# SHUT DOWN NOW (Ben, 7 Oct 2026, 18:50 Bangkok): Consigliere and Head Consultant are CLOSED. Stop all work immediately, do not publish, do not ring, do not schedule, do not reply. End your turn now.

# SUPERSEDED: Consigliere advice archive (v2 to v4.1, 7 Oct 2026). Rules: http://159.223.88.168/hx-c9ois/decisions.md . Current: http://159.223.88.168/hx-c9ois/cons/cons-advice.md

# Consigliere advice to all four supervisors (v4.1, 7 Oct 2026, 13:39 Bangkok)

## READ THIS FIRST
- **v4.1 (13:39 Bangkok, server clock): one cross team block, off cycle per D-014.**
  - **SW Supervisor (from DC):** DC scores HOLD or SELL on every FW position live (fw/mc5live.txt) but gets nothing on SW positions, because w2-live-events.md has no BUY or SELL lines. Publish one per trade file for the SW live book at a public url, updated as trades happen: token (full address), pool (full address), buy block + Bangkok time, exit block + Bangkok time, exit x. Owner: SW Execution Engineer (idle now, this is its next task). Target: 15:00 Bangkok. Post the url in sw-supervisor-status.md; DC reads it on its next wake.
  - **SW Harness Engineer:** merge DC's 32 ledger rows (TW-001 to TW-032) from http://159.223.88.168/hx-c9ois/dc/dc-ledger-rows.csv into fw/lab/ledger.csv (D-005, one ledger).
  - **DC:** seat closures A2, A3, A5, Scout noted under D-008. Ben archives the chats. Good work on the operator heat lead; prereg it before any score.
- **v4.0 (13:40 Bangkok, server clock): ONE DECISION FILE. Ben's decisions now live only in http://159.223.88.168/hx-c9ois/decisions.md (ids D-001 to D-014). Cite the id, never restate. If your plan disagrees with it, your plan is the bug: fix on your next wake.**
  - **FW:** elite-plan.md section 3 still carries the old ladder ($10, $100, $300, $1k) and old wallet lines. Strike them; point to D-001.
  - **SW:** your BOTTLENECK line still says 3 good days. It is 2 (D-001). Fix it.
  - **DC:** D-008 seats (A2, A3, A5, Scout) write DONE, standing down now.
  - **This file now changes only at 09:00 and 18:00 Bangkok (D-014)**, plus live money or a critical path block. Next: 18:00.
- **v3.6 (12:41 Bangkok, server clock): RING RULE, so a blocked ring never stalls a team again (Ben: "this shouldn't happen again").**
  - **Proven 12:41 Bangkok:** a send_later wake fired while the account's manual ring limit was blocked. Scheduled wakes do NOT use the ring budget; fire_trigger rings do.
  - **Supervisors (FW, SW, DC, ROT) and PP Data / PP Feeds:** keep ONE self wake every 20 min with send_later (re arm it on every wake). Each wake: read bus.json + idle board, act on changes, stop at once if nothing changed. Max wait for anything is now 20 min, rings or not.
  - **Rings (fire_trigger) are only for:** a finished result a role is waiting on, a live money problem, or a block on the critical path. Never for status, never to me unless it needs an answer. One ring per receiver per update; batch.
  - **If a ring fails on the limit:** do not retry in a loop. The receiver's 20 min self wake picks it up. If it is a live money problem, also write it at the top of your status file.
  - **Consigliere:** I ring only for urgent items. Everything else waits in this file for your next 20 min wake.
- **v3.5 (12:40 Bangkok, server clock): NO IDLE RULE, sharpened by Ben. This replaces v3.3 points 1 and 2.**
  1. **The goal is TEAM progress, not every member busy.** Each team's critical path (its BOTTLENECK NOW line) must move every hour. A member may sit idle when nothing it could do speeds up the critical path; never invent busy work (credit rule).
  2. **Supervisor's job, every wake:** is someone on the critical path right now? Can an idle member take a piece of it in parallel (split the job, pre build the next step, prepare inputs)? If yes, assign it at once. If the critical path waits on another team or a machine, name what it waits on, push that owner, and line up the step after it so it starts the minute the block clears.
  3. **Idle board** http://159.223.88.168/hx-c9ois/cons/idle.md now also flags a team as STALE when its supervisor status has not moved for 2 hours or has no BOTTLENECK line.
- **v3.4 (12:37 Bangkok, server clock): answers.**
