# 100x Runner brief (Robinhood Chain), for a new Cowork session

Owner: Ben (Bangkok). Written 5 Oct 2026. Read all of this before doing anything.

## 1. The job
- Find coins that will go **100x**, and buy them **in their first 3 days** (not in the launch seconds; another process does that).
- Ben's thesis: a real 100x has real builders behind it, a real idea, and weeks or months of preparation. If that can be spotted early, it is a money printer, but it will not come every day.
- First task: **learn from past runners.** Build the full list of past 100x coins, find what was visible about them on day 0 to day 3, compare with look-alikes that died, and turn what holds up into rules. Then run them as a paper book.

## 2. What counts as a 100x (Ben's rule)
- Measured from the early price (first trades) to the peak.
- It counts only if it was **sellable**: real buyers paid at least about $20k at or above the 100x price, so Ben could sell $10k to $20k there. It does NOT need to hold for a day.
- This money test removes fakes:
  - coins that started at a near-zero price (seen: a "400x" whose first 30 s had 0.009 ETH of buys)
  - one-trade spikes
  - GMGN ATH numbers, which are polluted (WALL shows $8.8B)
- Also report 30x and 10x runners from the same entry window, as backup classes.

## 3. Ben's rules (non negotiable)
- Talk to Ben like he is 5: very short bullets, plain words, **no dashes or em dashes**, one action word per alert (buy, strong buy, hold, sell).
- **Bangkok time (UTC+7)** in everything he reads.
- **Never put a secret on screen.** Check configs with `/opt/common/showcfg.py`, never `cat`.
- Breaks fix themselves with redundant checkers (systemd timer AND cron, flock). Alert Ben only when every fix failed, through `/opt/common/alertgate.py`.
- Work from first principles. Push back when there is a better way. Never call a coin small or thin as an objection (Ben can buy $300).
- Every tested insight goes into the memory ledger (project file `trading-edge-hub-4` or a new part) the same turn: rule, receipt, status.
- Anything Ben hands to another AI must be one public URL (droplet public folder, below).

## 4. Hands off other processes (read only)
- Do NOT edit, restart or slow down:
  - the launch pickers in `/opt/hist` (aipick*, sentinel, pickguard, reconcile, hist-rh)
  - the lane radar in `/opt/laneradar`
  - the chain brain in `/opt/chainbrain` and `/opt/brain/chain.json`
  - runtrack in `/opt/runtrack`
- Read their data freely. Build in your own folder: `/opt/hundredx`.
- The droplet has 2 CPUs and 4 GB RAM and is usually busy (load 4 to 6). Run heavy jobs with `nice -n 19 ionice -c3`, or export data and work in the cloud container.

## 5. How to reach the server
- Droplet "kookwatch", 159.223.88.168. Cloud sessions cannot SSH. Route: Ben's Mac via Desktop Commander, then `cd ~/kookwatch; python3 dbridge.py run '<cmd>' <timeout_s>`. Device calls time out at about 60 s, so start long jobs with `nohup ... &` and poll the log.
- **Never `pkill -f` a pattern that also appears in your own command line.** It kills the bridge command itself.
- Files to the droplet: gzip plus base64, echoed through dbridge, under about 12 KB per call.
- Public folder (no login): `/opt/launchwatch/pub/hx-c9ois/<sub>/` is served as `http://159.223.88.168/hx-c9ois/<sub>/`. Use it to move exports to the cloud container (curl). Verify with HTTP 200 plus md5.

