rule-1 fires but the coin shows a CROWD at minute 1 (GMGN bot/degen rate > 10%, OR 3+ smart-degen tags, OR holders growing > 30 per minute), then skip it: crowd coins lose (-24% per trade, negative in all three time thirds, n=60); rule-1 without the crowd = rule-2: +98% per trade, n=49, positive in all three thirds (+9/+220/+59), 5th pct +24%, random drop of 60 coins matches it 0.4% of 2,000 tries. Live paper test tags it (paper1.py md5 7b9671878d346d0805a82f6526c5f084, trades.crowd), replay matched backtest 109/109 and 49/49 | receipt: ~/ppan an8/an10 | TESTING
- [stated] 2026-10-03 | cowork: PRICE PREDICT speed | The minute-1 edge disappears with slow entry: +31% at the reading, +9% if entered 15 s later, +2% at 30 s; same rule judged at 30 s age +53%, at 90 s -7%, at 180 s -7%. Any real use needs an auto-buy within ~10 s of the curve filling; GMGN board readings come every ~39 s per coin, so a faster feed (chain logs) is the bottleneck, not the model | receipt: an8 ideas 03/04 | PROVEN
- [stated] 2026-10-03 | cowork: PRICE PREDICT holder/volume dynamics (Ben: track holders, buy and sell volume per coin on 5 s to 1 min) | Already collected every 10 s for ~6,000 of 51,000 coins (GMGN info, first 3 min); fast holder growth does NOT predict further upside (AUC for 5x 0.44, holders +50/min & buy share >60%: 0 of 69 hit $500k); buy share by count no signal (0.48); size of 1-minute buy volume does (AUC 0.75 for 5x) = the big-buyer effect rule-1 already uses | receipt: an9 | PROVEN
- [stated] 2026-10-03 | cowork: PRICE PREDICT exits | Trailing stops lose (-4 to -11%), 20x target loses (-3%), half at 3x then 10x weaker (+16%); 10x with -40% stop slightly better than -50% (+36% vs +31%, both halves); time stop 6h or 24h same as holding (+29%); costs up to 15% round trip still positive (+15%) | receipt: an8 ideas 05/08 | TESTING
- [stated] 2026-10-03 | chat: fomo top 50 one-hit test (Ben: "Visi is a one-hit wonder, useless; show coins traded and how many coins made the profit") | If one coin made 80%+ of a trader's 30d profit, then treat the leaderboard return as one lucky coin, not skill, and do not copy. Reading per trader: coins traded (repeat buys of the same coin count once), winners (coin profit >= $1k and >= +50% on cost), coins needed for 80% of profit, top coin share | receipt: 30 days to 1 Oct, Robinhood Chain fomo router legs (18,862 receipts, ETH size from WETH legs, CoinGecko daily ETH/SOL) plus Solana parsed so far; one-hit (top coin >= 80%): Visi235 AI 96%, ExactTallTakin SI 97%, maxiplaysgames ORBIO 99%, vladsbutler PONS 99%, LowRivalRat PONS 98%, Milliardi PONS 98%, roovera AI 97%, TheOldNite PONS 96%, tremendoustoad STONK 99%, insentos CASHCAT 97%, khomenei EMBER 90%, igivethesis MEME 100%, WuKong365 fone 94%, brrrgrrrz AI 82%, dtrainz ZCAT 80%, The__Solstice Index 80% | PROVEN
- [stated] 2026-10-03 | chat: fomo top 50 spread winners | If you want traders whose profit is NOT one coin, then the candidates are (coins traded / winners / coins for 80% / top coin share): notanicecat69 62/15/3/AI 56%, bluntz_capital 55/16/5/DELTA 49%, kyle 129/17/5/AI 36%, Carver___________ 193/25/5/STONK 57%, bigbabba 198/30/10/22%, Tekkerrss 180/23/12/29%, AvgJoesCrypto 149/11/3/PONS 48%, CardinalSaint2 71/7/3/AI 51%, _bnbull_ 107/15/7/34%, 0xsneakky 609/113/79/5% (grinder) | receipt: same run; Solana side incomplete for dreamloaderxo, macdegods, Carver, 0xsneakky, Iknowwhyy, brrrgrrrz (public RPC throttled) | TESTING
- [stated] 2026-10-03 | chat: fomo board unexplained profit | If chain data for a trader's fomo wallets cannot explain their board profit, then do not judge them yet (other wallets, Base, or unparsed Solana): AnselFang, ganymede_0x, Iknowwhyy, elliotrades, 000000, MetaHunter168, solo666, RZDxbt, vickycrypto, qwerty888 | receipt: found coin profit under 20% of board 30d profit | TESTING
