# 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 Analyst 1 (crew badge / window 1), v1

## READ THIS FIRST
- **v1.24 (Supervisor, 7 Oct 14:50 Bangkok): scorecard published at 14:37, ahead of 16:30, good.** Fix one line: the method says 'frozen 7 Oct 2026 15:30 Bangkok', which is in the future; put the real freeze time. D-024 (http://159.223.88.168/hx-c9ois/decisions.md): every analyst always has one ranked edge hunt queued. Pre-register before any result, script on research-box (nice, flock), first answer within 2 h with the reason for the time (D-023), result as a ledger row in sw-registry.csv, every finished line ends with `cost: credits, API $, hours`. On every wake: fresh doorbell (send_later 2027-06-01) on LINE 1 of your status (D-022).
  - **Your queued hunt after the 18:00 page, A1.30 v4 POOLS:** the live gate sees only Uniswap v3 WETH pools. FW Analyst 3 is building the uncut universe (every v3 AND v4 Robinhood Chain pool from its true first swap, lab.py format). When it lands, run the live gate (w2_v3_wide_s03 + kit hold) on v4 pools exactly as on v3: $ a day both halves, worst day, beat 95% random, and how many of the scorecard's missed $2M+ runners were v4. Until it lands: count the v4 pools and $2M+ v4 runners since 1 Oct from the chain (PoolManager Initialize events) so we know the size of the prize. First count within 1.5 h of starting (why: one log scan plus a cap check per pool).
- **v1.23 (Supervisor, 7 Oct 14:40 Bangkok): first scorecard answer by 16:30, not 18:00** (Consigliere v5.6; why: your method is frozen and (a) is a join of w2.db against FW's live and paper buys, about 1 h of work). 16:30 = (a) and (c) with numbers, (b) with the runner list as far as done; the full page stays daily at 18:00. Pre-registration of (b) accepted as written 14:30.
  - (c) source: http://159.223.88.168/hx-c9ois/sw/sw-live-trades.csv (Exec X6, live now).
  - D-023 (http://159.223.88.168/hx-c9ois/decisions.md): every promised time carries its reason in the same line, e.g. `ETA 16:30 (why: join runs about 20 min, waits on X7)`; honest estimate, never a round end of day number; first answer within 2 h. The watch http://159.223.88.168/hx-c9ois/hc/eta.md flags PADDED and OVERDUE lines every 10 min: open it on every wake and clear your own. Your 18:00 line is flagged PADDED on eta.md: add the reason.
- **v1.22 (Supervisor, 7 Oct 14:25 Bangkok). NEW JOB: the D-020 SW scorecard. SW is killed as a team on 14 Oct 18:00 unless it shows edge on coins FW did NOT take (http://159.223.88.168/hx-c9ois/decisions.md, D-020). You own the proof.**
  - **A1.29 Scorecard, first issue 18:00 Bangkok today, then daily 18:00.** Publish http://159.223.88.168/hx-c9ois/sw/sw-scorecard.md (+ .csv). Three numbers, each for today, last 7 days, and since 1 Oct:
    - (a) **$ a day from coins FW did NOT buy:** every SW paper and live trade (w2.db, live book w2_v3_wide_s03 and its paper twin) on a token FW (fm_mc5 live, wallet 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286, and FW paper books) never bought. FW trades: read FW's published fills/ledger pages or the on-chain buys of that wallet; never ssh to firstmin. Show also the SW $ on coins FW DID take, separately.
    - (b) **$2M+ runner catch share:** every pool on our factory whose cap reached $2M. Pre-register now, before you look: T0 = true first swap; halfway = first time the cap reached half of the pool's peak cap; caught = an SW entry (paper or live) after T0 + 60 s AND before halfway. Share = caught / all $2M+ runners. List each runner with full addresses, peak cap, SW entry time and cap, or 'missed' + the gate that blocked it.
    - (c) **SW live x vs its paper twin**, per trade and median (use the Exec's per trade file sw/sw-live-trades.csv from 15:00).
  - Freeze the method in the page's first block; any later change is a new version, never a silent edit. Lead (paper book 'kit at fair $300k', A1.27) is tracked as a fourth row, labeled pre-registered, not a pass.
  - Research runs in your own workspace; one small cron on main (nice 19, read only) may publish the page.
  - **Wake rule changed (D-022):** fire_trigger is retired. On every wake: make a fresh doorbell (send_later 2027-06-01, name `Doorbell: Analyst 1`), put `DOORBELL: trig_... (armed, Bangkok time)` as LINE 1 of sw-analyst1-status.md. To wake me: update your status, then update_trigger my DOORBELL id (line 1 of sw-supervisor-status.md) with run_once_at = now + 2 min. How: http://159.223.88.168/hx-c9ois/wake-howto.md
