---
name: chain-brain
description: Project copy of the Robinhood/Solana chain regime brain (/opt/chainbrain, chain.json, daily gauge with lifecycle phases). Older history (v1 to gauge v1) is in the global /areas/chain-brain.md. Read before touching /opt/chainbrain or any "is the chain bottoming / topping" question.
sources: [cowork]
aliases: [chainbrain, chain brain, chain gauge, chain phases, lifecycle model]
---
- Full history v1 to gauge v1 (5 Oct 2026): see global /areas/chain-brain.md. Code /opt/chainbrain (chainbrain.py 15 min, macro.py hourly, gauge.py daily 08:47 Bangkok, chainfix, score). Backup gauge.py.bak-v1.
## Gauge v2: lifecycle phases (5 Oct 2026)
- [stated] Ben: model must be chain-wide (not only meme price), say "more downside likely" or upside, and have phases: turning signals at the bottom, slow wallet growth = not overheated, parabolic wallets + other signs = take profit, turning signals at the top. Price Predict DB is read-only for this session: never write into it.
- Phases (daily, replayed, no look ahead): FALLING (40%+ off 90d high, under 20d avg, fees 7d < 0.8x 28d or users 7d < 0.9x 28d) = do not buy; BASING (cheap, under 20d, bleeding stopped) = get ready; TURNING UP (cheap, above 20d, BTC above 50d) = buy small; CLIMBING = hold; HOT (fees 7d > 1.5x 28d and 20%+ above 20d) = take some profit; TOPPING (was hot, back under 20d) = sell rest; COOLING.
- Test: 12 chains via Artemis (DAU, PRICE, FEES 2020 to 2026, solana ethereum bsc avalanche sui base arbitrum tron polygon near aptos ton) + Solana memes 2024 to 2026. Next 60 days, 25% drop before 30% rise: FALLING 43% chains / 51% Sol memes; BASING 35 / 39%; CLIMBING 23% chains (safest, 10 of 12 chains positive). HOT is not a sell-all: chains 49% and Sol memes 66% rose 30% first.
- Killed: "parabolic wallets = top" (12 chains: wallets 28d vs 90d > 1.7 gave 44% rise-first; Sol memes new-user surge 1.5x = +57% next month). On RH wallets were a bad heat gauge (airdrop spike 14 to 18 Aug 1.27M, flat during the Sep wave, peaked 12 days after the price top). Fees (money paid) led; wallets lag.
- RH replay: FALLING since 18 Sep at index 0.77 (now 0.42, another -45%), HOT 29 Aug (1.47), TOPPING 8 Sep (1.31), TURNING UP 21 to 23 Aug (0.95).
- 5 Oct live: Robinhood FALLING (fees 56% of normal, users 89%), Solana memes CLIMBING.
- Price Predict tokens.db (read-only): daily launches (birth) RH 30 Apr to now. 5 Oct ~12k to 14k a day = same as the 19 to 23 Aug bottom (12k); peak 26k 8 Sep; launches lagged the price top by 5 days.
## Gauge v3: meme phases (5 Oct 2026 ~16:50 BKK, five supervisor iterations)
- [stated] Ben: run five more iterations, act as own first-principles supervisor, elite level; he is backfilling chain data into Price Predict (read-only for me).
- DATA TRAP: runtrack has only coins alive ~2 Oct (15 crossings in July of 879), so the RH coin index before late Aug is survivor-biased (wave 1 drop understated). Unbiased RH inputs: DefiLlama fees + DEX volume, Blockscout wallets, tokens.db births, and Ben's chainday.db backfill (bridge_raw/bridge_eth, building at 16:50 BKK, back to 28 Aug).
- Chain coins trend, memes mean-revert. v2 phases held on 12 chain coins (FALLING days 42% and 44% in both halves, block bootstrap 90% CI 38 to 49% vs base 29%) but NOT on CMC meme caps (ETH, BNB, Base, Solana 2021 to 2026: FALLING 43% vs base 44%). As an episode entry signal FALLING is not bad (33%); it only marks long bleed regimes.
- Meme phases v3 (5 meme markets: CMC caps of memes >= $1M with vol filter for ETH/BNB/Base/Solana + Solana panel), 25% drop first vs 30% rise first in 60 d: STRETCHED (30%+ over 20d avg) 59/40, both halves 55% and 63% = TAKE PROFIT; DEEP STOPPED (60%+ under 90d high, under 20d, fees 7d >= 0.8x 28d) 30/66, halves 39% and 25% = BUY scale in; DEEP BLEEDING (same but fees < 0.8x) 40/59, Solana 73% = GET READY; RECOVERING 46/54; NORMAL 44/42. Days since the high did not matter.
