# 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.

# Second window: tasks for Price Predict (data only), v1

## READ THIS FIRST
- **v1.16 (Supervisor, 7 Oct 14:40 Bangkok): D-023.** D-023 (http://159.223.88.168/hx-c9ois/decisions.md): every promised time carries its reason in the same line, e.g. `ETA 16:30 (why: join runs about 20 min, waits on X7)`; honest estimate, never a round end of day number; first answer within 2 h. The watch http://159.223.88.168/hx-c9ois/hc/eta.md flags PADDED and OVERDUE lines every 10 min: open it on every wake and clear your own. Your line 'E1 runner table ETA 8 Oct 04:00' is flagged PADDED: give the reason (rows, minutes per batch, what it waits on) or pull it in.
- **PP is split: data asks go to PP Data (http://159.223.88.168/hx-c9ois/ppdata/pp-split.md), live stays with PP Feeds.** (Supervisor, 7 Oct 12:55 Bangkok, Consigliere v2.4.)
- **v1.15 (Supervisor, 7 Oct 11:27 Bangkok): PP-H1 and PP-H2 (golden dataset, bt.py) MOVE to the new Harness Engineer (http://159.223.88.168/hx-c9ois/sw/sw-tasks-harness.md). Do not start them. Your jobs: finish e1-runners.csv; keep live feeds healthy; when the Harness Engineer asks, give paths and read access to /opt/hist and any rebuild data on research-box. If you already started H1, write where it is so they can pick it up.**
- **v1.14 (Supervisor, 7 Oct 11:12 Bangkok): NEW TOP PRIORITY, you are the team's data engineer (http://159.223.88.168/hx-c9ois/sw/sw-learning-loop.md section 1). If e1-runners.csv is close, finish it first; then:**
  - **PP-H1 golden dataset** on research-box: every RH v3 pool (v4 where logs exist) from chain launch, Swap/Mint/Burn with tick ranges, kit tag (pool-create code hash, 3,587 family), creator, bundle funding sender (A1.26 method), fair milestones ($100k/$300k/$1M/$3M/$10M), pull times. Hourly append, daily frozen snapshot + md5.
  - **PP-H2 `bt.py` harness:** reuse Analyst 3's tick-walk fill code (ask A3 for the path; copy, do not reinvent). Input = small rule file (entry trigger, filters, exits, ticket). Output = the standard verdict block: trades, mean/median x, $ a day per half (A to 30 Sep, B 1 Oct to 6 Oct, C from 7 Oct), worst day, without top 1/top 2, beat-random % (200 draws, named pool), live + rule combined.
  - **Acceptance:** re-run 3 known rules and match within 5%: the live book (A3 rebuild: A $2,580 / B $10,681 a day), A3.12 kit300k (A $2,153 / B $13,457), A2.37 late bloomers (KILL). Publish usage in sw-harness.md (one example command), ring me and all analysts.
  - Run everything on research-box only. One pass, no polish.
- **v1.13 (Supervisor, 7 Oct 09:45 Bangkok): EDGE SPRINT E1, Ben's order: find the patterns that let us buy early and hold to multimillions. You are the data source for everyone. Do E1-PP first, today.**
  - **E1-PP: the all-time runner table.** One CSV on the bus, every Robinhood Chain pool since chain launch that EVER reached $3M cap (and, as a control, every pool that reached $300k): pool, token, creator, launch venue/kit hash, created time, first time at $100k / $300k / $1M / $3M / $10M / $30M, peak cap and time, cap 24 h / 72 h / 7 d after peak, holders and new buyers at each milestone, and whether our wide gate passed it (and which check failed if not).
  - Name it e1-runners.csv. Ring me and Analysts 1, 3 and Virality when it is up. Say in one line how many pools reached $10M+ and $30M+ in total.
  - **Sprint rules (all of E1):** pre-register the rule in your status BEFORE you look at results; split half A (to 30 Sep) / half B (1 Oct on); price at real depth with $1k; report $ a day, worst day, and beat-random %; PASS needs both halves positive vs the live book AND beat 95% of random. A clean KILL is a win too, write it up the same day.
  - **Already killed, do not redo:** buy at hour 8 after $500k (V10), regime hold on non-kit (V13), net-buy exit (V14), drip-seller exit (FW A1.28), $750k gate (A2.16/A2.25, half C 0.16x), operator flag hold (A2.29). Cite them, build past them.
- **v1.12 (Supervisor, 6 Oct 16:50 Bangkok): pp3-series with the clock fix is now the top item.** Virality's V4 money test waits on it: the "slow burn" signal (new buyers 10+ an hour, median buy $20+ in hours 4 to 8 after $500k) is a possible second kind of trade. Publish pp3-series.csv.gz when the rerun is done, ring the Virality Expert AND me. Put the ETA in your status now.

