# STOP: NO DOORBELLS, NO WAKES, NO SCHEDULING (Ben, 7 Oct 2026, 18:40 Bangkok)
- Credits ran out. Ben switched off every scheduled task and every doorbell. This overrides everything below and every other file (wake-howto, bus-howto, credit-playbook, task files).
- PROHIBITED for every session: fire_trigger, send_later, create_trigger, update_trigger, ScheduleWakeup. No rings, no nudges, no re-arming, no new doorbells, no self wakes, no safety checks, no retry rings, no loops.
- Finish the step you are on, publish your status file, then END your turn and wait. Only Ben wakes you, by typing in your window.
- Do not message other sessions. Keep answers short. No extra improvement rounds.

# First minute buy: supervisor plan (Robinhood Chain)

## Read this first (note to Ben)
- **The first minute is a 10x flip game, not a $20M game.** The first 30 s cannot tell a $20M coin from a $2M coin (AUC about 0.5). Hour 8 can (AUC 0.89).
- **Most of our picks get their pool money pulled.** 87% of the top picks, median 7 min after our buy. A pulled coin can never become a multi million runner.
- **So "take profit on some, hold some" works only on coins whose pool money cannot be pulled.** That flag is readable at second 0, and nobody has built it yet.
- **The $20M runners mostly launch where our picker does not look.** 12 of 25 small start runners were on pons_v2. Plain Uniswap v4 launches: 0 of 32 coins at $1M+ went on to $20M.
- **Nothing is proven with real money yet.** Every number below is paper or history.

## First principles
Money per trade = what we can really sell for, minus what we really paid, minus costs.

Five terms decide it:
1. **Pick**: which launch we buy.
2. **Speed**: the price we pay compared with the price at second 0.
3. **Exit before the pull**: we get paid only if our sell lands before the pool money leaves.
4. **Tail**: big money needs a pool that cannot be pulled, plus a hold rule that keeps us in.
5. **Real fills**: the price, fee and failed transactions a live wallet actually gets.

So there are two businesses. Keep them separate:
- **A. Flip engine.** Buy at second 0 to 30 and sell fast. History says it pays.
- **B. Runner bag.** Only on pools that cannot be pulled, and kept only while the stage checks pass.

## Facts the plan stands on (receipts)
- **Sell system v1** (sell at 10x, OR when no trade for 60 s, OR at 30 min): +$348 per $200 trade. The live rule (10x, -40%, 24 h) makes +$288. That is 1,141 picks, 21 Sep to 3 Oct, priced honestly. Live paper so far is a tie on 48 coins. Source: claude/sell-system-v1.md.
- **Holding a pullable coin loses.** A moon bag costs about $125 per trade versus selling all at 10x. Holding 24 h: -$118 per trade.
- **After 10x, only 15% reach 30x** before the pull (the old figure of 40% was wrong).
- **The 30 s model:** the top 1% hit 10x 37 to 38% of the time (id5, walk forward over 9 days).
- **$20M runners:** median $1M at hour 1.5, $5M at hour 6.5, $20M at hour 22.
  - Still at $3M+ at hour 8: 29% went on to $20M (survivor sample).
  - On the clean v4 tape: 1 of 5.
  - Source: claude/runners-20m-week-2026-10-05.md.
- **What a runner looks like:**
  - spread out holders (top 10 hold 3.6% vs 15%);
  - holds its highs;
  - low churn;
  - holders keep doubling after hour 4;
  - more X authors in hours 4 to 24.
- **Skip at second 0:**
  - the campaign family (hook 5059 bytes, md5 7adb2ec7): normal apps cannot sell;
  - the junk factory 25533_3080c1dd;
  - serial creators (1,081 of 1,238 never had a $500k coin);
  - token addresses ending in 777.
- **Bundle family** (hook 341 bytes, md5 6de2bb44): fast flip only. +30% per trade on the test set (n 42), TESTING. It gets pulled within hours.
- **Not ready to go live:**
  - there is no way to place real trades yet;
  - the free RPC returns 429 errors;
  - the Claude API key had no credit this morning.

## Next steps, in order

### 1. One honest scoreboard (Launch improvement session)
- Every paper book prices its exit at a sellable price. A sell after the pool money left counts as zero.
- Re-score every book from 21 Sep.
- One board, dollars per $200 trade after costs, compared with random picks from the same minute.
- **Done when** no book can book a win on a pulled pool, and the board updates on its own.