- [stated] 2026-10-03 | chat: method traps, fomo wallets on Robinhood Chain | Sells go to the fomo router 0xb92fe925dc43a0ecde6c8b1a2709c170ec4fff4f; USDG sent to 0x4cd00e387622c35bddb9b4c962c136462338bc31 is the gas sponsor fee (ERC-4337 entry 0x4337084d9e255ff0702461cf8895ce9e3b5ff108), not a trade; WETH 0x0bd7d308f8e1639fab988df18a8011f41eacad73 must be dropped from the coin list or it shows up as a fake top winner; coins bought before the window and sold inside it count as profit with unknown cost (flagged OLD) | receipt: this run | PROVEN
- [stated] 2026-10-03 | chat: raw data | Container of this chat (temporary): /tmp/f50/positions.json (per trader per coin flows), /tmp/f50/rep.txt (the table), /tmp/f50/report2.py, /tmp/f50/conc.py, /tmp/f50/rhrc.jsonl (receipts), /tmp/f50/solfull.jsonl (Solana, partial)
- [stated] 2026-10-03 | cowork: CORRECTION to "PRICE PREDICT speed" (marked PROVEN above, now TESTING) | The +31% to +9%/+2% delay numbers were partly noise: median entry cap does not move between 60 s and 90 s (+0%), 78 to 80% of rule-2 trades end the same; the change comes from a few winners that jump within seconds (FOX, PURSER, ZZZ: 10x win at 60 s became -42% at 90 s). Rule-2 with 10x/-40%: +100% per trade at 60 s vs about +35% at 90 to 120 s. Late entry still makes money, just much less | receipt: an12 coin-by-coin delay | TESTING
- [stated] 2026-10-03 | cowork: PRICE PREDICT weighted scores | Learned points on all 34k launches (score-1) overfit: early +35%, late -30%. Points among real candidates only (curve >= 90% at 1 min, n=197, 18 winners): logistic score-2 late +15 to +27%, evidence-points score-3 late +40% (top 30%) BUT reversed split -3%; rule-2 holds both ways (+79% early, +119% late). Too few winners to learn weights reliably yet; strongest learned signs agree with rule-2: holders +30/min strongly bad, 1 to 2 smart wallets good but 3+ bad, bots > 10% bad, night 20:00 to 04:00 BKK good, fresh wallets > 20% good | receipt: an11 to an13 | TESTING
- [stated] 2026-10-03 | cowork: rhfast chain feed live | /opt/price-predict-rh/rhfast.py md5 ecc08d43a2a358a4a57692a1b29c5989 (rhfast.service): all RH Transfer logs every ~2 s via dRPC (0.12 s per call, ~500 logs per 5 s), new launch = first transfer is a mint from 0x0, trigger at 10 to 75 s when (chain curve >= 50% and 10+ holders) or 30+ holders, then one GMGN info call (price-predict signals.normalize, fields match the board: holders 0.98, progress 0.94, smart wallets agree on >=3 in 99%) and paper decision for rule-1, rule-2, score-3 (fast.db: trades, evals, path). Replay of today's 6 full-curve launches: chain trigger at 10 to 25 s after launch vs board reading at 13 to 44 s. GMGN birth time equals first chain transfer (median 0 s, 87 of 109 rule-1 coins within 60 s). Guard alerts if stale | receipt: fastcheck3, birthcheck | TESTING
- [stated] 2026-10-03 | cowork: droplet disk | 100 GB DigitalOcean volume ($10/month, Ben approved) mounted at /mnt/volume_sgp1_1791007409047 (fstab UUID, nofail); /opt/price-predict-rh moved there (symlink, byte-identical check), recorder cap raised to 85 GiB; root disk 86% to 28% used | receipt: movepp.log | PROVEN
- [stated] 2026-10-03 | cowork: COMEBACK rule, the high timeframe bottom system (Ben: "find the system today", only bottoms that lead to new highs, RH + Solana) | If a coin already made 1 or 2 comebacks (fell 50%+ under its ATH, then made a new ATH), is now 50 to 80% under its ATH (ATH 1+ day old, after it first reached $500k), trades under its 7-day EMA, and its ATH cap is under $3M, then STRONG BUY, goal new ATH, stop 50% under entry, max 30 days. 0 comebacks = random, 3+ = exhausted (worse than random) | receipt: 984 coins (RH + SOL, 4h candles from launch, runtrack + GeckoTerminal), 396 with full 30-day outcomes; time split: train 37% vs 20% new ATH, test 41% vs 20% (n=107, 63 coins); same-week vs other dips +19.7 pts new ATH [+15.5,+26.0], +17.6 pts growth; holds both halves, both chains (Solana weaker on money); under $10M: test 37% vs 20%, same-week +17.4 pts; median 6.7 days to the new ATH when it comes, median pain about -50% first. Live in BOTTOM LAB v4 (lab.py md5 4bcfef5eaea24d2e0636016cc6567f59), alerts RH + Solana, first list of 12 sent 3 Oct 13:45 BKK; sellerwatch silenced (/opt/bottomlab/LIVE) | TESTING (live scoring)