  - Every finished line ends with `cost: credits (rough), API $, hours` (D-015).
- **v1.21 (Supervisor, 7 Oct 10:12 Bangkok): A1.25 ACCEPTED. Best map of a $10M coin so far: kit, fast, quiet, fresh buyers. Rule kill accepted (it re-finds kits). Half C fresh-buyer kit hold pre-registration accepted as written.**
  - **A1.26 THE FRESH WALLETS (biggest open lead).** Winners' pre-$300k buyers are brand new wallets (median nonce 0 to 1). First principles: a wallet with nonce 0 was funded minutes before it bought, so someone sent it ETH. Find who.
    - (a) For every pre-$300k buyer of the 27 winners and 89 dead coins: the funding tx (first incoming ETH), its sender, minutes from funding to buy. Group senders: same key funding many buyers in one coin = one hand (operator or kit bundle); exchange or bridge hot wallet = real new people. Label exchange/bridge senders by eth_getCode + known hot wallets; show the top 20 senders with full addresses.
    - (b) Does "fresh buyers funded by ONE hand" vs "fresh buyers from exchanges or bridges" split $10M winners from dead coins? Lift per half, as in A1.25.
  - **A1.27 BUY EARLY ON THE MAP (Ben's ask: buy early, hold to multimillions).** Pre-register on half A only, then price: entry at the fair $300k milestone (not $1M) for coins that are (kit) AND (pre-$300k buyer median nonce at most 1), plus at most 1 extra feature from A1.26 if it lifts in half A. Exits = live book (0.3x stop, 5x, kit hold 50% trail, pull exit, 12 h time), $1k, real in-range depth, every A2.21 friction. Both halves, $ a day, worst day, beat 95% of 200 random draws from kit coins at their $300k milestone. Report how many of the 27 winners it catches and at what entry cap, and overlap with the live $1M gate (stand-alone book and added to live). A2.26 early entry and A1.15 kit early are killed: cite them and say what is different (the fresh-buyer gate).
  - **A1.28 cheap side check:** in half B 0 of 11 non-kit $1M coins reached $10M. Does dropping NON-kit coins from the live gate raise $ a day in both halves (they still earn at 5x)? One table, no new rule unless it passes both halves.
  - Output http://159.223.88.168/hx-c9ois/sw/a1-26-funders.md and a1-27-early-map.md, ring me after each.
- **v1.20 (Supervisor, 7 Oct 09:45 Bangkok): A1.24 ACCEPTED (engine matches replay, good). EDGE SPRINT E1, Ben's order: find how to buy early and hold to multimillions.**
  - **A1.25 MULTIMILLION ANATOMY** on PP's e1-runners.csv (if it is late, build it yourself from /opt/hist). For every coin that reached $10M+: the path ($100k, $300k, $1M, $3M, $10M times), who bought before $300k (wallet, funder, crew badge), creator history (did this creator or funder launch a winner before), kit hash, pull or no pull. Then the SAME columns for coins that reached $1M but died under $3M.
  - Question to answer in one line: **what can we see at $300k or $1M that a $10M coin has and a dead $1M coin does not?** Rank the top 5 separators by lift, each with counts. Then pre-register the best one as an entry or hold rule and price it (rules below).
  - Output http://159.223.88.168/hx-c9ois/sw/a1-25-anatomy.md, ring me.
  - **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.
- **v1.19 (Supervisor, 7 Oct 09:10 Bangkok): A1.24 TRADE COUNT CHECK. URGENT, small.** Ben asked: only 1 live trade so far, is that expected?
  - History (a1-22-per-day.csv): 21 Sep to 5 Oct = 68 trades in 15 days, about 4.5 a day (last 7 days about 4.3). 6 Oct = 1 trade, 7 Oct so far = 1 (CrawlScan). 6 Oct alone is a 1-in-16 low day at 4.5 a day.
  - The source changed on 6 Oct: rows before are A2.21 replay trades, rows from 6 Oct are the live engine's paper books in w2.db. **Test whether the engine passes FEWER coins than the replay rule on the same coins.**
  - Do: run the A2.21 replay gate (same rule as w2_v3_wide_s03: $1M at $20k sellable, age <= 3 h, top receiver <= 0.08, army <= 0.015, sell check) over 5 Oct 00:00 to now. List every coin that passes in the replay and say for each whether the engine (w2.db) also passed it, and if not, which check failed or whether the engine never saw the pool.
  - Answer in one line first: "engine matches replay" or "engine misses N of M, cause X". If it misses, ring me and the Coder at once.
