# 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.

# Asks for Price Predict from the first minute supervisor (6 Oct 2026, 08:50 Bangkok)

## Read this first
- Goal: find out what really makes money, priced the way a real wallet gets paid, before any real money.
- Order = priority. 1 to 3 are the ones that decide whether we go live.
- FIRST do the wash fix (http://159.223.88.168/hx-c9ois/brx/pp-wash-fix.md): mark a wallet as crew by who funded it first, not only by repeat trading. Asks 1, 4 and 5 below depend on a correct crew / outsider split, so rerun them only after that fix passes.
- My live data for you (refreshes every 10 min): http://159.223.88.168/hx-c9ois/fm/q7tz/lane1_trades.jsonl
  - Every pond launch (v4, fee 100, native ETH, no hook) bought on paper 10 s after the first trade, $5, sell half at 3x and the rest at 10x, priced on chain (eth_call buy+sell simulation, GMGN 1% each side and gas charged). Field `root` = your crew root.
  - First night (59 coins): 3x hit 24%, 10x hit 7%, net -25% per coin. 4 crews (0x04f2cb48fa0c60da3a6241077f1b7c549b9b9094, 0x3aef953b35654b76ae8fea2dfe36edaa72834da3, 0x1577c6f8c5a203ab00c5af9fb2c128a0bbe6db08, 0xf433ba5beddb56fce495abb54aefb891945305fb) gave 38 coins with ZERO 3x hits. The other 20 coins made about +120% per coin.

## 0. URGENT: test the second-0 launch fingerprint on all history (new, 6 Oct)
- Live last night, 82 pond coins, bought 10 s after first trade, priced on chain:
  - Token is the standard launcher token (runtime code 3855 bytes, md5 of the "0x..." code string starts 17ad80) AND the launch tx (PositionManager 0x58daec3116aae6d93017baaea7749052e8a04fa7 multicall) sent 0 ETH: 28 coins, 17 hit 3x, 5 hit 10x, +94% per coin.
  - Everything else (custom token bytecode, launch tx sends 1.6 to 2.0 ETH): 54 coins, 0 hit 3x, -100% per coin.
- Both are knowable at second 0. Please run it over every pond launch 20 Sep to today: 3x / 10x rate and $ per trade (entry 10 s after first trade, half at 3x, rest at 10x, pull = 0) for pass vs fail, by day. Also: is "pass" the same thing as your WASH crews, and does it also split the money book and L-BROAD20 picks?
- This decides go live. Please do it before everything below.

## 1. Crew recent hit rate as a feature (highest value)
- For each crew root, rolling hit rate of its last 10 and last 20 coins: share that reached 3x and 10x from the price 10 s after first trade, and median seconds from first trade to pull.
- Add it to the picker and to L-BROAD20 as a filter. Test walk forward by day (only past coins of that crew count at decision time).
- Report: $ per trade with and without the filter, 20 Sep to today.

## 2. Is L-BROAD20 luck? (selection bias test)
- It is the best of ~1,500 ideas. Rerun the whole search with a nested walk forward: choose the winner using only days before D, test it on day D, roll D over every day you have. Report the chained out of sample result.
- Also a reality check: how many of the 1,500 ideas look this good on 3 and 4 Oct by chance? (shuffle the coin outcomes across coins of the same day, 200 times, and count.)
- Only the chained result counts for going live.

## 3. Paper vs real fills
- Publish every lab book's picks live as jsonl, same fields as picks.jsonl (model, token, pool, t_pick, venue), e.g. /hx-c9ois/fm/q7tz/labpicks.jsonl. I run them through the on-chain watch bot so we see the real number next to paper.
- For each paper exit, also store what a sell of that size would have returned in that block (eth_call, not last trade price). Report the average gap.
- Exit latency test: for every 10x / 15x / 25x exit in your books, would it still fill if the sell lands 1, 3 and 5 s later? (Crews pull ~2.5 s after their last trade.) Report $ per trade at each latency.

## 4. Pull timer per crew
- From the crew map: time from first trade to pull, per crew, distribution. If a crew always pulls at a fixed time, a timed sell just before it beats the 3x / 10x ladder. Test: sell all at (median pull time minus 20 s) vs the ladder.

## 5. Size on crew coins
- Your new finding (big pond crews pull after only 2 to 3% of outside buys of $270 or more) is good news. Extend it: pull rate within 60 s after an outside buy of $20 / $50 / $100 / $270, per crew. That sets the size we can use.

## 6. Housekeeping
- Paper books: hook family kill plus buy/sell simulation before any win is booked (pp-check.md, confirm done).
- Hourly model retrain on the new server (Ben's call, 6 Oct), automated, with a stale check that retrains by itself. Guard: a new model goes live only if it is not worse than the current one on the last 24 h walk forward; otherwise keep the old one. Retrain must pause for the live pickers so it never slows a pick.
- Finish 0x956c7a0d50b164cf8442d1ec983724e0383f5c7a (real crowd or crew?). Its pool (fee 0.05%, tick spacing 1, no hook) is pulled.

## What I do with the answers
- If 1 or 4 gives a crew rule that stays positive walk forward, and 3 shows the real gap is small: Ben puts in $5 a coin, live, 30 coins, then $20.
- If not: the first minute lane stays paper and we look elsewhere.

Version 4, 6 Oct 2026 11:00 Bangkok (new item 0: launch fingerprint test). Latest: http://159.223.88.168/hx-c9ois/brx/pp-asks.md