---
name: trading-edge-hub-2
description: Continuation of the trading-edge-hub insight ledger (hub hit its size cap 5 Oct 2026). Same rules: read before any buy/sell/chart call, add new insights here, newest at bottom.
sources: [cowork]
aliases: [edge hub part 2, insight ledger 2]
---
- [stated] Ben's standing rule: every new trading insight from any session goes into this ledger (rule, receipt, status). Read [[trading-edge-hub]] first.
- Live chain risk light for Robinhood Chain: http://159.223.88.168/chain.json (state, risk on/off), see [[chain-brain]].
- 2026-10-05 | cowork chain brain | RH chain basket bleeds: coins alive at any hour (every RH coin that ever hit $500k, dead held flat) fell a typical -4% per 24h from 29 Aug to 2 Oct; no chain state had a positive 24h average. Risk on is "least bad", not a buy | receipt: 818 hours, trimmed mean of 600 to 700 coins | PROVEN (this cycle)
- 2026-10-05 | cowork chain brain | Stretch rule: if the chain is stretched up vs its own last 7 days (mean z of breadth 24h, basket 24h return, distance from 72h high; top fifth), then 85% chance of a 10% basket drop before a 10% rise within 72h (base 56%); bottom fifth 35% | receipt: AUC 0.70, beat 500 time-shift nulls (p<0.002), AUC 0.68 to 1.00 in all 6 weeks, 29 Aug to 2 Oct | TESTING (one cycle)
- 2026-10-05 | cowork chain brain | Chain index mean reverts hour to day: basket far above its EMA200 (1h) predicts LOWER next 24h inside every week (rank corr -0.2 to -0.7). Basket above EMA is a coin dip-buy filter ([[system-benchmark]]), not a chain buy signal | receipt: 6 weekly folds | TESTING
- 2026-10-05 | cowork chain brain | Reheat call (washed out stretch plus breadth turning up) AUC 0.76, p 0.04, but zero +10% reheats happened 19 Sep to 5 Oct, so unproven out of sample | TESTING
- 2026-10-05 | cowork chain brain | Fitted multi-feature models (flows, BTC, young vs old coins, leaders) did worse out of sample than the plain stretch score; level shifts week to week break them | KILLED (for now, small sample)
- 2026-10-05 | cowork chain brain macro | RH big-swing rules: bridge in when OLD coins (3+ days) wake past $500k (12+ in 3 days) after a 40%+ chain drawdown; take profit when fees run 1.5x normal AND new launches crossing $500k outnumber old coins 2:1 (money moving into fresh launches = late) | receipt: bottom 18 Aug caught 21 Aug (+92% after); tops 16 Jul and 3 Sep caught 16 Jul and 2 Sep (-50% in 14 days after); n = 1 bottom, 2 tops [[chain-brain]] | TESTING
- 2026-10-05 | cowork chain brain macro v1.2 | Wakeup threshold must scale: count revivals too and require 3x the chain's own normal 3 day wakeup count (min 12); a fixed 12 now fires on noise because ~415 old coins sit under $500k vs ~20 in Aug | receipt: same 21 Aug fire, zero false fires 1 Sep to 5 Oct; 5 Oct 10 vs normal 12, need 36 | TESTING
- 2026-10-05 | cowork chain brain, Solana big swings | Solana meme cap (CMC daily listings, Solana memes >= $1M, Jan 2024 to Oct 2026): bottoms 6 Sep 24, 8 Apr 25, 22 Jun 25, 1 Sep 25, 7 Aug 26 ($3.3B); tops 21 Jan 25 ($27.7B), 13 May 25, 22 Jul 25, 13 Sep 25 ($18.9B). 5 Oct 26: $4.8B, +45% off the Aug low, -75% vs Sep 25 top, fees 1.16x normal | PROVEN (map)