- [stated] 2026-10-03 | cowork: Ben's bottom confluences tested (3,103 first touches after 3x+ runs, 234 coins RH + SOL, control = same coin and episode, level shifted 5 to 15%) | fib 0.5 / 0.618 / 0.66 / 0.7 and 50 to 70% depth: no better than the shifted level (within 0 to 4 pts, CIs overlap), deeper = worse; untested 12h OB at the level: no edge (52% vs 60% back to -38.2%), 1D OB worse (40%, new ATH 12%); 4h RSI under 30 at the level much worse (23% win, new ATH 7%), 1D RSI under 35 worse (13%, 0%); MA100/200 near the level no edge; 2+ confluences stacked worse (33% vs 61%); 4h bullish divergence small hint (new ATH 35% vs 28%, n=62, 27 coins). Price chart patterns do not pick bottoms; coin history (comebacks) does | receipt: study2.py / rep2.py 3 Oct | KILLED (divergence: TESTING)
- [stated] 2026-10-03 | cowork: Ben's chain rotation (big coins run first, then mid, then small; big coins topping = wave over) | RH: big tier led mid by about 3 days on the way up (corr +0.54 vs +0.30 reverse), mid did not lead small; at the top small coins rolled over first (week of 6 Sep: small -9%, mid +15%, big +72%), big last; leaders near ATH -> chain median coin next 14d -15%, leaders 30 to 65% under ATH -> +11 to +12%, 65 to 80% -> -41% (4 days). Solana: tiers move together, no order. Only one RH wave in the data, so a hint, not proven | receipt: study3.py, 62 RH days, 125 SOL days | TESTING
- [stated] 2026-10-03 | cowork: price features cannot pick which dip makes a new ATH | Out-of-time gradient boosting on 109k dip hours: new ATH AUC 0.63 only from distance to ATH and ATH size; fixed 2x/3x targets AUC 0.51 to 0.55 (coin flip). Same-week contrasts: breakouts (close above 7d high) worse for new ATH (-9.6 pts); buying weakness (RSI under 40, 7d down, near the low) better on money but lower new-ATH odds; "chain cold" breadth effect was one late-August episode, not repeatable | receipt: ml.py, xsec.py, harness.py 3 Oct | PROVEN
- [stated] 2026-10-03 | cowork: droplet trap | Another session moved /opt/fillwatch to /opt/RETIRED-20261003 at 06:31 UTC, but rhledger.py and BOTTOM LAB import fillwatch: the next restart would have crashed both. Fixed by copying only the module to /opt/runtrack/fillwatch.py (import verified). Rule: before retiring a service, grep /opt for imports of it | receipt: find + import test 3 Oct 13:35 BKK | PROVEN
- 2026-10-03 | cowork: basket buy window | If a basket breakout (close > 30d high, market gates on) is bought k days late, then 2025 to 2026 PF falls 1.39 (day 0) / 1.02 (1) / 0.96 (2) / 0.93 (3) / 0.55 (5) / 0.61 (13); BUY phase = signal close + next close only, then HOLD (no adding) | 14 CMC baskets, dead coins kept, 2023 to 2026 | PROVEN
- 2026-10-03 | cowork: basket rule v4 filters | If Solana baskets break out while SOL is +20% on BTC in 4 weeks, or fewer than 25% of other baskets are risk on, then skip; both halves improve: 2023-24 PF 3.76 to 4.29, 2025-26 1.39 to 1.83 (weekly clustered 0.97 to 1.49); 109 trades +9.4% avg, 61% wins | thresholds 15 to 35% all agree | TESTING
- 2026-10-03 | cowork: basket rotation by phase | If new money goes only into baskets on their BUY day (3 slots) and is held until a 15d low close, then Jan 2023 to Sep 2026 +364% with worst drop -24% and ~21% money in, vs BTC +403% / -53%; holding the 3 strongest risk on baskets daily +389% / -37% | rh50/it_c.py | TESTING
- 2026-10-03 | cowork: fast rule for young baskets (Robinhood, Ink, Arc) | If a basket has under ~30 days of history, then use 10d high in / 5d low out: on the 14 long baskets PF 2.69 (2023-24 3.14, 2025-26 2.10); on rebuilt Robinhood 30/mid caps (runtrack prices, point in time) it caught late Aug and exited 9 to 10 Sep (+37.5%, +67.5%) while 30/15 bought the 1 Sep top (-34 to -36%) | n=4 RH trades | TESTING
- 2026-10-03 | cowork: basket index construction | Daily equal weight reset vs weekly drift (official Robinhood Indices style): same picture (2025-26 PF 1.87 vs 2.95, 6 baskets); top 30 members beat top 15 recently (2.89), top 10 worst (0.92) | rh50/it_f.py | TESTING