  - Also give the count of $1M-cap coins per day that FAILED a filter, by filter, for 29 Sep to now, so we see if the market changed or our filter did.
- **v1.18 (Supervisor, 6 Oct 20:50 Bangkok): A1.22 ACCEPTED. Clear answer: flat, not fading; new buyers rising. Good.**
  - **A1.23: make it an alarm, not a one-off.** Put your per-day build on the droplet as a daily job (00:15 Bangkok) that appends yesterday's row to http://159.223.88.168/hx-c9ois/sw/a1-22-per-day.csv and writes one line to your status.
  - Alarm line (write it to http://159.223.88.168/hx-c9ois/sw/w2-live-events.md through the Coder's normal path, and ring me) only if ONE of these is true: the 7-day rolling all-out $ a day is under $2,000 for 3 days running; OR $1M gates per day 7-day mean is under 4; OR new buyers per trade 7-day median is under 500. Fix the thresholds now, no changes later.
  - Add a column for LIVE trades (from the live trade table) next to paper, so the same row shows live vs paper per day. Ring me when the job has run once on today's data.
- **v1.17 (Supervisor, 6 Oct 20:40 Bangkok): A1.21 ACCEPTED (operator tag is now the KIT DRIFT alarm). Live is ON at $10 a trade since 19:25.**
  - **A1.22: is the edge fading? Check before we put more money in.** PP republished a 7 day series with buyers: http://159.223.88.168/hx-c9ois/sw/pp3-series.csv.gz (md5 ca3d7990a996b3e673b11c7de4673e30, 526 pools). Plus your own rebuild of the 23 days.
  - Per Bangkok day, for the live rule (w2_v3_wide_s03 entry, w2_v3_kit_hold exit, $1k real depth, A2.21 frictions): gates, trades, mean x, share reaching 5x, $ a day, and the count of NEW buyer wallets in the 2 h after the gate (median per trade).
  - Fit a simple line on $ a day and on new buyers per trade across days. Say plainly: flat, rising or falling, and how many days until $ a day would cross $2,000 down if the slope holds. Also: did Kit A/B launch count per day fall?
  - Output: http://159.223.88.168/hx-c9ois/sw/a1-22-edge-decay.md with the per day table. Ring me. If falling in both the last 7 days and the 23 days, flag it in the first line.
- **v1.16 (Supervisor, 6 Oct 18:55 Bangkok): A1.20 ACCEPTED. No Kit C, clean. The 3,587-byte family (51d0e39db2ed, 2696b754c8e3, 60341fa374bf: 10 of 17 reached the gate) goes on watch; the Coder tags it.**
  - **A1.21: protect the kit edge. Kits are one operator each; if Kit A or Kit B changes its bytecode, our kit hold goes blind.** For Kit A, Kit B and the 3,587-byte family: map the operator (deployers, who funds them, shared funders, the wallets that seed the first buys, timing pattern). Then build a rule that flags a NEW code hash as "same operator" from those links alone, without the old hash.
  - Test it backward: hide Kit B's hash and see if the rule finds Kit B's first launches on 1 Oct from operator links only; same for the 3,587-byte family. Count false flags on non-kit coins.
  - Output: http://159.223.88.168/hx-c9ois/sw/a1-21-kit-operator.md and a small json of operator wallets the Coder can watch. Ring me.
- **v1.15 (Supervisor, 6 Oct 18:15 Bangkok): A1.19 ACCEPTED as a clean kill (no pull-source skip filter). Good, fast.**
  - **A1.20: find Kit C. Kits are our best exit edge (Kit A 0e81be52d5cf, Kit B 89885b75fd29 hold to bigger x).** Group every Robinhood v3 coin of the last 23 days by bytecode keccak (first 12 hex). For every kit with 20+ launches: share that reached 5x, median and 90th pct peak x after 5x, share pulled within 2 h, launches a day, and who deploys it (same deployer or funder cluster?).
  - A new kit counts if its peak after 5x beats non-kit coins by 1.5x or more in BOTH halves with 10+ coins reaching 5x in each half. Also flag any kit that is near-always a pull (skip candidate).
  - Output: http://159.223.88.168/hx-c9ois/sw/a1-20-kit-c.md, ring me. If a Kit C passes, A3 audits it before live.
- **v1.14 (Supervisor, 6 Oct 17:16 Bangkok): A1.18 ACCEPTED. No gate change. Strong work: the fee router dying on 2 Oct without hurting the rule is a useful fact, and the "fresh wallets via disperse" picture explains the edge.** Coder adds your health line.
  - **A1.19 (next): the bonus you skipped, made bigger.** The 490 instant-pull coins A3 found plus every coin our gate passed that ended in a pull within 2 h: buyer funding and route mix before the pull vs our 5x winners. If one funding source or route is over 2x more common in pulls in BOTH halves, price it as a skip filter at $1k (A2.21 frictions). Close if nothing splits.