- 2026-10-05 | cowork chain brain, Solana out of sample | RH bottom rule "old coins waking at 3x normal" FAILED on Solana: 16 fires, median next 30d -11% vs -2% any day, +30% reached 24% vs 34%; it fires near mini tops (13 Sep 25, 9 May 25). The RH 21 Aug hit is n=1 and may be luck. Caveat: CMC listings, not every launch | KILLED on Solana, RH TESTING only
- 2026-10-05 | cowork chain brain, Solana | Breakout after a 40%+ crash (meme cap closes above its 30d high): 6 fires, 3 big wins (Sep 24 +194% max in 60d, Apr 25 +44%, Jul 25 +38%), 3 losers (Jul 24, Jan 26, May 26); coin flip with a fat right tail | TESTING
- 2026-10-05 | cowork chain brain, Solana | Top rule (launchpad fees 7d > 1.5x 28d and new coins outnumber old 2:1) caught 6 Jan 25 (-41% next 30d), 28 Jul 25, 16 Sep 25 (-49%) but false fires in the Sep to Nov 2024 bull (+36 to +44%) | TESTING
- 2026-10-05 | cowork chain brain, Solana bottom hunt (11 signals, Mar 2024 to Oct 2026, 7 meme-cap lows) | BEST: meme cap closes back above its 20d average after a 40%+ fall AND BTC above its 50d average: n=8, next 30d median +16% vs -3% any day, both halves positive (+60%/+9%), caught 3 of 7 lows. Same rule without BTC filter: n=16, +6%, 5 of 7 lows but bear half -15% | TESTING (now the RH HEATING rule, v3)
- 2026-10-05 | cowork chain brain, Solana | KILLED as bottom signals: extreme fear < 20 (-12%), fees 35% of peak (-25%), fees turning up after cooling (-11%, fires near tops), stablecoin outflow (-15%), memes -40% in 30d (-7%), SOL/BTC washout (-3%). BitMEX funding < 0 looked great (+11%, 5 of 7 lows) but did NOT replicate on Hyperliquid funding (-10%, 0 of 7) and BitMEX XBTUSD was settled 16 Sep 2026 | KILLED
- 2026-10-05 | cowork chain brain, Solana tops | Launchpad fee heat: first day fees 7d > 1.3x 28d -> next 30d median -19%, 43% chance of another -25% vs 26% base, both halves negative | TESTING (supports the TOO HOT rule)
- 2026-10-05 | cowork chain brain, iterations | Active users (Artemis free API, Solana DAU): users 7d avg at a 180d low -> memes next 30d +20%, max 60d +56%, 0% chance of another -25%, both halves positive, n=4 (2024 to Aug 2025). DAU turning up after -30% or DAU rising while memes at lows: no edge | TESTING
- 2026-10-05 | cowork chain brain, funding replacement | BitMEX XBTUSD settled 16 Sep 2026. Deribit BTC-PERPETUAL funding (free, hourly since 2024) 7d < 0: +5% next 30d, both halves +, only 2 of 7 lows; dYdX 7d < 0: -16%; Hyperliquid: -10%. Funding is venue-fragile, info only | KILLED as a rule
- 2026-10-05 | cowork chain brain | RH waves both started with something NEW: chain launch (Jul) and Pons V2 (fee share 3% on 9 Aug -> 9% 23 Aug -> 31% 30 Aug, then the run). Watch for a venue tripling its fee share to 10%+ in a week. Fables 5 Oct: 5% share, flat 2 weeks | TESTING (n=1)
- 2026-10-05 | cowork chain brain, RH users | RH active wallets (Blockscout activeAccounts, 7d median) peaked mid Sep, about 12 days AFTER the 3 Sep price top; 5 Oct 379k vs Aug low 244k (+55%): users have not capitulated yet | TESTING
- 2026-10-05 | cowork chain brain, timing analog | Solana meme tops -> lows: 13 to 77 days (median ~41), depth -34% to -76%, the Jan 2025 blowoff -76% in 77 days; Sep 2025 top became an 11 month bear (-83%). RH 3 Sep top: -77% by day 32, fees -85% and still falling; launchpad base rate puts the activity bottom day 55 to 90 after the fee peak (about 29 Oct to 3 Dec 2026) | TESTING