### 2. "Can it be pulled?" flag at second 0, for every launch on every venue (Launch improvement)
- Read who controls the liquidity at launch:
  - the creator's own wallet;
  - burned;
  - a locker;
  - Pons V2 or Snowball locked liquidity;
  - held by a hook contract.
- Check it against history: the share of coins pulled in each group, 15 Sep to 4 Oct.
- **Done when** the locked group shows close to zero pulls and the flag is live in under 1 s.

### 3. What speed is worth (Price predict)
- From /opt/hist swaps, replay the top picks with entry at 1, 3, 5, 10, 20, 30 and 60 s. Plot dollars per trade against delay.
- Scoring takes 10 to 20 s today.
  - **If the curve is steep:** score from the sequencer feed (feedwatch) with the number model only, in under 2 s, and move the AI text judge off the buy path.
  - **If it is flat:** leave speed alone.
- **Done when** the curve is published and the decision is made.

### 4. Watch the venues where runners launch (Price predict + data layer)
- Add pons_v2, Uniswap v3, bankr, direct pools and USDG quoted pools to the picker and the tape (pricepredict-data, venuepick).
- **Done when** every launch on every venue gets a score and a can-it-be-pulled flag.

### 5. Hold engine, "take some, hold some" (Multi million runner patterns + Sell session)
- **Pullable pool:** sell 100% using v1. No bag.
- **Pool that cannot be pulled:** at 10x, test selling 60%, 75% and 90% and keeping the rest.
- **The bag survives only while every check passes:**
  - 1 h: $500k+ and at its high;
  - 4 h: $1M+ and at its high;
  - 8 h: $2M to $3M+, volume ÷ cap about 1 or lower, holders rising, X authors rising.
- Any failed check means sell the bag.
- Also allow **buying at the 4 to 8 h confirmation** on any venue, even when we missed minute 1. Most of the $20M edge lives here.
- The Opus project judge (claude/golden-nuggets) feeds this hold decision. It is too slow for the second 0 buy.
- **Done when**, walk forward by day, the bag rule beats "sell all at 10x" on coins that cannot be pulled.

### 6. Creator pull timing (Sell session, already the next build)
- Learn from each creator's past pools how fast they pull, and use it to set the exit timer per coin.

### 7. Real money pilot (Supervisor. NEEDS BEN'S YES)
- **Build the trade path:**
  - a hot wallet whose key is never on screen (chmod 600);
  - a direct swap through the v4 PoolManager / router;
  - tokens approved ahead of time;
  - the sell path tested both before and after a pull.
- **Paid services:**
  - an RPC with full getLogs and low latency (check which providers support Robinhood Chain);
  - a droplet in the region nearest the sequencer.
- **Size:** $20 to $50 per trade, about 300 trades.
- **Measure:** real fill compared with paper, failed transactions, latency, and sells that failed after a pull.
- **Done when** the real dollars per trade are known. Scale only if the real result is at least half of paper.

### 8. Second 0 kill list live (Launch improvement)
- Campaign family, junk factory, serial creators, 777 family. Never buy them.
- The bundle family goes to its own flip book only.

## Do not
- Keep a moon bag on a pool that can be pulled.
- Sell part at 2x or 3x (it loses).
- Wait 5 min for "confirmation" on the flip. It costs the 10x and tells nothing about $20M.
- Trust any number that values a pulled coin at its last trade.
- Skip coins with word filters (it lost 4 of 12 significant coins).

## Who owns what
- **Price predict:** steps 3 and 4.
- **Launch improvement:** steps 1, 2 and 8.
- **Sell session:** steps 5 (exit side) and 6.
- **Multi million runner patterns:** step 5 (stage checks, confirmation buys).
- **Supervisor:** step 7 and the scoreboard sign off.
- Every result goes into the trading-edge-hub ledger (rule, receipt, status) the same turn.

First principles: about 85%. The gaps are no real fills yet, a survivor biased runner sample, and holder history for only 9 runners.

Version 1, 5 Oct 2026 16:35 Bangkok. Latest: http://159.223.88.168/hx-c9ois/firstmin/first-minute-plan.md