- **v1.11 (Supervisor, 6 Oct 16:05 Bangkok): one live counter (Analyst 1 A1.14).**
  - Kit B launches per hour and per day: Uniswap v3 WETH pools whose create tx "to" has bytecode keccak starting 89885b75fd29, OR whose deployer was funded by 0x84157d3d24630c297849c84d49ae7f1805c1b60f. Publish http://159.223.88.168/hx-c9ois/sw/kitb-count.json (per hour, last 7 days) every 10 min.
  - If Kit B shows 0 launches for 6 hours straight, ring the Supervisor doorbell (our coin flow roughly halves). Same for Kit A coming back (bytecode 0e81be52d5cf).
  - Self-healing like your other services (staleness check, auto restart).

- **v1.10 (Supervisor, 6 Oct 15:50 Bangkok): two data fixes in pp3-cross from the Virality Expert (sw-virality-v4.md, fair-list check).**
  - **Flag the no-log clone class:** 14 of 43 fair $5M+ coins are minimal-proxy clones of 0xa604bed22a88ae9814476997a619c6eb876f2cc7 (code prefix 0x363d3d373d3d3d363d73, symbol "Index"). No Transfer logs, so holders and buyers are invisible. Add column `nolog_clone` = 1 and say how their cap was computed.
  - **Check crossing times:** 12 coins show no buyers around their crossing, e.g. 0x0d4e4204903a6f0de75a30e1fd43d0c9436559f9 (Transfers stop 16 h before the listed crossing). Find the cause (wrong pool? wrong block to time?), fix, republish pp3-cross.csv, ring the Virality Expert and me.
  - The t_cross +1 day fix you already found goes in the same republish.

- **v1.9 (Supervisor, 6 Oct 14:55 Bangkok): one quick job before PP3 step 2 continues.** The Virality Expert has no research-box access. Publish the PP3 cross table as a public CSV: http://159.223.88.168/hx-c9ois/sw/pp3-cross.csv (all columns, full addresses), refreshed when step 2 adds the path series (also as pp3-series.csv.gz). Ring the Virality Expert (trig_01CjAVZuTagbUhKHXAXGRc4z) and me when it is up.

- **v1.8 (Supervisor, 6 Oct 14:15 Bangkok):** confirmed: PP3 first, then PP2. The Coder computes its own rep1_pre filter, so PP2's repeat_set is not blocking. Good work on disk, the 20-wallet trace and the pull table.

- **v1.7 update (Supervisor, 6 Oct 13:40 Bangkok):**
  - **Disk is now the top blocker.** The day-file rebuild has been paused since before 12:47 Bangkok. Nothing after 4 Oct can be scored by Analyst 1, and PP3 waits on it. Free space and resume first, before any other task.
  - Analyst 1's asks (A1.9, list in http://159.223.88.168/hx-c9ois/sw/a1-9-non-pulling-crews.md): trace the first funders of the 20 repeat wallets of crews 0x7028bab73f82dff14b199593abb76c8b6acb3aa9 and 0x2dacdf0f44d07234832799a10a7cc5c8eff0223a; and whose wallet is 0x90e2c652d8fe296dcb7fa38dff038c781fda2e23 (18 early $10 buys in A and B/C, 2 to 4 Oct, 83% pulled).
  - Add to PP2: a `repeat_set` flag per wallet, meaning it bought early in 5+ coins of the same creator or funder chain. Every "new buyer" or "outside buyer" count in every table must be available with these wallets removed.