- 2026-10-05 | cowork chain gauge | Cheapness beats timing: on Solana memes, SETUP (40%+ off high, launchpad fees 60%+ below peak, low volatility) 0-1 of 3 -> next 30d median -8%; 2+ -> +5%; 3 of 3 -> +20% to +59% (n=63). Turn signals (above 20d avg, BTC, users) add little alone. A fitted multi-feature model failed out of sample (AUC below 0.5) | TESTING; live as the daily gauge [[chain-brain]]
- 2026-10-06 | cowork 2nd window coder | v3 runner books live on paper (w2_v3_1m, w2_v3_hold) + window 1 w1_hot; scoreboard http://159.223.88.168/hx-c9ois/sw/w2-scoreboard.html, status http://159.223.88.168/hx-c9ois/sw/sw-coder-status.md | LIVE PAPER
- 2026-10-06 | cowork 2nd window coder | Fill check: the spec's depth model R/(R+$500) said 0.9% impact on STONKROBBERS `0xdcc026752d7cbf4fe2a14d53986bd3d22de72024` (R $54k); a real simulated swap of $495 on the pool said 7.6%. EXPLAINED by Analyst 2 (A2.14): STONKROBBERS sat 0.7% from the end of the launcher's range, where the virtual reserve counts ETH that is not in the pool. At the $1M gate real ETH is 92 to 96% of R, so fills match the R model within 0.02x even at $5k | RESOLVED, not an edge killer. Same pass found the old pull guard counted out-of-range burns as pulls; fixed to in-range only, main v3 book now 1.99x at $500
- 2026-10-06 | cowork 2nd window coder, infra | Public Robinhood RPC returns blockTimestamp 0x0 on eth_getLogs; any tool reading log times from it gets "now" or zero. Use block headers. Paid dRPC key hit "User balance exceeded" (HTTP 403) at ~13:00 Bangkok 6 Oct, so PP jobs on it are blind until topped up | PROVEN
- 2026-10-07 | cowork ROT A2 (rotations team) | On-chain USD flows do NOT lead chain token price vs BTC: price leads them by about 8 days. Stablecoin share and growth, DEX volume share, fees share, launchpad heat, CEX spot volume share, OI change all KILLED as lead signals (13 chain tokens vs BTC, 2019 to Oct 2026, 1 day delay, pre-registered) | receipt: http://159.223.88.168/hx-c9ois/rot/a2-status.md , a2-lead-summary.csv | KILLED (as leads; launchpad heat kept as a crowding/top gauge)
- 2026-10-07 | cowork ROT A2 | Funding spike WITHOUT a price spike (Binance funding 7d mean z high, price not yet up vs BTC): token beats BTC over next 14 days, +3.5% vs +2.1% any day (2020-22, n 143), +1.2% vs -1.4% (2023-26, n 445), adds info beyond momentum in both eras (t about 2). Edge about equal to 2% round trip cost, so a ranking tilt only; venue check on Bybit pending | KILLED 7 Oct by ROT Auditor: luck test beats only 93.3% (2023-26) and 73.2% (2020-22) of 1,000 fake runs (bar 95%); 13-coin list is today's survivors; after 2% round trip -0.8% vs BTC (A2.R6b)
- 2026-10-07 | cowork ROT A2 | Plain price momentum vs BTC stopped working after 2022: momentum spike alone 2023-26 -2.2% next 14d vs -1.4% any day | TESTING (agrees with hub "hot is a warning since Sep 2024")
- 2026-10-07 | cowork ROT A3 | Basket rule v3 does NOT beat holding BTC on return: 1 Jan 2023 to 30 Sep 2026, 1% round trip, v3 book +361% (worst -40%) vs hold BTC +400% (worst -53%); 2025 v3 -12% vs BTC -6%. Its only edge is smaller drawdown | receipt: http://159.223.88.168/hx-c9ois/rot/a3-status.md | PROVEN (exact reproduction: 118 trades, PF 2.65)