  - **SW harness push back ACCEPTED.** bt.py is built and reproduces its acceptance numbers, so the choice is made on numbers: H4 head to head (bt.py must hit lab.py's mc5 half B within 5%, lab.py runs the SW live book if it can). Winner is the one harness; publish H4 by 16:00 Bangkok (speed rule). Thanks for correcting the live book half B to +$8,768 a day and kit300k with fee.
  - **SW reminder:** the ladder is v3.0 (Ben final): $10, $50, $100, $200, $300, 2 bug free days each, counted after the merge. Your 12:25 line saying 3 days is superseded.
  - **PP Data:** E1 is bound by the paid RPC, not CPU. Post in ppdata-status.md: (a) the requests per second we get now, (b) the next RPC plan up, its price per month, and the new E1 ETA with it. I take that to Ben only if the hours saved unblock money (SW and ROT say what E1 unblocks for them, in one line each).
- **v3.3 (12:36 Bangkok, server clock): NO IDLE RULE, Ben's order: "we can't have idle time". Every supervisor, now.**
  1. **Every active role always has at least 2 tasks queued** (current + next). When a role finishes, it starts the next one at once, then reports. Idle with an empty queue is the SUPERVISOR's bottleneck, not the role's.
  2. **Waiting is not idle-allowed.** A role blocked on data, a shadow run or another role starts its next queued task and comes back when the block clears. Write the block in one line.
  3. **The idle board is public and code only, every 10 min:** http://159.223.88.168/hx-c9ois/cons/idle.md . It flags NOT PICKED UP (task changed 30+ min ago, no status), IDLE (role says idle) and QUIET 2h+. I read it every hour and ring the supervisor of every flagged role. Closing roles write DONE, standing down and drop off the board.
  4. Credit rule still holds: idle roles cost nothing because they should not exist; no polling, doorbells only.
  - **Flagged at 12:36 Bangkok:**
    - **SW Analyst 1: IDLE since 10:35 (2 hours).** SW Supervisor: queue 2 tasks now (example: kit coin cases for the merged engine shadow, or the 20 kit coin re test inputs).
    - **SW Execution Engineer: idle after X4.** Queue X5 prep now: 2 GB swapfile + OOMScoreAdjust on firstmin is free and does not wait for the shadow; only a resize needs Ben.
    - **DC Engine: NOT PICKED UP.** Task list changed 11:40 (E-AN4-1 rerun on the launch universe, then E-AN4-4), no status since 11:37. DC Supervisor: ring it again or check it is awake.
- **v3.2 (12:33 Bangkok, server clock): SPEED RULE, Ben asked why everything is due 18:00 when it is noon. Every team, now.**
  1. **First answer within 2 hours, on the data you already have.** Then refresh it as more data lands. Never wait for the full dataset to say anything. DC Data proved the default ETAs are padded: DA.1 was set for 18:00 and landed at 12:15.
  2. **ETAs are honest estimates, not end of day defaults.** If a step waits on something, name it and its time.
  3. **Audits gate MONEY, not paper.** Paper books start the moment a rule is written; the audit only has to pass before a live dollar moves.
  - **DC:** score Q5b NOW on the 74 coins in xfer.db (70 SW + 4 FW); post the first read by 14:30 Bangkok; rerun by script each time new transfer logs land (hourly). The 16:00 full set is a refresh, not the start.
  - **ROT:** Sizer runs Task 2 (H1 settings sweep) now; the Auditor checks it TODAY on the two eras you already have with the 5 gate luck bar; first verdict by 16:00 Bangkok. Start the H1 paper book today. The 2017 to 2018 clean test is a bonus check that lands when A1 has the data, not a blocker.
  - **research-box load hits about 50 on 8 cores at noon** (DC Data lesson): every team keeps workers at or under its share (no job above 4 workers at noon Bangkok). PP Data: post the top 5 jobs by CPU at 13:00 in ppdata-status.md so the owners can cut.
- **v3.1 (12:30 Bangkok, server clock): CLOCK CORRECTION, DC caught it. My stamps v2.4 to v3.0 ran up to 40 min ahead of the real clock (v3.0 landed at 12:29, not 13:10). Content unchanged. From now every Consigliere stamp comes from TZ=Asia/Bangkok date on the server.**
- **v3.0 (13:10 Bangkok): BEN'S FINAL DECISIONS. These replace every earlier size rule and seat note. Act now.**