- 2026-10-03 | cowork: rule v4 raw data | http://159.223.88.168/hx-c9ois/edge/basket-rule-v4-2026-10-03/ (all.tgz: 70 iterations, evidence5.json, rebuilt RH lines, runtrack kline); handover dash-h5 http://159.223.88.168/hx-c9ois/rotation/dashboard-handover.html
- [stated] 2026-10-03 | cowork: BUY FRAMEWORK v1 ([[buy-framework]]) | If a coin is 70%+ under its ATH, the ATH is 3+ days old and X has gone QUIET (contract posted about once a day or less), then buy (depression stage); with price near the 72h low or RSI 1h < 45 it is a strong buy | receipt: one trade per coin, 2x/-50%/14d, 3% costs; out of sample 36 coins EV +25%, halved 19% vs random -2.5%, halved 47%; with trigger + no chain knife 23 coins EV +30%, halved 13%; 63-coin blind set agrees (EV +23% vs +4%) | TESTING (small n, forward paper test from 3 Oct)
- [stated] 2026-10-03 | cowork: ANGER STAGE ([[buy-framework]]) | If a coin is 70%+ down but X is still busy or angry, then it is NOT the bottom: wait | receipt: out of sample 87 coins EV -6%, halved 60% (worse than random); deep dip alone EV -4%, halved 52% | TESTING
- [stated] 2026-10-03 | cowork: EUPHORIA ([[buy-framework]]) | If a coin is within 30% of its ATH and its X posting rate doubled in 3 days with 30%+ bullish words, then no buy / trim | receipt: n=25, next 7d mean log return -0.47 vs -0.12 other near-ATH moments | TESTING
- [stated] 2026-10-03 | cowork: sentiment word lists | "dead / over / cto" despair words and LunarCrush sentiment % add nothing; X post RATE does | receipt: 2,011 snapshots, despair-word deep dips -0.09 hit vs same-day average | KILLED (word lists)
- [stated] 2026-10-03 | cowork: GOLDEN POCKET as a zone, retested the way Ben trades it (zone 0.5 to 0.666 of the run low -> ATH, run 3x+, entry at the 0.618 touch, out on a 1h or 4h CLOSE under 0.666, hard stop wick 0.75, goal the ATH, 30 days, 2% costs) | Zone touches made a new ATH 13 to 19% of the time, same as 0.5, 0.75 or 0.85 entries (flat curve, no bump at 0.618); back to the 0.382 level within 30 days 72 to 79% (smooth with depth). The 0.618 -> 0.75 drop is about 32% of PRICE on a typical 10x+ run (not 10%); with the 4h close rule the average exit loss is ~13 to 15%. 1h vs 4h close rule: no difference; wick-only stop slightly better | receipt: gp.py / gprep.py, 4h data 2,356 zone touches 280 coins, hourly data 829 touches 104 coins | PROVEN (zone not special), OK as a waiting area
- [stated] 2026-10-03 | cowork: what raises the odds INSIDE the golden pocket (4h data 280 coins; hourly 104) | 4h bullish RSI divergence at the touch: new ATH 47 to 51% vs 24% in the same coins without it, avg +52 to +56% vs +2 to +18%, both halves, both chains (36 touches, 17 coins: TESTING); untested 12h order block in the zone 24% vs 17%, avg +20% vs +3% (86 touches, 40 coins); 2+ OB timeframes 26%, +25%; 1h OB in zone 28% vs 22% (hourly); MA100/200 inside the zone avg +10 to +21% vs +3 to +5% (same ATH rate); slow drop to the zone (3+ days) +15% vs +2.5%; run 10 to 50x 30% ATH, 50x+ 16%; ATH under $3M 24% vs 16%; RH better than Solana. WORSE: RSI 1h under 30 at the touch (1% new ATH), RSI 4h under 35 (12% vs 18%). Divergence anywhere else in a dip (213 cases, 76 coins): no edge vs control. CORRECTS the earlier line "OBs worse": at any level yes, inside the golden pocket they help. 15m RSI not testable (no 15m history) | receipt: gpconf.py, divx.py 3 Oct | TESTING
- [stated] 2026-10-03 | cowork: sentiment at real bottoms (twitterapi.io cashtag sample, 40 latest posts per coin-day every 2 days, 35 coins, 1,051 coin-days) | At each coin's post-ATH low: X post rate at 8% of its peak (median), down 80% vs ~10 days before (134 -> 27 posts/day), bearish words spike AT the low (10% of posts vs 4% before, 1% after), then attention returns (72/day) and price median 2.0x off the low. As a buy filter on all dip days: attention under 10% of peak 2x in 14d 63% vs 52%; falling attention + angry talk median +22% vs +9% 14d; new-ATH rate not improved. Agrees with [[buy-framework]] (quiet X = depression = buy); the anger spike is the last day, not a wait signal, in this small set | receipt: sent.py / sentan.py 3 Oct | TESTING