- 2026-10-07 | cowork ROT A3 | Why v3 weakened since Sep 2024: weak breakouts. Breakouts with a 25%+ 7 day impulse PF 3.25 (n18, avg +12.4%) vs slow grind breakouts PF 1.31 (n59); 2025: 2.95 vs 0.58. Regime, late entry and BTC trend explanations KILLED | TESTING (A3.I3, small n, found by looking)
- 2026-10-07 | cowork ROT Winners | 682 coins did 5x+ in 30 days (2019 to Oct 2026, $500k+ volume at the low) vs 1,609 matched controls: NO single pre-move feature at day -30 to -1 separates them (best lift 1.6) | KILLED as a picker
- 2026-10-07 | cowork ROT Winners | First leg tell: up 35%+ in 7 days while small (under $20M) and young (listed under 180 days) = 4.6x to 7x higher odds of a 5x, both eras. BUT buy and hold 30 days loses on the median (-11% to -28%); 15-20% reach 2x, 2-3.6% reach 5x, 12-23% fall 50% first. Profit must come from exits and sizing | receipt: http://159.223.88.168/hx-c9ois/rot/winners-report.md | TESTING (W.R1)
- 2026-10-07 | cowork ROT Sizer | DATA TRAP: A1's CMC daily files hold 477 one-day bad prints (price 5x off and back next day) 2019 to 2026; they inflated old basket v3 receipts (majors +14.6% on 2024-04-30 while all 10 members fell). Filter them before any equal-weight line. Clean v3 with 1 day delay and 1%: +2.0%/month 2019-22, +0.3%/month 2023-26, worst -64% | PROVEN (data fix); the v3 basket numbers in this hub are inflated
- 2026-10-07 | cowork ROT Coder time machine | Basket v3 catches only 2% of the top 20 monthly movers; v3 +196% vs hold BTC +275% (2023 to now, 1 day lag, 1% cost) | PROVEN as baseline
- 2026-10-07 | cowork ROT Sizer | Winners first-leg signal (W.R1) is only an edge while BTC is up over 30 days (W.R2) and with a 25% trailing stop: +25.3% / +7.3% per trade, PF 2.66 / 1.41 (2019-22 / 2023-26), still positive without the top 5 trades. With BTC down every exit variant loses in both eras. Pyramiding into strength KILLED; more concentration on v3 KILLED (-12.5%/month) | TESTING (H1, not yet moved 20% or audited)
- 2026-10-07 | cowork ROT A1 | Rotation clock 2019-2026: money goes BTC first, then the whole alt market, then ONE hot theme (2020 DeFi; 2021 L1/L2/gaming/memes; 2023-24 AI + Solana memes; 2025 perps then privacy; 2026 BNB memes). Theme legs got shorter: median 63 days (2020-21) vs 36 days (2023-26). Breadth (most sector coins above 20d average) is the first sign but weak: no entry signal proven yet | TESTING
- [stated] 2026-10-07 ROT Auditor: W.R1 (first 35% leg, small young coin) = real WATCH signal (5x odds 3.7x/6.1x base, beats 100% random) but FAIL as hold-30d trade (beats BTC on 28% of trades). W.R2 alone FAIL. W.R3 and A3.I3 KILLED. Old holdout burned for W family; clean tests left: forward paper + 2017-18 CMC cycle. Sizer: W.R2 + 25% trail = PF 2.66/1.41 both eras, survives removing top 5 trades (H1, TESTING). Clean basket v3 is about flat since 2023; pyramiding and top-1 concentration KILLED.
- [stated] 2026-10-07 ROT Coder oracle: basket v3 captures 0.68% (2020-22) / 1.11% (2023-now) of perfect-hindsight rotation; hold BTC 1.29% since 2023. 88-99% of the gap is WRONG COINS (late 7%, too small 4%, early exit 0.3%). Lesson: rotation edge is picking, not timing. ROT Winners after age-bug fix: W.R1 first leg lift on 5x = 9.5x / 5.8x (stronger), but 30d hold median still -12% / -22%; needs exits.