- Caveat: summed caps gain from new coins joining (inflates rises); meme results are from 5 series, Solana memes only one chain of the RH type.
- RH v3 replay: DEEP STOPPED 15 Aug (0.91, low 0.84 18 Aug, top 1.86) good; STRETCHED 27 Aug (1.43, top 1.86 3 Sep) good; DEEP STOPPED 13 Sep (0.94, then 0.42) bad; DEEP BLEEDING since 18 Sep. Live gauge.py v3 (backups gauge.py.bak-v1, -v2); message leads with the meme phase; v2 phase kept in JSON.
- Remaining drop when FALLING: median 30% more (chains), low ~53 days out; barely worse than all days (24%), so it says little about size.
- Launches signal: RH 12k to 14k a day = Aug bottom level with more wallets (326k vs 263k), so launches per wallet are lower than at the Aug bottom. Untested elsewhere: CMC new listings too sparse, Artemis has no pump.fun launch count, Dune skipped for the credit budget.
## Gauge v4 (5 Oct 2026 ~17:50 BKK, ten more iterations; [stated] Ben asked for 10 iterations then run the backfill)
- KEY FIX: summed meme caps (CMC, vol filter) fake outcomes because coins drop in and out of the filter. Rebuilt labels as TRUE HOLDINGS (coins in the basket on day t, their caps later; delisted = 0) for ETH/BNB/Base/Solana memes 2021-26 (/home/claude/cb/memes/fixedpaths.pkl, multi/memeh.py build(True)). Normal = 41% chance of a 25% drop before a 30% rise in 60 d.
- Survives true holdings: DEEP BLEEDING (60%+ under 90d high, under 20d avg, fees 7d < 0.8x 28d) 52% (halves 58, 49; Solana 84%) = more downside likely. Deeper (-80%) 54%. DEEP STOPPED (fees >= 0.8x) 38% = danger gone, no strong buy edge (halves 53, 28 unstable).
- Killed on true holdings: STRETCHED as a sell (42%), STRETCHED + BTC down (looked 84%, is 40%), DEEP STOPPED + BTC up, fees hot as a top (41%), fees washed out, reclaim after deep. The 16:50 v3 numbers (59% stretched, 30/66 deep stopped, trade sim +128%) were the filter artefact. No reliable daily top signal for memes; use trim slices.
- Stablecoin supply change (Artemis, chain level) adds nothing at meme bottoms.
- Price Predict chainday.db (READ ONLY): bridge_eth daily 1 May to 5 Oct (L1 escrow), stables_day (USDG, U, USDe, rwaUSDi). Bridge 7d net: wave 1 peak 80k ETH/wk 14 Jul (same week as price top 16 Jul); Aug low 7.6k/wk on 19 Aug (low 18 Aug), never negative; Sep wave peak 74k/wk 6 Sep (3 days after top); first net OUTFLOWS ever from 27 Sep (-10k to -13k ETH/wk). RH stablecoins flat ~1.06 to 1.10B since 9 Sep. Bridge = coincident to lagging at turns (n=2), shown in gauge as a receipt, not a rule.
- gauge.py v4 live (backup gauge.py.bak-v3): RH DO NOT BUY YET (DEEP BLEEDING, fees 56%, bridge -8,877 ETH 7d); reads chainday.db with mode=ro only.
## Gauge v5 (5 Oct 2026 ~18:30 BKK)
- [stated] Ben: never said memes only, wants the entire chain; questioned "97% under the high"; asked if we are near the bottom, how far and how many days; five more iterations.
- CORRECTION: "-97% vs 90d high" came from the equal weight log index (macro.py gi), where dying coins drag the average. CMC robinhood-placeholder memes (cap weighted): 25 biggest coins on 3 Sep top are -67% as a basket, median coin -65% (AI 246M to 130M, CASHCAT 297M to 165M, UP 387M to 7M). gauge.py v5 now uses a chained top-20 cap weighted CMC basket (seed /opt/chainbrain/rh_cmc_seed.json built by rhseed.py on the droplet, 89 days from 8 Jul; daily append via the Solana CMC fetch into rh_cmc.json); reads -65% vs 90d high. chain.json/macro.py still on the old index.