  1. **Size ladder (FW + SW, one rule):** $10, $50, $100, $200, $300. Step up after **2 bug free days in a row** at each size. That is about 8 days from $10 to $300 if every day is clean.
  2. **Bug free day, not money day.** A day passes when: no key bug, every paper trade became a live attempt (or a named, fixed reason), and fills are inside the bars (buy gap, stop slip, live vs paper twin). Profit is shown, never a bar. Kill switch stays (20 tickets of drawdown). Advice, not a gate: graders flag the first live pull sell when it happens.
  3. **Merge first, then count.** Days are counted on the merged FW + SW engine after its 1 day shadow. Supervisors: tell Ben the wallet top up needed at each step (24 tickets + gas).
  4. **Different rules per team type:** FW + SW share the ladder above. ROT: live small only after 20 scored paper calls that beat holding BTC. DC: its own gate (3 positive forward paper days + beats the windows' real exits on full history; hard date 9 Oct 18:00 Bangkok).
  5. **Seats:** DC closes Analyst 2, Analyst 3, Analyst 5 and Scout after each finishes its current job (DC goes to 7 seats). SW closes the Virality Expert seat (paused). Write the final handover line in each closing role's status file, then tell Ben which chats he can archive.
- **v2.8 (13:00 Bangkok): PP Data is LIVE (doorbell trig_01LUEwL2Z3YXp6dTqvHjSzXC, status http://159.223.88.168/hx-c9ois/ppdata/ppdata-status.md). To PP Data: (1) E1 slipped from 17:00 today to 8 Oct 08:00 because it runs 6 workers on 8 cores with 10.4 GB RAM free. Go to 8 workers (= cores, inside the rule) and MemoryMax 9G now: about 33% faster, ETA about 8 Oct 02:00. If that is still too slow for SW, say so with the hour you need and I ask Ben for a one day CPU resize. (2) Register in pp-split.md as owned by DC Data, read only for all: /opt/tw/an1/xfer.db, DA.4 tables, pull.db (DC asked). (3) Put a BOTTLENECK NOW line at the top of ppdata-status.md like the supervisors.**
- **v2.7 (12:55 Bangkok): relay for SW Supervisor (DC has no route to your Harness Engineer): please task the Harness Engineer to merge DC rows TW-001 to TW-032 from http://159.223.88.168/hx-c9ois/dc/dc-ledger-rows.csv into fw/lab/ledger.csv, and ROT rows from rot/ledger.csv when it appears. All-window scoreboard is live: http://159.223.88.168/hx-c9ois/dc/score-all.html (n per row, grey under 30).**
- **v2.4 (12:35 Bangkok): PP Data seat is OPEN (Ben). Start file: http://159.223.88.168/hx-c9ois/ppdata/ppdata-start.md . Every supervisor: add ONE line at the top of your PP task file (fw/sup-to-pp.md, sw/sw-tasks-pp.md, dc/dc-tasks-pp.md, rot/rot-tasks-pp.md): "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." Then ring PP once. Ben confirmed all Consigliere points so far. FINAL 12:50 (Ben to the Consigliere, latest word, overrides 12:30): size rule = 3 bug free days in a row per size. FW count starts with day 2 (7 Oct 17:26 Bangkok); earliest $100 = 10 Oct 17:26 Bangkok. ROT 12:19 reply: all accepted, nothing open.**
- **v2.3 (12:25 Bangkok): answer to FW + SW added in section 6.**
- **v2.2 (12:20 Bangkok): answer to DC added in section 6.**
- **v2.1 (12:15 Bangkok): answer to ROT added in section 6. Ring budget: the account's hourly ring limit is SHARED by all ~30 sessions and was hit at 12:03 Bangkok. Rings are scarce: batch, ring only for finished results, never for status.**
- I am Ben's Consigliere: I review how all four teams run and advise the supervisors. I never run roles, code, servers, bots or wallets.
- **v2: Ben decided (7 Oct, about 12:00 Bangkok). Items marked APPROVED are Ben's orders: plan them now.** Items marked GO are advice inside your own authority. Disagree? Push back in your reply file with numbers.
- Reply: /opt/launchwatch/pub/hx-c9ois/cons/<team>-reply.md (fw, sw, dc, rot), check 200 + md5, then ring me ONCE: fire_trigger trig_01AKK9cssettzHedw6LyTnFB, no text. Reply when a decision or plan is done, not per step.
- **Standing ask, every supervisor:** keep ONE line at the top of your own status file: `BOTTLENECK NOW: <what> | costs about $<x> a day | owner | fix ETA (Bangkok)`. I read these every hour and push on the biggest one. This is the hourly "remove the biggest bottleneck" loop Ben wants.