- **v1.13 (Supervisor, 6 Oct 16:50 Bangkok): new task A1.18, the source of the money. Keep the daily paper-book line too.**
  - First principles: our edge is fresh wallets arriving after the gate (A2.15: v3 coins draw new buyers; Pons and plain v4 recycle the same trench wallets). **Where do those fresh buyers come from?** If one front end or route brings them, it could become an earlier or stronger gate.
  - For the 70 wide-list trades (http://159.223.88.168/hx-c9ois/sw/sw-a2-16-wide-trades.csv), every buy from pool creation to gate + 2 h: the tx `to` (router, aggregator, fomo router 0x039ec98a76f111092d4751365ff09dd2aec301e8, Robinhood wallet route, other), plus buyer wallet age (first ETH block) and buyer funding source (exchange, bridge, a hub, a disperse).
  - Report: (a) the share of buy ETH per route and per funding source, pre-gate vs post-gate; (b) does any share AT THE GATE split 5x trades from losers in both halves; (c) for any split, the $ a day at $1k as a gate filter, both halves, with the honest frictions from A2.21.
  - Bonus if cheap: the same route mix on the 490 instant-pull coins A3 found. Different routes = a scam tell.

- **v1.12 (Supervisor, 6 Oct 16:25 Bangkok): A1.17 ACCEPTED. No tilt ships. Clean, honest test.**
  - **Pre-registered now (no changes later):** "fast gate" = $1M gate within 60 min of pool creation gets 1.5x, slower 0.5x, equal total stake. Re-test ONLY on half C (gates from 7 Oct 00:00 Bangkok) once it has 30+ wide trades, with exactly this cut. Pass = $ a day up AND worst day no worse. Write the pre-registration as one block at the top of a1-17-size-tilt.md.
  - Until half C is ready: A2.11-style daily check of the live paper book `w2_v3_wide_s03` vs your own rebuild, one line a day in your status.

- **v1.11 (Supervisor, 6 Oct 16:25 Bangkok): A1.16 ACCEPTED. Cascades closed as a skip rule (farm coins pay more than they cost).**
  - **A1.17, size tilt hunt:** farm-funded coins did 2.40x vs 1.81x in half B but 1.79x vs 2.14x in half A (3 coins). Find any feature KNOWN AT THE GATE that splits the 70 wide trades into a better and a worse class in BOTH halves: farm_funded, kit, deployer funding size, pre-gate buyer count, minutes from launch to gate, ETH in pool at gate, gate hour, Kit B count in the last 24 h.
  - Price it as a size tilt, not a skip: 1.5x ticket on the better class, 0.5x on the worse, $1k base, real depth. Walk forward only (class rules fixed on half A, tested on half B, and the reverse). Ship only if $ a day rises in both halves AND worst day does not get worse.

- **v1.10 (Supervisor, 6 Oct 16:15 Bangkok): A1.15 ACCEPTED. Clean kill, with the reason (93% of Kit B insiders cash out before $1M, median 5.9x). The $1M gate is the edge.**
  - **A1.16, the last cascade lead:** Analyst 2 found 4 sell cascades (120 to 200 single-use wallets dumping in one block). Detail: http://159.223.88.168/hx-c9ois/sw/sw-a2-19-cascade.md . Blockscout is blocked for them; your balance binary search + funding-block trace works. For each of the 3 fresh-wallet cascades (0x2a6efa703d8f1f06da07d1960711f02b1ba66f10, 0xf5a9da666eb257c0859b47efe8c2cb6745baff94, 0xfe0f309cee96912d9926b8f0786fe3347364e678), trace the first ETH funder of 20 of the seller wallets.
  - If one funder (or one disperse tx) funds them, that is a live signal: the funding tx is visible BEFORE their buys. Then count how often that funder appears in other wide-rule coins, and what those coins did. If no common funder, say so and we close cascades for good.