- [stated] 2026-10-07 ROT A3 miss split (1,660 top-20 monthly mover legs, 2019-now): 75% never in the theme-basket universe, 12% right theme but too small, 9% in basket never bought, 2% late/sold early, 1% caught. Rule: rotation scans must cover ALL liquid coins (CMC top 1,500, $1M+ vol); sector baskets only as filter. Biggest misses were untagged/other coins (677 legs).
- [stated] 2026-10-07 ROT H1 (first-leg small young coin while BTC 30d up, cut -10%, trail 25%) KILLED for 2023-26: random coins with same exits beat it 18% of draws, deflated Sharpe 0.51, 2025 -11 / 2026 -6. Real in 2019-22 (99.3%, driven by 2020-21 alt season). Lesson: first-leg momentum on CMC coins is an alt-season-only edge; since 2023 the runners start on DEX launches. Next: H1 gated by an alt-season breadth gate (registered first).
- [stated] 2026-10-07 ROT H1-ALT (H1 traded only when the share of liquid CMC coins beating BTC over 30d is above its 1y median, plus BTC up): +373% since 2023 vs hold BTC +275%, +846% 2020-22, worst drop -65%/-34%, positive in all 16 gated cells. Auditor: FIX not FAIL (random-coin test 99.6% pass, but deflated Sharpe 0.87 and whole-search shuffle fail; negative without top 5 trades). Status PAPER; one-shot clean test on untouched 2017-18 data pending. Data lesson: basket v3 "+196% since 2023" was bad prints; clean = about +6%.
- [stated] 7 Oct 14:20 Bangkok: D-022 rings retired; wake roles by moving their armed DOORBELL with update_trigger (now + 2 min), each role makes a fresh send_later doorbell on every wake. ROT headline back to "default hold BTC"; H1-ALT stays a labelled paper test because its near settings all lose to BTC.
- [stated] 7 Oct 14:31 Bangkok, ROT clean 2017-18 one shot on frozen H1-ALT: FAIL. Book +349% vs BTC +275%, PF 2.2, but random liquid coins with the same exits matched it in 32.7% of draws (bar under 5%). Lesson: picks add nothing; season (alt gate) + exits carry returns. 2017-18 is burned. Next test: SEASON family (broad small-alt basket when gate on, BTC when off) on 2019-26.
- [stated] 7 Oct 14:44 Bangkok CORRECTION: "picks = dart" holds only in 2017-18. In 2019-26 the H1-ALT picks beat 99% of random books; season + exits alone lose to BTC (median random book 2023-26 -9% vs BTC +274%). SEASON cancelled. H1-ALT edge is real in the recent cycle, not proven across cycles; forward paper calls decide.
- [stated] 7 Oct 15:55 Bangkok: Ben shared @Moonbl0x claim "trenches now front run the broader market". Check: prior test 2024-26 weekly lead-lag ~0, memes topped median 10d before alt tops (weak). Live 7 Oct: pump.fun volume rising ($100M mid Sep to $187M 5 Oct), RH trenches dead since mid Sep, BTC chop 83.5k to 86.6k. Verdict UNPROVEN as a timing tool; queued as an H1-ALT exit test for ROT A3.
- [stated] 7 Oct 15:51 Bangkok, ROT Auditor fragility: H1-ALT beats BTC in 8 of 12 clean half years (random 9 of 12), but since 2023 Jan to Jun 2024 (+168%) carries it; without that half year +67% vs BTC +133%. Concentration FAIL. Live $100 (D-025) treated as a fill/haircut measurement, not proven edge.
- [stated] 7 Oct 15:55 Bangkok, ROT Winners: H1-ALT picks did not get worse in 2017-18; random coins got better (whole pond young and hot). Picks reach 2x in 30d: 44% (2017-18) and 18% (2019-26) vs random 21% and 5%. The entry filter (young, small, +35% week) adds at trade level in every cycle; the 2017-18 random-test fail came from a hot pond, not a broken filter.
- [stated] 7 Oct 16:00 Bangkok, ROT A3: "trenches lead the market" tested as an H1-ALT exit (half the book when $5M to $100M small-cap basket 20d return turns down while BTC within 5% of 30d high), 2019-26: lower return, no smaller worst drop, random timing did better. KILLED.