- Chain-wide cycle clock (12 chains Artemis fees+price, 2020-26, + pump.fun/Solana memes): event = fees 7d fall to 25% of their 60d peak. In 14 cases 15 to 50 days after the fee peak with price already -35%+: price low came median 62 days later (p25 44, p75 115), median another -40% (random day baseline -25%), only 1 of 14 had the low within 10 days. RH event day 29 Sep (fee peak 7 Sep, fees now -80%, DEX volume only -50%) gives low window early 12 Nov, mid 30 Nov, late 22 Jan. Printed daily in the gauge.
- Not done: CMC meme cap indices for ETH/BNB/Base are too noisy for chained indices (fake 2x days), dropped.
## Gauge v6 + out of the box round (5 Oct 2026 ~19:00 BKK; project doc claude/out-of-the-box-thinking.md)
- [stated] Ben asked for 5 iterations thinking: what if the whole approach is wrong, are the gaps significant, unused data, missing data; save as "out-of-the-box thinking" in the project folder.
- New chain survival (Artemis fees, 19 chains): after fees -80% from the launch-era peak, only 6 of 19 regained 50% of that peak within 180 d (Base +64 d then 8x, Abstract, Unichain, Sei, Blast, Soneium); 13 never did. RH comeback likely but ~1 in 3 long-fade risk.
- RH big-coin daily returns vs BTC/ETH/SOL/Solana memes: corr 0.2 to 0.3, RH runs on its own catalysts.
- Meme cap / chain stablecoins: RH 0.49 now = Aug low (0.3 to 0.5), Sep top 1.3; on Solana the ratio did not time bottoms (stables grow structurally). Info only.
- CATALYST SIGNAL (best lead found): new app with >=3% of chain fees that had <=1% 28 d earlier. Pons V2 fired 10 Aug (8 d before the 18 Aug low), ~$1.6M/wk plateau 14 to 23 Aug, then exploded to $10M/wk by 29 Aug. Fables firing since 25 Sep, $1.5M/wk flat ~9 days, 4.6% share. IGNITION = new app weekly fees >= $1M and 2x prior week.
- Chain fees $33M/wk vs Aug bottom week $16M; falling ~25%/wk, reaches Aug level ~20 to 25 Oct at that pace.
- gauge.py v6 (backup gauge.py.bak-v5) adds catalyst line and weekly fees vs Aug bottom.
## Gauge v7 (5 Oct 2026 ~19:30 BKK)
- [stated] Ben: the chain is much bigger now, so a return to the first bottom's absolute fee level ($16M a week) is not what he expects and would not be a good sign; asked whether Fables is a buy signal.
- Fables = chain-wide early warning (a new app pulling in fresh money, like Pons V2 before the Aug wave), not a coin buy signal; Fables has no token of its own.
- Fees per active wallet per week (7d fees / 7d median Blockscout active wallets): Aug bottom ~$56 to 73 (airdrop days excluded), Sep top ~$375, now ~$86 to 93. Scaled to today's ~350k wallets, the Aug-bottom level is ~$21 to 23M a week (now $33M, falling ~20 to 25% a week). Measured from its own peak, wave 1 fell 52% (weekly fees), wave 2 has fallen 79%.
- gauge.py v7 (backup gauge.py.bak-v6) replaces the "$16M Aug bottom week" line with fees per active wallet.
## Fables check (5 Oct 2026 ~20:00 BKK)
- CORRECTION: Fables does have a coin path: PROLOGUE 0xb9972CA7188e511174947E3936a5315ac7073277 (pools.trade launch 17 Aug, pair 0x8651e656738064177752a395dbde2b2a9e3fc469edc2a9212e6060c0990bb7eb), redeemable into FABLES.
- Fables (fables.fi docs, read 5 Oct): dynamic-fee ve(3,3) DEX on Uniswap v4 for tokenized stocks (Aerodrome-style), not a launchpad like Pons V2. Fees $1.5M last 7 d, $4.8M in 49 d, all to LPs today (no protocol fee; 10% claim fee to treasury).
- FABLES launch 20 Oct (snapshot 17 to 18 Oct; any outgoing PROLOGUE after the snapshot voids the claim): 25M FABLES for PROLOGUE holders, worst case 40 PROLOGUE per FABLES (better when tokens sit in pools/exchanges). 75M liquid at launch, 240M allocated at launch, 1B full plan (500M emissions). veFABLES voters get the fees of staked markets.
- PROLOGUE 5 Oct: $0.00573, cap $5.7M, pool $562k, 24h volume $339k. Worst-case implied FABLES $0.23 -> liquid launch cap ~$17M, 240M ~$55M, 1B FDV ~$229M vs ~$78M a year fee run rate.
- 20 Oct = dated RH catalyst for the gauge's catalyst watch.