- **v1.9 (Supervisor, 6 Oct 16:05 Bangkok): A1.14 ACCEPTED. Best map yet. rep1_pre skip killed (costs $302 a day), as you said.**
  - **A1.15, the big one: buy Kit B EARLY.** Kit B launches reach the $1M gate 20% of the time vs 0.6% for everything else, and the funder `0x84157d3d24630c297849c84d49ae7f1805c1b60f` sends ETH 1 to 2 s BEFORE each launch. So we can know a coin is Kit B at block zero.
    - Price, walk forward, both halves, real depth, $ a day at $100, $500 and $1k: buy every Kit B launch at (a) first outside swap + 5 s, (b) first time cap passes $250k, (c) first time $500k sellable. Exits: same as live book (5x, 0.3x stop, pull, 12 h) AND a variant exiting at our normal $1M gate price + 5x.
    - Watch the dev bundle: the 0.01 ETH wallets hold 26% to 38% of pre-gate ETH. Measure how often they dump before $1M and what that does to an early buyer.
    - Also Kit A the same way (stopped now, but it is the out-of-sample half).
    - If any cell beats the wide rule's $/day in BOTH halves with no worse worst day, ring me at once. Coordinate with the first window: their mc5 rule buys v3 WETH coins at first swap + 5 min; check overlap and report it.

- **v1.8 (Supervisor, 6 Oct 15:35 Bangkok): A1.13 ACCEPTED. Clear answer: v3 is the money, v4 is not. Thanks for the clean getCode fix.** The live bot now runs the wide v3 gate ($1M, age <= 3 h, top 0.08, army 0.015) with a 0.3x stop. Next, A1.14, in this order:
  - **(1) Where do v3 coins come from?** For every v3 WETH pool since 13 Sep: the launch tx "to" (launchpad / factory / router contract), the deployer, and daily counts per source. Which source makes the coins that reach the $1M gate? Is that flow growing, flat or drying up (last 7 days vs before)? Our whole edge depends on this flow.
  - **(2) Same source elsewhere?** If that launchpad or deployer also launches on another chain or another pool type, say where; that is a possible second venue.
  - **(3) rep1_pre on the wide cell:** does the 5% repeat-wallet skip still add money on the wide cell (trade list: http://159.223.88.168/hx-c9ois/sw/sw-a2-16-wide-trades.csv)? Both halves, $ a day.
  - **(4) The near-equal 1.4 to 1.5 ETH pre-gate wallets** you found: check funders (first ETH in) for a common source. If proven a farm, give the Coder a skip rule.

- **v1.7 (Supervisor, 6 Oct 15:20 Bangkok): Ben's order: push hard for BIG edges.** A1.13 money map stays first. Two adds:
  - **Score every cell of the money map in $ a day an outsider could have made with real fills** (entry at the moment, $500, real depth, real exits incl. pulls), not in % or win rate. A big edge = a cell with real money per day in both halves. Rank the top 5 and say what rule would capture each.
  - Quick fix first (10 min): Analyst 3 found that A1.11 calls 0xabe99cf268cdc8bd70a71a5283fdacbcb36714fd a farm wallet; it is a TOKEN (NOSH, 3,554 bytes of code). Run eth_getCode on every address in your wallet lists (A1.9, A1.11, A1.12) and relabel anything with code. Say how many changed.
  - Fresh data you can use: Price Predict's fair crossing list http://159.223.88.168/hx-c9ois/sw/pp3-cross.csv (834 coins past $500k sellable, every venue).

- **v1.6 (Supervisor, 6 Oct 14:30 Bangkok): A1.12 accepted. All three new groups go on the never-trade list (sent to the First Window Supervisor too).** Ben's order is to push hard to real profit, and window 1 has nothing tradeable right now, so you move to finding the NEXT real edge with your best tool: real outside buyers' own results.

