# Phase 2 runners: the process (v1)

Owner: phase 2 session. 5 Oct 2026, 16:00 Bangkok. Live buying is OFF. Data collection is ON.

## Read this first
- **Start (the gate):** a coin is worth $250k AND has really traded both ways (3+ buys, 3+ sells, $5k+ volume in the last hour). Any coin, any venue, new or old.
- **Win:** sellable at $2M, meaning $20k+ of buying happened at $2M or more, so we could sell there. Big win $5M, jackpot $20M.
- **Two shapes:** quick (within about 6 hours) and slow (up to 45 days).
- **What the data says today (15 Sep to 4 Oct, new launches):**
  - about 130 coins a day pass the gate;
  - about 1.7% of them become sellable at $2M (about 2 a day);
  - 0.19% reach $5M and 0.04% reach $20M.
- **Almost every quick win comes from one launch type, the "bundle" coins.** At launch, about 40 wallets buy together in one transaction. About 6 of these a day pass the gate. 30% of them become sellable at $2M, in about 16 minutes.
  - Paper test: buy at the gate, sell at $2M or at minus 50%. Result: +30% per trade in old weeks and in new weeks.
  - Selling works: the middle bundle coin had 192 different sellers in its first 30 minutes.
  - Caveat: these coins often have their liquidity pulled (sell session finding), so this needs a live paper check before trust.
- **Plain coins (no hook)** pass the gate about 115 times a day, but only 0.1% reach $2M. Today these are noise.
- **"Campaign" coins** (one private router, outsiders cannot sell): always skip.
- **Slow wins are not visible yet.** The launch tape stops at 6 hours, and slow runners often move to another venue or a new pool (Agrippa, AGE). The new collector fixes this from today. The Price Predict data fix fills the past.
- **Sell plan (for when we trade):**
  - sell most at $2M;
  - keep a small part only when the signs stay strong.
  - Reason: of the coins that reached $2M, only about 1 in 9 reached $5M within 6 hours.

## Stage signs already found (runners 20M study, 16:10 Bangkok; from a survivor sample, so they are re-tested on the unbiased collector)
- The clock starts at the first hour with $5k+ volume. The middle small-start $20M runner was at $1M by 1.5 h, $5M by 6.5 h and $20M by 22 h. 1 in 4 took more than 3 days.
- **At 30 minutes:** runners had a lower top-10 holder share (3.6% vs 14.8%) and more holders (482 vs 174).
- **At 8 hours:** a coin still at $3M+ reached $20M within 7 days 29% of the time (this caught 12 of 20 runners).
- **Churn** (volume divided by cap) was lower for runners: 1.0 vs 2.1. This matches the low churn sign in the 100x study.
- **X:** real authors in hours 4 to 24 were 24 for runners vs 2 for stallers.
- **Bundle caveat:** bundle coins were back near $6k by 6 h, and liquidity was removed within 8 h in 4 of 6 checked. So the quick lane must sell at the target and never hold.

## The process, step by step
1. **Gate (every 5 minutes, all venues):** the coin is worth $250k AND it traded both ways in the last hour. This writes the trigger time, cap, buys, sells, venue and pool age.
2. **Screen (seconds):**
   - launch family from the hook code: campaign = skip, bundle = quick lane, plain or other = slow lane;
   - is it a new launch or an old token waking up on a new pool;
   - can outsiders sell (public sells exist);
   - supply and decimals checked on chain.
3. **Odds (at the gate, then at 15 min, 1 h, 4 h, 8 h, 24 h, then daily):** chance of reaching sellable $2M, $5M and $20M. Two scores, one per shape. Today only the family and two gate signs are proven (below). Every other sign is collected and tested weekly.
4. **Paper position (later, not now):** buy only when the odds beat the base rate on new weeks.
5. **Sell ladder:**
   - At sellable $2M, sell 75%.
   - Keep 25% as a runner bag only if all of these hold:
     - new buyers per hour are still rising;
     - sells are being absorbed (price holds while sells happen);
     - holders are rising;
     - real people on X are rising.
   - Close the bag at minus 40% from its peak, or if buyers dry up for 6 hours.
   - At $5M, sell half the bag. Hold the rest for $20M with the same stop.
   - All of this is tested on paper before use.
6. **Learn (weekly):**
   - fit on older weeks, test on newer weeks;
   - count hits among the top picks;
   - keep a sign only if it works in both halves.

## Five iterations (what each one found)
1. **Gate definition.** "Price" alone is fake in thin pools. The gate needs real two-way trading, and the win needs $20k of buying at the target. Built from the v4 tape (Sep 15 to Oct 4): the first $100+ sell at $250k+ marks the moment. 261 gates a day raw. 132 a day after removing old re-pooled tokens and campaign coins.
2. **Base rates.** Sellable $1M 5.1%, $2M 1.7%, $5M 0.19%, $20M 0.04% (within the first 6 hours of the pool). A random sample of 376 gated coins, followed for up to 20 days on GeckoTerminal hourly (every venue), found almost no extra slow wins (sellable $2M in 1.1%, all within hours). Slow wins are rarer than 1 in 300 and need more coins and more days.
3. **Shapes and families.** Quick = bundle family (30% reach $2M, middle 16 minutes) and some other hooks (1.3%). Plain pools 0.1%. Campaign skipped. Slow = rare, often on other venues: of 375 sampled coins, 7 trade mainly on Pons V2 and 8 on pools.trade, which the launch tape does not see.
4. **Signs at the gate (non-bundle coins, 2,522, target $1M because $2M had only 6).**
   - Killed: "ran far past $250k before the first real sell". Only 1% of non-bundle coins do that, so it is not usable.
   - Low sell share before the gate is better, in both halves.
   - Buyer counts and size flip between halves, so they are not proven.
5. **Money test.** Buy at the gate plus 5% slippage, sell at sellable $2M, stop at minus 50%, 3% cost.
   - Bundle family: 1.31x per trade in old weeks, 1.30x in new weeks (n 91 and 42).
   - Plain pools: 0.70x. Other hooks: 1.01x in old weeks, 0.43x in new weeks.
   - **Only the bundle quick lane makes money today, before the liquidity pull check.**

## Data source (changed 5 Oct, 16:40 Bangkok)
- Phase 2 reads the Price Predict database, read only, and never writes to it. That covers the token birth table, supply, every swap, liquidity and sync log of every pool from 15 Sep, and the token minute bars once they are built.
- The GeckoTerminal collector built earlier (/opt/phase2) is stopped. Its database is no longer used.

## What is still missing (owner and when)
- Clean minute data for all venues back to July: Price Predict (brief: http://159.223.88.168/hx-c9ois/hx/pp-datafix-brief.md).
- Wallet level data for 72 h after the gate (fresh buyers, retention, dev and sniper exits): Price Predict wallet tape, item 5 in the brief.
- Liquidity pull check for the bundle lane: next phase 2 step, using ModifyLiquidity events, the same method as the sell session.
- Enough slow winners to learn from: about 4 to 6 weeks of collector data, or sooner once the July to September backfill lands.
- First weekly scoring: 12 Oct 2026, Bangkok morning.

Version 1.2, 5 Oct 2026 16:40 Bangkok (data source = Price Predict DB read only, own collector stopped, one sign killed). Latest: http://159.223.88.168/hx-c9ois/hx/phase2-process.md