- [stated] 2026-10-03 | cowork: LunarCrush on RH coins | Covers PONS only from 4 Sep (limited data mode), sentiment stuck 72 to 91 (useless); mentions and engagement work: PONS posts 8,347/day (6 Sep) -> ~2,000 (3 Oct), engagement 10.5M -> 0.55M | receipt: topic_time_series $pons 3m | PROVEN
- [stated] 2026-10-03 | cowork: GOLDEN POCKET WATCH live | /opt/bottomlab/gpwatch.py, cron :25 hourly, RH + Solana: STRONG BUY when price reached 0.618, no 4h close under 0.666 and the last 4h candle completes a bullish divergence; message carries zone, 4h-close exit at 0.666, hard stop 0.75, goal ATH; in-zone list in /opt/bottomlab/gpzone.json; watchdog in stale.py | receipt: dry run 3 Oct 14:26 BKK, 8 coins in zone (PONS, BONER, UBIK, PRIORS, Bucket, SLIPPY, STBL, AMETHYST), 0 fires | TESTING
- [stated] 2026-10-03 | cowork: BOTTOM LAB universe bug | runtrack never set crossed_ts for 317 of 641 RH tokens, PONS and AI among them, so BOTTOM LAB skipped half the chain; lab.py now computes the first hour over $500k itself (md5 61d17ecbfe4978584d498b4f6ae4e958, 896 coins loaded vs 659); runtrack supply for PONS is 1.46x DexScreener's, so gpwatch prices caps from DexScreener at alert time | receipt: tokens table check 3 Oct | PROVEN
- [stated] 2026-10-03 | cowork: PONS and AI read 14:20 BKK (PONS 0x39dBED3a2bd333467115dE45665cC57F813C4571, AI 0x2E8c31162b855A2ffa90F6F8634643Ad6F111e18) | PONS $289M, run $10M -> ATH $666M (5 Sep): 0.5 $338M, 0.618 $261M, 0.666 $230M, 0.75 $174M; untested 4h OB $261 to 265M sits on the 0.618; RSI 4h 27 (oversold, the worse case), no 4h divergence yet; X still loud (buy-framework: WAIT anger, 19.7 posts/h vs 8.6 three days ago); LunarCrush attention down 75 to 90% from peak. AI $128M on the 0.666 line ($129M), 4h bullish divergence present, but a 4h close under $129M voids it; X 13 posts/h vs 47 three days ago (calming) | receipt: live_gp.py, buyscore.py | TESTING
- [stated] 2026-10-03 | cowork: BUY FRAMEWORK v2, relative quiet ([[buy-framework]]) | If X chatter about a dipped coin fell to 20% or less of its rate at the ATH, then that alone is NOT a buy: it marks dying coins (one trade per coin, 2x/-50%/14d: 63 coins EV -17%, halved 68%); ABSOLUTE quiet (contract posted about once a day or less) beats busy in every market cap band (<$1M, $1M to $3M, $3M to $10M, $10M+) and quiet per $1M of cap (bottom third) also works out of sample (+21%, 26 coins). CONFLICTS with the line above "attention under 10% of peak 2x in 14d 63% vs 52%" (different measure: cashtag posts/day, no stop); both stay TESTING until a forward test settles it | TESTING
- [stated] 2026-10-03 | cowork: GOLDEN POCKET confluence, hourly event study ([[buy-framework]]) | 965 coins RH + SOL (runtrack + 422 GMGN-only RH coins + blind GT set), 3x+ legs, first 0.618 touch, goal ATH, 30 days, 3% costs: plain touch EV -2% to -5% under every stop style (1h close <0.666, 4h close <0.666, wick 0.75, 1h close <0.75 best); deeper first touches pay more (0.75 +15%, 0.8 +24%); 1h order block inside the zone +18% vs +1% (n=269), buy volume shifting to green candles (6h vs 24h) +19% (n=131, all 4 chain x time cells positive), X quiet +12% vs -3%; OB + X quiet +28% (n=67); RSI 1h <25 at the touch -34%; 1h and 15m divergence entries no edge (n=242 / 42); 4h divergence entries raise the ATH-retest rate (48% vs 32%) but lose money entered at the divergence close (-9%, n=105, 84 coins) because the entry sits far above the pocket | TESTING
- [stated] 2026-10-03 | cowork: X API cost | A twitterapi.io search returning 20 posts costs about 180 credits; ~15,000 searches in one research session emptied the shared balance (2.65M -> -44,790 credits, about $27) at 14:35 BKK 3 Oct and stopped every X-fed service on the same key (x-alert, xdeck, followwatch, casig, trench). Rule: check https://api.twitterapi.io/oapi/my/info before any batch and stop at a stated credit cap; a failed X call must never be read as "quiet" (buyscore.py v2 refuses a verdict instead) | PROVEN