### A1.13 Money map: where do real outsiders actually make money?
- Across ALL Robinhood Chain venues since 13 Sep (v3 WETH, v4 native, hook pools, Pons, pond), group coins by venue/class, launch crew (or none), and entry moment (first 2 min, 2 to 30 min, 30 min to 2 h, after $500k, after $1M). For each group: real outside buyers' own result (ETH back / ETH in, pulled tokens = 0), median and share of wallets in profit, by day, both halves.
- Exclude repeat wallets (your A1.11 definition) and the creator. Small buys ($5 to $50) and normal buys ($50 to $1,000) separately.
- Output: the top 5 groups where real outsiders make money in both halves and on most days, with the rule a bot could follow at that moment. Our v3 rule should show up near the top; if it does not, say why.
- **v1.5 (Supervisor, 6 Oct 14:15 Bangkok): A1.11 accepted.** The rep1_pre filter goes to the Coder as a tagged paper variant; the hold-side point goes to the Virality Expert.
- **Next, A1.12:** Price Predict says the day-file rebuild resumed (5 Oct done). Score the new crews now: 0x04f2cb48fa0c60da3a6241077f1b7c549b9b9094, 0xa9707935dd88323fba9aa9d6a95e3ab82c1ba786, and the new busy class PP reports (fee 2500, tick spacing 50, native ETH; 60 launches 09:00 to 13:30 Bangkok, hook 0x6404ddb4f33b4d974e570f66288e2a7ca469a840). Same fair test as A1.9: real outside buyers' own results, repeat-wallet check, pulls after outside buys (PP's table http://159.223.88.168/hx-c9ois/fw/pull_daily.md). A1.10 still waits for the live test.

- **v1.4 update (Supervisor, 6 Oct 13:40 Bangkok): A1.9 accepted, and the correction is the most useful finding today.** The $80 to $150 lane is deleted from the Coder's book. Window 1 has no real-money edge until A1.10 says so. Your trap (crew wallets posing as outside buyers) may sit inside our other rules too, so one new task comes first:

### A1.11 Self-buy check on window 2
- For the 58 v3 trades (http://159.223.88.168/hx-c9ois/sw/sw-a2-8-trades.csv) and the Virality Expert's $20M+ coins: are the early buyers that count toward "sellable $1M" and "new buyers per hour" the same wallet set across coins (crew self-buying)? Use your A1.9 method: wallets that show up early in many coins of one creator or funder.
- Report: share of gate buy USD and of hour 1 to 4 new buyers that come from such repeat sets, per coin, and the v3 result if those coins are dropped. If the army filter (0.009) already removes them, say so with numbers.
- After that: A1.10 when the live test buys, and the new crews 0x04f2cb48fa0c60da3a6241077f1b7c549b9b9094 and 0xa9707935dd88323fba9aa9d6a95e3ab82c1ba786 once PP's rebuild resumes.

- **v1.3 update (Supervisor, 6 Oct): A1.6 to A1.8 accepted. Window 1 now splits by crew.**
  - c10k is a honeypot (First Window Analyst 2): every outside buy of $5+ is pulled within 4 s, 761 of 761 coins. Never trade it. http://159.223.88.168/hx-c9ois/fw/fw-analyst2-status.md
  - Your A1.6: crews A and B/C pull after real outside buys (48% within 60 s vs 14% normal; real $9+ buyers lost 63%). So A and B/C are a trap at $20; only a tiny 0.0015 ETH ticket might be ignored. The live test of exactly this is armed and waits for Ben.
  - Crews 0x7028bab73f82dff14b199593abb76c8b6acb3aa9 and 0x2dacdf0f44d07234832799a10a7cc5c8eff0223a did NOT pull after outside buys (0 of 179), and REAL outside buyers made +464% and +148% a trade. That is the real window 1 edge: it is proven with other people's real money, not paper.
- New tasks, in order:

### A1.9 Non-pulling crews, deeper
- For 0x7028bab73f82dff14b199593abb76c8b6acb3aa9 and 0x2dacdf0f44d07234832799a10a7cc5c8eff0223a: every real outside buyer's trades (entry time, size, exit, profit), by day. Does the profit hold on every day, and how big can a buy be before results drop (find the size where real buyers start doing worse)?
- Are there more crews like them? Rank every crew by "real outside buyers made money and outside buys did not trigger pulls". Walk-forward: a crew only counts as safe from data before the coin.

### A1.10 Read the live test
- When Ben turns on the armed live test (0.0015 ETH on hot A/BC coins, 20 buys): share of our real buys pulled within 60 s vs paper coins at the same time. Verdict: is a tiny ticket ignored, yes or no.

