# RH big runners, round 2: earliest signs (5 Oct 2026, ~14:20 Bangkok)

## Read this first
- **The first report was wrong on one big point.** It used runtrack, which only lists coins still alive around 2 Oct. It missed 98% of coins that pumped and died fast. It also counted coins that *launched* at $50M to $160M as "runners". Its "STRONG = 17% reach $20M" was mostly that bias. On clean data the same rule gives **1.8% (1 of 55)**.
- **The new data:** every Uniswap v4 launch on Robinhood Chain from 15 Sep to 4 Oct. That is 46,343 coins, about 2,200 a day. The study tracks each one's first 6 hours, swap by swap, then 7 days hourly for the 640 coins that crossed $500k.
- **Most fast $2M coins are two operations, not organic coins.** Each is recognised by its hook code at second 0:
  - **CAMPAIGN_WETH**: about 3 launches a day. Buyers using normal apps never sold in the main pool. PANDAKEY had 61 such buys and 0 sells. Price Predict booked PANDAKEY as a +8.8x paper win, and that win is probably not real. **Avoid this family.**
  - **BUNDLE_ETH**: about 7 a day. The launch transaction buys through about 40 wallets at once. Normal trading works. From a 30 s entry, the median gain is about 2.5x (top quarter 4.9x) and the peak comes after about 14 minutes. **Only good for a fast flip.**
- **Organic coins:** about 26 a day cross a sellable $500k, but only about 1 a day reaches $2M. A model can pick out the $1M coins early (AUC about 0.85 to 0.9). It **cannot** tell which ones will reach $2M+ (AUC about 0.5 to 0.7).
- **$20M+ coins come from somewhere else.** Of 48 runtrack coins at $20M+:
  - 27 were already worth $2M+ in their first hourly candle;
  - 11 started between $200k and $2M;
  - only 10 ran up from under $200k, and those took **15 to 799 hours (median about 4.5 days)** to peak.

  Many traded on venues our launch data does not see: Uniswap v3 (SC, HI, SIF, SI), Pons V2 (AGE), and a bridged token (TIBBIR). Others re-ignited days later on a new pool, like Agrippa: launched 17 Sep, ran on 24 Sep.
- **What that means:** the launch minute is where 10x pops happen, which is Price Predict's game. The $20M game is played over **days**. The earliest real sign is a coin that crosses a sellable $250k to $500k and then **keeps going for days**. That has to be tracked on **every venue**, starting at the crossing, not only for survivors.

## Answers to Ben's questions
- **First-minute identifiers of a runner:**
  - The **hook code family** sorts out the operations: 81 to 86% of family coins reach $500k, against 1.2% for organic coins.
  - For organic coins, what marks a likely **$1M** runner:
    - 100+ distinct buyers in the first 5 minutes: 1.4% reach $1M, about 7x the base (AUC 0.81);
    - alumni wallets, meaning wallets that were early in past $1M coins (AUC 0.78 to 0.80);
    - $5k+ bought in the first 2 seconds (3x the base);
    - the first buy's size is weak (1.5x at most).
  - **None of these separate $2M+ from $1M.** Big organic coins have *fewer* alumni wallets (median 0 against 20).
- **The next earliest signs, by time:**
  - 30 s to 10 min: the $1M model works (AUC 0.84 to 0.89), but its top 3 per day only hit 14%.
  - 30 to 60 min: top 3 per day hit $1M 22 to 28% of the time.
  - For $2M+, nothing gets reliable inside the first hour.
- **When can we say $10M to $20M with good accuracy?** Not from launch data. The fast crossers peak within hours and almost never come back: the 7-day peak equalled the 6-hour peak for all 11 coins that reached $5M+. The ones that went big did it over days, from places the launch tape misses.
- **"Outside the launch pages" means:**
  1. a token whose real pool was on Uniswap v3, Pons V2, or came over a bridge;
  2. a token minted earlier whose real pool opened later;
  3. a token that opened with deep liquidity at $10M to $160M.

  The 30 s picker misses all three. Type 3 is not a runner at all.
- **Is around $500k to $1M where we spot it?** Not by itself. Of organic coins crossing a sellable $500k, 18% reached $1M, 4% reached $2M, 0.6% reached $5M and 0.2% reached $20M within 7 days. The crossing is a starting gate, not a signal.