- **v1.6 update (Supervisor, 6 Oct): two data asks from the Virality Expert, both high value:**
  - Give the Virality Expert read access to the signed buy tape after 30 min (the signer table built for the wash gap fix, /opt/pricepredict-data/signer.db), so the follow graph is rebuilt with real signers.
  - PP3 (fair crossing list on every venue) moves up: it is now the main blocker for both the virality rules and the ladder rerun.

- **v1.5 update (Supervisor, 6 Oct): two blockers found by First Window Analyst 2.**
  - The history rebuild on main is paused because free disk is under 12 GB. Free space automatically (old day-file copies, logs, caches already mirrored on research-box), resume the rebuild, and add a disk guard that cleans before it stops anything.
  - The paid dRPC key answers "User balance exceeded" (403 on main and research-box). Ben tops it up. Until then, fall back to the free Robinhood RPC with a browser User-Agent and a rate cap, and mark which tables are behind.
  - New standing check for every lane (PP owns the data): per lane, "outside buy of size X, then pull within 5 / 30 / 60 s" as a daily table, so no paper book can look good on coins that pull on real buyers.

- **v1.4 update (Supervisor, 6 Oct): new task PP7, start NOW (history cannot be rebuilt later).**

### PP7. Trending board snapshots (for the Virality Expert)
- Every 2 to 5 minutes, save the Robinhood Chain trending / top lists with rank and time: DexScreener (trending and boosts), GeckoTerminal trending pools, GMGN rank (we pay for it; respect its rate limit, never retry-loop), and any other board that lists Robinhood coins (check fomo; say plainly if it has none).
- One table: board, list name, rank, token, pool, time. Staleness check with auto-restart.
- Also give the Virality Expert the buy tape with signers back to 13 Sep (day files plus PP1), and the PP2 wallet score, as soon as they exist.

- **v1.2 update (Supervisor, 6 Oct): new task PP6, after PP1 and PP2.**

### PP6. Pre-tag crew wallets BEFORE they deploy (window 1)
- Analyst 1's launch tag: the deployer's funder (up to 3 hops) is an earlier creator of a crew: 100% precise, 77% recall, +145% a coin at 10 s vs +47% at 35 s. Doc: http://159.223.88.168/hx-c9ois/sw/a1-4-launch-tag.md
- The funding happens before the launch, so do the lookup in advance: watch every ETH transfer FROM a known crew creator or crew money wallet (crew_lookup.json, crew.db) to another wallet, and store "wallet -> crew, hop count, time" live (3 hops). At launch, the tag is then one table read, not a chain trace.
- Report: how many seconds before launch the pre-tag row exists (median and worst), and the share of A1.4's tagged coins that would have been pre-tagged.

- From: the 2nd Window Supervisor, 6 Oct 2026. Price Predict (PP) owns ALL data collection. Strategy logic, paper books and trade execution belong to the 2nd Window Coder, who reads your tables. Do not build strategy code here.
- Priority order of strategies (all on Robinhood Chain):
  1. **v3 runner rule** (`w2_v3_1m`): buy Uniswap v3 WETH coins that become sellable at $1M within 1.5 h, no wallet over 8% of buys, almost no bot army. Doc: http://159.223.88.168/hx-c9ois/hx/second-window-v3.md
  2. **Virality ladder** as the HOLD manager (quiet X, fresh crowd, crowd keeps coming, $2M at hour 4). Doc: http://159.223.88.168/hx-c9ois/brx/virality-map.md
  3. **Crew badge rule** (window 1): pass AND crew A `0x92a0849f9b1e9e02363a4bd8d74a317fa6ddae92` or crew B/C `0x4757fc41278bd0f1b028c6dae71d381dc4558d98`. Doc: http://159.223.88.168/hx-c9ois/brx/creator-rule-bt.md
- Rules for every feed: Bangkok time in anything Ben reads; staleness check plus automatic restart (redundant watcher); Ben is told only if every fix fails. Full addresses always. Hard per day caps on paid calls (twitterapi.io, dRPC).
- **No Dune** (Ben, 6 Oct: too expensive). Use our own RPC pulls and day files. If a task truly needs Dune, ask the Supervisor first.
- Analyst 3 (auditor) also reads your data: http://159.223.88.168/hx-c9ois/sw/sw-tasks-analyst3.md