- [stated] 2026-10-03 | cowork: PRICE PREDICT themes (Ben: "anime thing went to $20M, runners share the theme") | If a rule-1 coin's TICKER equals the ticker of a coin (RH or Solana) that reached $1M before this launch, then it does far better: +161% per trade vs +1% (n=24 vs 85, early +141% / late +176%); outside rule-1 theme copies LOSE (-30 to -38%, n=15 to 35): theme needs the big-buyer structure; shared name WORD only flips in the late half (not used); meta waves (5+ same-word launches in 60 min) no help | receipt: ~/ppan an14-an15, names from forward.db board rows, $1M timing from runtrack kline + forward prints | TESTING
- [stated] 2026-10-03 | cowork: ZZZ correction | The ZZZ in the rule-2 delay example is a COPYCAT, 0x880d596c70ac2d2868bc75df36b65faa3d9ea4a6 (22 Sep, $47,901 at 60 s, $333,729 at 90 s, peak $605,725, now ~$7k); the real ZZZ is 0x7dbf38976f6D3b9c529e7D9484A71898B409eE6a (5 Sep, ATH $82.2M, ~$13M on 3 Oct), not in the 20 Sep+ data. "Loss at 90 s" = it had already done 7x in 30 s, then only 1.8x more | receipt: zzz.py, DexScreener 3 Oct | PROVEN
- [stated] 2026-10-03 | cowork: PRICE PREDICT score-4 (equal points, Ben: use ALL good and bad signs, strong good can outweigh a mild bad) | Inside rule-1: +1 each for theme ticker, fresh wallets >20%, 1-2 smart wallets, launch 20:00-04:00 BKK, creator had a $100k coin, liquidity/cap >0.35, snipers >5%; -1 creator 10+ launches (11/11 lost), dev sold, telegram link (9/9 lost); -2 crowd. All 11 signs picked on the early half held direction on the late half. Rule-1 & score-4 >= 2: n=34, +182%/trade (early +186%, late +179%), 9 of 13 winners kept, permutation null 0.2%; >= 3: n=18, +313%. Live in rhfast and paper1 (score4.py md5 c4d0d319c62e837a098cf1083a37a0d3, columns score4/score4_buy/signs, daily scorecard line) | receipt: an16 | TESTING
- [stated] 2026-10-03 | cowork: SUPERVISOR (Ben: every process must have a tracker that fixes it within an hour) | /opt/supervisor/super.py, super.timer every 5 min: auto-discovers every systemd unit under /opt (40), restarts dead ones (max 3/h), checks output freshness (paper.ok, fast.ok, collector status, rhledger.ok, runtrack, watchdog, cleaner), young-coin feed and rhfast evals, disk; Telegram (watchdog bot) DOWN after a failed fix, STILL DOWN after 45 min, BACK; runtrack-guard watches super.ok. Paper test had been running all along: 0 coins qualified (0 full-curve launches 11:20 to 14:20 BKK); its log now prints coins judged per run. Fault test: stopped moowatch, restarted in the next pass | receipt: deploy 3 Oct 14:28 BKK | PROVEN
- [stated] 2026-10-03 | cowork: where to anchor the fib (Ben asked) | Tested 5 placements on the same golden-pocket trade (4h data, 280 coins): run low = lowest low in the 45 days before the ATH (start of the move) to the ATH wick, 14-day low, leg low since the last 50% dip, launch low, and plain % from the top. 45-day and launch anchors best and equal (0.618 touch: new ATH 14%, back to 0.382 74%); plain % from the top worst (new ATH 8 to 11%). Rule: draw from the lowest wick before the run to the top wick | receipt: anchor.py 3 Oct | PROVEN
- [stated] 2026-10-03 | cowork: GOLDEN POCKET SCORE v1 (Ben: odds tiers from tested confluences) | Points at the zone touch: +2 4h bullish divergence, +1 1h bullish divergence, +2 order blocks on 2+ timeframes in the zone (else +1 each for a 12h/1D, 4h, 1h one), +1 MA100/200 (1h or 4h) inside the zone, +1 ATH $1M to $50M, +1 run 10 to 50x, -2 oversold (RSI 4h < 30, daily < 35 or 1h < 30), -1 RSI 4h 30 to 40, -1 run under 10x. Tiers on 1,597 touches (291 coins, RH + Solana): 4+ VERY HIGH bounce to 0.382 71% / new ATH 27% / avg +18%; 2 to 3 HIGH 62% / 30% / +23%; 0 to 1 MEDIUM 44% / 20% / +6%; below 0 LOW 28% / 15% / -2%; same in both time halves and both chains. Big-coin penalty dropped: ATH over $50M = average (17% new ATH, n=49, 18 coins). A logistic model on the same flags only reaches out-of-sample AUC 0.58 to 0.60, so treat the tiers as odds, not certainty | receipt: evb4.py, hfeat.py, points.py, score.py 3 Oct | TESTING