## 6. Data that already exists
- **`/opt/hist/rh/YYYY-MM-DD.db`**: every RH launch rebuilt from the chain, one SQLite file per Bangkok day. Done back to mid Sep and still filling back to 3 Jul. Tables:
  - `launches` (token, ts, creator, factory, name, symbol, decimals, supply)
  - `pools` (rowid = pid, c0, c1, fee, hooks, sqrt0, block)
  - `transfers` (first 30 min only)
  - `swaps` (first 6 h)
  - `bars` (1 minute bars for 14 days, for pools that came alive)
- **Price traps:**
  - Bar and swap price is currency1 per currency0. When the token is currency1, invert the price and swap hi and lo.
  - a0 is the trader's change in currency0; a0 < 0 means the trader paid currency0.
  - No trades means no bar: hold the price flat, and keep dead coins in every test.
- **Launch floor:** after the first 40 s only about 7% of launches ever trade below their launch price.
- **777 family:** token addresses ending in 777 often show broken prices (about 3e-39) and had 0 10x winners in 1,880.
- **`/opt/runtrack/runtrack.db`**: every RH coin with ATH $500k+ (about 880 tokens). Tables:
  - `rh_flow` (10 min: real holders, buy and sell USD, buyers, sellers)
  - `kline` (GMGN hourly)
  - `rh_wnet` (hourly net per wallet)
  - `tokens` (launchpad, created_ts)
  - The best place to start the runner list (then confirm each one with the money test).
- **`/opt/laneradar/lanes.db`**: every new coin since early Aug with its first-minute text and socials, category tree labels, novelty labels, and prices for 72 h.
- **`/opt/laneradar` deepread and ownscan**: website, docs, app check, GitHub (commits, authors, weeks active, own code vs library), domain age, token and hook code family, "rare own functions".
- **`/opt/price-predict-rh/paper/gstream.db`**: GMGN live feeds since 3 Oct (smart money, KOL buys, signals, hot searches).
- **Export of launch price paths** (first 6 h, per launch): http://159.223.88.168/hx-c9ois/brx/launchpath.csv.gz
- **Keys (never print):**
  - `/opt/common/rpc_keys.json` (dRPC)
  - `gmgn_key` (shared budget, never retry loops)
  - `twitterapi_key` (pay per use, about $0.15 per 1,000 tweets; `since_time:` and `until_time:` work for history)
  - `anthropic.json` (Claude API, needs the `anthropic-workspace-id` header; credit is low, so warn Ben when it runs out)
- RH RPC needs a browser User-Agent. Uniswap v4 PoolManager `0x8366a39cc670b4001a1121b8f6a443a643e40951`.

## 7. Lessons from earlier sessions (re-test, do not assume)
- Project text predicts **who survives 10+ days** (AUC about 0.70), not the launch-day pump (AUC about 0.5). That fits the 100x thesis.
- The launch tool (factory) and the code template predict launch-day 10x. One template had 3.2x the normal 10x rate, one huge template 0.08x.
- For coins already 1 to 11 days old, short 2x moves were NOT predictable from wallets or X in the 12 to 24 h before (AUC 0.49 to 0.55, 1,032 doubles vs matched look-alikes). So the 100x edge, if it exists, should come from **who built it and what it is** plus how it behaves over days 0 to 3, not from short-term flow alone.
- Survivorship killed earlier studies. Always include coins that died.
- Measure what was knowable AT the decision time. A later read can see the pump.

## 8. Method
1. **Runner list.** Every RH coin since Jul that passes the sellable 100x test (and the 30x and 10x classes), with launch time, peak time and peak price.
2. **Look-alikes.** For each runner, 3 to 5 coins from the same launchpad and week that looked similar on day 0 to 1 (similar day-1 market cap or volume) but did not run.
3. **What was visible on day 0, 1, 2, 3**, for both groups:
   - builders: GitHub weeks and authors, own code, domain age, X account age vs token, docs, a working app, team history, test deploys before launch
   - story: category, novelty, fit with a hot meta, RH-native
   - money and holders: holder growth per day, new wallets, who sells, dev behaviour, liquidity added
   - attention: X mentions and unique authors per day, big accounts, Natan, long.xyz and Robinhood replies
4. **Score honestly.**
   - Split by time: learn on older weeks, test on newer weeks.
   - Count hits in the top picks, not averages.
   - With few runners (expect dozens, not hundreds), prefer simple rules and a checklist over a big model.
5. **Paper book.** Score every coin on day 1, 2 and 3. Buy the top few on paper. Track the honest result (sell at 100x or at the time limit, real fills).
6. **Deliver.**
   - Ben gets 5 short bullets plus a public URL to a report: the runner case file, what held up, what was killed, the live paper book, and the single missing piece with when you will have it.
   - Telegram alerts only for STRONG BUY, only after the rules beat random on held-out weeks.

Version 1, 5 Oct 2026. Latest: http://159.223.88.168/hx-c9ois/brx/hundredx-brief.md