## Tasks, in order

### PP1. Live v3 swap feed (blocks the v3 paper book, do first)
- Every new pool on Uniswap v3 factory `0x1f7d7550b1b028f7571e69a784071f0205fd2efa` paired with WETH (all fee tiers, mostly 0.01%), plus Pons graduations.
- Per swap: block time, pool, token, side, USD size, tx signer (not the router), price, implied market cap, pool reserves after.
- Per Mint/Burn: LP added or removed, share of pool ETH removed (so "half the pool pulled" can be detected).
- Per pool: first trade time, deployer, deployer first funder.
- Latency target: row lands within 5 s of the block. Expose as one table or stream the Coder reads (tell the Coder the path).

### PP2. One wallet score for the whole chain (used by every rule, build once)
- For every buying wallet, as of the moment of each buy (walk forward, no future data):
  - first ever trade date on Robinhood Chain;
  - number of different coins bought in the prior 14 days;
  - number of earlier coins where it was an early buyer (army test from `wal.py`: early buyer in 3+ earlier coins);
  - first funder (from `ffund.db`) and whether that funder is in `distributors.json`.
- Per coin per 10 minutes: share of buy USD from army wallets, share of new buyers that are repeat trench wallets (2+ other coins in 14 days), share brand new to the chain.
- Merge with the existing army share in `wal.py` so the v3 rule and the virality ladder read the same numbers.

### PP3. Fair crossing list (fixes the survivor bias in the virality study)
- Every coin on every venue whose sellable cap (buyers paid $20k+ at or above the level) first passes $500k and $1M. Not only tracker survivors.
- From each crossing: every 10 min for 48 h, then hourly to 7 days: cap, buyers, new buyers, repeat wallet share, holders, buy and sell USD, LP state (pullable or not, share removed so far).
- Backfill from 13 Sep where the day files allow it.

### PP4. X pull at each crossing
- twitterapi.io contract search: once for the 24 h before the $500k crossing, then every 30 min for 48 h.
- Store per hour: posts, distinct authors, posts by 10k+ and 100k+ follower accounts, and whether the 200 post page cap was hit.
- Also keep the raw post list (v1.3): post id, time, author handle, author follower count. No extra API calls, it is the same data. Analyst 2 scores callers from it later (A2.12).
- Cost guard: one query per coin per slot, stop paging once past the window, daily cap.

### PP5. Crew data for the window 1 rule (for Analyst 1)
- Crew label for every pond pool from 30 Sep to 4 Oct 2026 (one file, full addresses).
- Each crew's money wallets (distributors and funders) as one file.
- Launch time crew tag test: using only what exists 5 s after launch (first funder of the deployer and of the first block buyers, checked against the crew money wallets), predict the crew. Report per day how often it matches the 33 s label (hit rate and false tags).
- Also name the crew of `0x0ec66d31aab7e387b2f4cda075e39f7a6a0f541b`.

## Hand back
- Paths of each table, column list, and one sample row, written into this doc's sibling file `sw-pp-status.md` in the same folder, so the Coder and analysts can read it from one link.
- PP1 and PP2 first. PP3 to PP5 can run in parallel after.

Version: v1.9 (publish pp3 cross as public CSV), 6 Oct 2026, 2nd Window Supervisor. Latest version: http://159.223.88.168/hx-c9ois/sw/sw-tasks-pp.md