## Five iterations (what each one changed)
1. **Data audit.** Found three gaps:
   - runtrack survivorship;
   - 71% of day-DB launch rows have no supply. That covers 24k traded coins, including 42% of the $5M+ ones. Fixed for this study with a Multicall3 backfill.
   - the launch tape covers Uniswap v4 only.
2. **Clean minute ladder** over 46k launches. Found the two hook families behind most fast $2M coins.
3. **Can outsiders actually trade them?** Checked router by router:
   - campaign family: 99.6 to 99.9% of swaps go through one private router per coin, and 0 public-router sells (CHIPINU, PPORT, NFLOAT, RBPERP, STKDOG, PANDAKEY; MCHAT had 4);
   - bundle family: public buys and sells both happen.
4. **Organic checkpoint models** (walk-forward by day). Results in the table below.
5. **7-day follow-up** of all 640 coins that crossed $500k (GeckoTerminal hourly, main pool), a re-test of the old score, and a look at how the 48 big runtrack runners were born.

| checkpoint | organic, target $1M+ after T: AUC / top-3-a-day hit | organic, target $2M+: AUC | all incl. bundle, $2M+: top-3 hit / median entry cap |
|---|---|---|---|
| 30 s | 0.87 / 14% | 0.63 | 36% / $484k |
| 60 s | 0.88 / 3% | 0.45 | 42% / $628k |
| 5 min | 0.89 / 6% | 0.70 | 44% / $1.0M |
| 10 min | 0.88 / 14% | 0.59 | 50% / $1.15M |
| 30 min | 0.91 / 22% | 0.68 | n/a |
| 60 min | 0.83 / 28% | 0.57 | n/a |

The $2M+ test period has only 10 organic positives over 12 days, so read that column as "no signal".

## Out-of-the-box stack
- **Are we using all the data?** No. Unused or underused sources:
  - runtrack hourly flows (rh_flow, rh_wnet, holders_hist);
  - casig X and website data;
  - deep-read cards;
  - owncode fingerprints;
  - sellerwatch logs;
  - the pool's own router mix (the strongest operation fingerprint found today).
- **Do we know all our data?** Two traps were found:
  - day-DB supply holes, which make mcap.py drop coins;
  - Price Predict's `pad` takes the hook from the mint_to pool, which misses the trading pool's hook for the campaign family.
- **Gaps that matter:**
  - Uniswap v3, v2 and Pons V2 pools are not in the launch tape;
  - swaps carry no router or sender, so you cannot see who can sell;
  - runtrack only has survivors;
  - there are no DexScreener "paid / boost" flags, and those are campaign markers (X showed "DexScreener Paid" bots and CMC vote spam on every campaign coin);
  - there is no multi-day, all-venue tracker of coins after they cross $250k.
- **Is the approach wrong?** Partly. "Market cap" (price times supply) is not money. In thin or private-router pools it is fake. The target should be **value an outsider can sell**. Big runners are a slow, multi-day, multi-venue story. Hunting them in the launch minute is the wrong place.

## Proposed code (nothing deployed, nothing in the other session edited)
- `family_check.py <token>`: finds every v4 pool of a token, fingerprints the hook code, and probes buys and sells per router. It returns AVOID (campaign or honeypot suspect), FAST FLIP ONLY (bundle), or no known operation family. Tested on PANDAKEY, where it flagged the campaign pool: 40 public buys, 0 public sells.
- `supply_backfill.py`: fills missing supply and decimals into a **new** side DB (`/opt/hist/analysis/supply_extra.db`) through Multicall3, about 4 minutes for 80k tokens. Read-only on the existing day DBs.
- **Proposed build, "ignition watch":** watch every venue (v4 PoolManager Swap, v3 factory PoolCreated plus Swap, Pons V2, v2) for any token, new or old, that crosses a sellable $250k. From that moment, record hourly: sellable cap, outsider buyers and sellers per router, holders, alumni wallets, and X authors. After about 4 weeks there will be an unbiased set with 30+ big runners to learn the day-1 to day-5 pattern from.
- **Hand to Price Predict:**
  1. add the hook *code* family of the trading pool as a feature;
  2. block the campaign family;
  3. re-check the PANDAKEY +8.8x paper win;
  4. backfill supply.

## Receipts (data and scripts)
- Data, read-only copies:
  - hist day DBs 15 Sep to 4 Oct;
  - runtrack export;
  - GeckoTerminal hourly OHLCV for 640 coins;
  - twitterapi.io search of the contract address, from 1 h before the $500k crossing to 1 h after (695 coins);
  - public RPC getLogs and eth_getCode;
  - Multicall3.
- "Sellable" means the highest market cap at which a sell of $100 or more actually executed.
- Scripts in `scripts_v2/`: tl.py (extract), build.py (features), prep.py (clean + families), an4.py (checkpoint models), an5.py / an7.py (milestone + 7 day), vets.py (alumni wallets), xq2.py (X), gt7.py (7 day), codes.py (hook code fingerprints).

First principles: about 80%. The facts come straight from swaps, routers and code hashes. Still missing: v3 and Pons V2 data, true outsider-sell checks on every coin, and enough $20M+ cases to learn from (n≈10 in the clean window).

Version: v2, 5 Oct 2026, about 14:20 Bangkok. Latest: http://159.223.88.168/hx-c9ois/runners/runner_patterns_v2.md