- [stated] 2026-10-03 | cowork: confluence detail inside the golden pocket (first touch per ATH, 505 events) | 2+ order block timeframes 57% new ATH (n=16), 12h OB 44% (n=20), 1D OB 35% (n=11), 4h OB 20% (n=62) vs base 14%; 4h divergence 36% (n=14); MA100 4h in zone 23% (n=78); hourly coins (101): 1h OB bounce 71% vs 46%, 1h divergence new ATH 33% / bounce 76% (n=22), 1h MA in zone 28% / 62%, 1h RSI < 30 12%. Drop speed: 2 to 7 days best (22%), under 2 days 15%, 7+ days 12%. Combos: slow drop + 4h divergence 38% (n=31), slow + OB + not oversold 28% / bounce 59%, (OB or 4h div) + not oversold 28% / 66% (n=182, 82 coins) vs 19% / 44% all. Confirmation entry (still above 0.666 three days after the touch): 34% new ATH but only 192 of 1,597 touches survive | receipt: fac.py, hfac.py, combos 3 Oct | TESTING
- [stated] 2026-10-03 | cowork: sentiment at the golden pocket touch (21 coins with daily X samples, 116 touches) | X attention under 25% of the ATH day: new ATH 56% vs 23%; optimistic talk (35%+ bullish posts) 18% vs 48% when bullish talk is low (Ben's counter-signal confirmed); angry talk (10%+) 17%; quiet and not optimistic 65% (n=12); loud and optimistic 18%. Small n: full test on all 1,597 touches is blocked because twitterapi.io credits ran out (402 "Credits is not enough") on 3 Oct ~14:40 BKK | receipt: sent.pkl x ev4 | TESTING
- [stated] 2026-10-03 | cowork: GOLDEN POCKET WATCH v2 live (/opt/bottomlab/gpwatch2.py, cron :25, RH + Solana) | Sends STRONG BUY ZONE (score 4+) or BUY ZONE (2 to 3) when price is in the 0.5 to 0.666 zone with no 4h close under 0.666, with the why-list, zone, exit and odds; X layer (+2 quiet, -2 loud optimistic, -1 angry) switches on by itself once twitterapi.io credits are topped up (key file /opt/common/twitterapi_key, chmod 600). First run 3 Oct 14:54 BKK sent 6: STBL 0x79a74fd91f8e1c4ab8e76253dec5c91f3094393f and UBIK 0x812486eaea648819853f8e372dc9f1516c7868bd VERY HIGH; AI 0x2e8c31162b855a2ffa90f6f8634643ad6f111e18, BONER 0x98096d17e191b3da1d5f99a6d7b3584351b11e18, PRIORS 0xedbf91223639800bcd5756815caf908df3b890be, SLIPPY 0xf0568863195770965a6d8abb0aa87f4314b80320 HIGH; PONS 0x39dbed3a2bd333467115de45665cc57f813c4571 LOW (-1: RSI 4h 27 / 1h 17 oversold) | TESTING
- [stated] 2026-10-03 | cowork: BUY FRAMEWORK v2.1, spam + mood ([[buy-framework]]) | X contract searches are 39% spam overall and 45% to 100% on live dumped coins; if real people go quiet but bots/scam links keep the coin loud, then NO BUY (-18% EV, halved 70% in full sample); keep the quiet test on ALL posts (deep + all quiet +25%); read humans' mood by meaning, not keywords: hopeful after a dump -12%, euphoric -32%, 0 to 1 humans +13% | receipt: gq.py, labels_*.json, live run 15:05 BKK | TESTING
- 2026-10-03 | cowork: DATE TRAP, CoinMarketCap vs OKX | CoinMarketCap historical label D = the close at the END of day D = the OKX 1Dutc candle that OPENS on D; keying OKX by open+1 puts every gate a day late (daily return correlation 0.01; fixed 0.94 to 0.9997). Check any new price source against BTC on both clocks before mixing | receipt: 298 days BTC match 0.05% only at that alignment | PROVEN
- 2026-10-03 | cowork: basket EXTENDED definition | Of 10 "extended" rules (HOT = 20d z>=2 in top of 90d range, % over 20d avg, 30d gain, RSI 80/85, gain from entry, day 8, sell all), HOT with half taken was best in 2025-26 (PF 1.83 vs 0.88 with no profit taking) | 14 CMC baskets 2023-26 | TESTING
- 2026-10-03 | cowork: rotate out of extended into fresh BUY | If a tested basket turns BUY while slots are full, move the half left in the weakest EXTENDED basket into it: Jan 2023 to Sep 2026 +568% (worst -31%) vs +364% without, BTC +403% (-53%); strong buys (bigger 7d/30d move, rising volume, higher RSI) did better in both halves | it100/it_i.py | TESTING
- 2026-10-03 | cowork: stablecoin inflows per chain | If a chain's stablecoins rise (7d or 30d) or its DEX volume rises, then basket prices are NOT reliably better next 14 days (Solana, Base, TON 2024-26, mixed signs); RH: ~$1.06B of $1.08B stables sit in Morpho | flowtest.py | KILLED as a signal