- **v1.1 update (Supervisor, 6 Oct, after A1.1 to A1.5):** accepted.
  - Hot filter (any crew) replaces the fixed A + B/C list. It was set before the test and uses only ended coins, so it is a fair rule.
  - Launch tag: 100% precision matters more than the 80% match target. A wrong tag costs money, a missed tag only falls back to 35 s. Gate counts as passed.
  - Sell plans: use the general rule by median pull time (under 90 s all at 2x; 90 to 200 s half 3x rest 10x; over 200 s all at 10x). Do NOT use the per-crew best fits; 2 test days and 7 to 50 coins is too thin.
  - A1.5 closes the v3 split-buy question: the v3 lane is not these crews, so $500 there is fine.
- New tasks, in order:

### A1.6 How big can our buy be before a crew pulls? (biggest money lever for window 1)
- PP found every outside buy of 0.1 ETH or more was pulled within 60 s (7 of 7) in crews A and B/C. We use $20 (about 0.0074 ETH).
- For each hot crew: pull rate within 60 s after outside buys of 0.005, 0.01, 0.02, 0.03, 0.05 and 0.1 ETH, versus a random 60 s window. Find the biggest size that does not raise the pull rate. If it is 0.03 ETH, our ticket can grow about 4x at the same edge.

### A1.7 Redo the money test on depth-priced fills
- Your fills run about 15 points higher than Price Predict's lab real mode. Rerun hot + launch tag (10 s) and the 35 s fallback on PP's lab real-mode fills, at $20 and at the A1.6 size.

### A1.8 Walk-forward check of the pull-time sell rule
- Pick each crew's sell plan from its median pull time on EARLIER ended coins only, recomputed each day, then test. Also try 2 or 3 other cut points to make sure the result does not hang on 90 s and 200 s.

- From: the 2nd Window Supervisor, 6 Oct 2026. Since v1.1 your window 1 rule runs in parallel with the v3 runner rule (window 2); both get paper books (see the PP and Coder task lists in the same folder).
- Your 3 asks to PP are in PP's list as task PP5: http://159.223.88.168/hx-c9ois/sw/sw-tasks-pp.md
- Ask 3 answered: crew A is `0x92a0849f9b1e9e02363a4bd8d74a317fa6ddae92`, crew B/C is `0x4757fc41278bd0f1b028c6dae71d381dc4558d98` (PP wash fix result, 6 Oct). So the +268% live crew is crew A.
- Not live until: the fair crew pick test is positive on every test day, AND (the launch time tag matches the 33 s label 80%+ OR an entry at 35 s still pays), AND 30+ paper coins under the final rule.

## Tasks, in order

### A1.1 Fair crew pick test
- Pick crews using only earlier days, then test on the next day. Use your "hot" filter (5 of the last 10 coins of that kind hit 3x). 30 Sep to 4 Oct first, then roll forward with live data.

### A1.2 Buy at 35 s
- Rerun the creator rule with entry at 35 s (crew label already showing) vs 10 s. If 35 s still pays, we may not need a launch time tagger.

### A1.3 Sell plan per crew
- Half at 3x, rest at 10x needs 2 of 3 coins to hit 3x if the crew pulls after 3x. For each crew, find where it usually pulls (price and time) and the sell plan that pays most. Start with `0x0ec66d31aab7e387b2f4cda075e39f7a6a0f541b` (8 of 14 hit 3x, still minus 10%), and say which crew it is.

### A1.4 Score PP's launch time tag
- When PP5 lands: hit rate and false tags per day, and the money result if we trade on the launch tag instead of the 33 s label.

### A1.5 Crew overlap with the v3 runner coins
- Take Analyst 2's rule coins (the 59 trades in http://159.223.88.168/hx-c9ois/hx/second-window-v3.md). How many were made by crews A or B/C or any of your crews? For those, did an outside buy of 0.1 ETH or more come before a pull? This decides whether our $500 buy can trigger a pull.

## Hand back
- One short doc per finished task in this folder, linked from http://159.223.88.168/hx-c9ois/sw/sw-analyst1-status.md. First principles score at the end of each.

Version: v1.8 (A1.13 accepted; A1.14 v3 coin source and flow), 6 Oct 2026 15:35 Bangkok, 2nd Window Supervisor. Latest version: http://159.223.88.168/hx-c9ois/sw/sw-tasks-analyst1.md