- 2026-10-03 | cowork: risk off states | Still making new 20d lows: turned bullish within 14d 12 to 18%; no new low 5d+ (basing): 30 to 35%; within 5% of the 30d high: 49 to 73% | 14 baskets 2023-26 | PROVEN (descriptive)
- [stated] 2026-10-03 | cowork: PRICE PREDICT copycats, shared socials, deployers, callers (round 4) | Sharing an X handle/site with other coins is the RH norm and NOT bad (3.9% vs 4.0% win, 34k launches); worse: brand-new handle (rule-1 -32%, 1/30 wins; casig era $500k 0.02% vs 0.07%), link to someone's tweet (-13%), site = social link (6/6 lost); order (first vs later) does not identify the official coin, so ownership is checked on X: rhfast xcheck.py (owner_posted_other = COPYCAT, own scorecard line). Deployer 10+ launches 1.0% vs 4.0%, serial no-hit 1.2%, relaunch <30 min 1.7%. Known caller posting the CA within 10 min: 2.9% reached $500k vs 0.05% (4/140, both halves), 12 of 23 $500k coins had one within 60 min: confirmation sign, live tag. rhfast now logs first 40 buyer wallets per launch | receipt: an18/an19, round-4 project doc | TESTING
- [stated] 2026-10-03 | cowork: PRICE PREDICT round 5 (64 ideas) | Dev skin: dev holds >5% wide 2.4x/1.8x win rate, rule-1 +150% vs +12%; dev holds 0 bad (rule-1 -12% vs +97%), but it overlaps score-4's sniper/liquidity signs (adds +30 pts, inside luck, 30% null) so it runs as shadow score-5 (+dev skin, -dev spray in last 10 min, -telegram dup). Sizing 2 units at score-4 >= 3: +226%/+227% per $ vs +186%/+179% flat (live scorecard line). KILLED: minute-1 flow (they mark coins already big, rule-1 EV -14 to -42%), GMGN wallet tags, widening with dev skin (14/14 lost), early-wallet quality from chain logs (flips sign between groups, 27 winners), strategy momentum, weekday | receipt: an20-an23, round-5 project doc | TESTING
- [stated] 2026-10-03 | cowork: SILENCE BOUNCE FAILED + PULLBACK TO NEW ATH STUDY ([[buy-framework]] v3.2, doc claude/pullback-ath-study.md; CORRECTS v3.1) | Silence at a fresh 80%+ first touch: 679 new events (425 coins) -6%/trade vs talking -7%; all 1,060 events +1% vs 0%. Not a buy; buyscore v3.2 says WAIT. 100 approaches (2,266 events): runtrack coins are survivor-biased (filled 2 Oct from top-now lists); its young/small-cap winners (32% vs 21% new ATH) vanish on other lists (19% vs 21%). Only Bottom Lab COMEBACK held off-list: +18 pts new ATH 14d (CI +3..+32, 47 coins) | receipt: bounce_an.py, test100.py, rw_check.py | TESTING
- [stated] 2026-10-03 | cowork: CROWD CYCLE v4 ([[buy-framework]] v4, doc claude/crowd-cycle-framework.md; 176k posts, 8k read by tone) | Same cheat-sheet cycle in all coins for volume (86% of 127), euphoria at top (72%), denial in crash (70%), disbelief in recovery (75%); NOT anger (57%). If contract chatter is quiet but humans still talk on the $TICKER, then good (0.618 touch: new ATH 29% vs 18%); if humans are gone everywhere (memecoin "depression"), then it is death not a bottom (new ATH 3 to 4%). Blind on newest coins: LLM crowd read 44% on 5 zones (chance 20%); counts-only model fails (34%). Old "quiet at pocket 56%" did not hold. Alerts stay off | TESTING
- [stated] 2026-10-03 | cowork: CROWD AT COMEBACK SIGNAL ([[buy-framework]] v5) | If a COMEBACK STRONG BUY fires, then the crowd mood does not change the odds: depression read 29% new ATH, optimism 36%, denial 42%, euphoric 44% (213 signals, 91 coins, base 35%); do not require depression, do not veto on optimism. Alerts now carry an X crowd line and store posts for later review | TESTING
- 2026-10-03 | chat: MONEYTAUR SYSTEM ([[moneytaur-system]], 1,441 public posts) | Fresh HTF body levels (OB, breaker, FVG, hidden OB behind 2+ FVGs), 1h refinement, 4h body-close stop, target PSH: every variant lost on memecoins, -0.04 to -0.66R, below shifted levels | receipt: 275 coins 1h, 5 iterations, 6,001 plain + 2,444 refined trades; new coins -0.46R | KILLED (memecoins)
- 2026-10-03 | cowork: basket rule v4.2 (78 ideas, x50) | Hot market 20d entry when 50%+ baskets on, breadth 35%, trail 20%: book +1006% dd -25 vs +511% dd -33, every year better, 161/168 neighbours win; strongest-hold daily, dual momentum, pullbacks, time exits KILLED | edge/basket-rule-v42-2026-10-03 | LIVE