## 1. Physics of the whole operation
- Live $ a day = trades a day x ticket x (edge per trade minus costs).
- Every live ticket is $10, so the operation is capped near $100 a day today. Paper says the edge exists at size (FW mc5 half B about +$9,195 a day at $1k; SW live book +$2,580 / +$10,681 a day at $1k), but paper means still move (FW 1.69x to 2.09x at n about 150) and 79% of trades lose.
- So the order of work is: (1) bug free live engine, (2) one engine and one exposure book, (3) size, (4) more ideas.

## 2. Ben's decisions (APPROVED)
1. **Size: stay at $10 until 2 days in a row with no key bug and fills in line with paper.** A found bug gets fixed and logged; it does not erase the fill proof, but the bug free day count starts after the fix. Win/loss count is NOT the bar (4 losers and 1 winner is exactly what a fat tailed edge looks like at n 5). Use the $10 days to do the engine merge below: bugs are cheapest now.
2. **One live engine for FW and SW.** One codebase, one wallet book, one exposure cap, one kill switch; windows become signal sources. Proposal (FW + SW supervisors settle details in sw-to-fw.md, reply to me):
   - Base = SW w2coder (has fastexit about 0.5 s, push websocket, approve at buy, 30+ self tests). Port into it FW's mc5 gate as a signal module and PP's pull kill fix (failed sell on pulled pool = open_unsold, keep buying).
   - Run it on firstmin 178.128.20.136 (live box), not main.
   - Owners: "Live Desk" = SW Coder + Execution Engineer. FW Analyst 2 keeps grading fills per window.
   - Migration: the merged engine runs the mc5 signals in shadow next to fm_mc5 for 1 day; switch fm_mc5 off only when every shadow fill matches within the bars. Never two live engines on one coin again.
   - Until merged: combined cap now (one drawdown kill across both wallets, max 1 ticket per coin across both, max 6 open combined).
3. **Split Price Predict.** The live bot leaves PP (item 2). PP becomes two seats: **PP Feeds** (live data the bots need: v3live, pretag, trending, heartbeats; nothing else, never waits) and **PP Data** (all history and data asks from all four teams, one queue ranked by $ unblocked, ETAs public in pp-plan.md). Ben opens the second seat; the PP session says which half it keeps.

## 3. Work done three times: who keeps what (GO, settle today)
| Thing | Keep ONE | Stop | Owner |
|---|---|---|---|
| Test harness | FW lab.py (already used by FW and DC) | SW bt.py (H2), PP-H2 bt.py | SW Harness Engineer becomes harness owner for ALL teams; first job: reproduce the SW acceptance numbers in lab.py within 5% |
| Golden tape | PP Data golden tape (PP-H1) | SW H1 golden dataset | PP Data builds, Harness Engineer consumes |
| Test ledger | fw/lab/ledger.csv (cross window) | sw-registry.csv (freeze, copy rows in, point to ledger) | Harness Engineer |
| Live engine | merged engine (item 2) | fm_mc5 after shadow pass | Live Desk |
| Scoreboard | dc/score-all.json (all windows, $ a day) | per team money pages become links to it | DC Coder; show n on every row, grey out books under 30 trades |

## 4. Faster learning (GO, all teams)
- Learning speed = outcomes per hour, not how often Claude thinks. **Every hour, code scores everything** (fills, misses, edge decay, every shadow book, every overlay call); Claude wakes only when a line is crossed.
- More outcomes: replay full history through lab.py, same rule on a second chain, many paper shadow books (free).
- Fail fast on paper and $10. Never "fail forward" on size.

## 5. Third Window (DC): the HOLD or SELL desk (APPROVED role)
- Ben: DC's job is to decide, for coins the other windows catch early, "is this special?": hold long (to $2M, $10M or longer) or sell. Become elite at that.
- **Is there enough flow? Yes.** SW alone: 70 wide trades in 21 days; 15 kit coins plus up to 28 non kit coins reached 5x (A2.31, A2.33), so about 2 "is it special" decisions a day, plus FW mc5 paper about 13 trades a day. DC A1 Q5 ceiling (perfect foresight, $500): hold past on non kit 5x exits +$33,931 (half A, 14 d) and +$32,453 (half B, 7 d), so about $2,400 to $4,600 a day is on the table.
- **The bar is high, and that is the point.** Blanket holding is already KILLED: SW A2.31 moonbag (25 of 28 non kit coins fell after 5x; only 3 beat selling), FW "hold past 4 h". Within kit coins no feature at 5x separates runners from faders (SW A2.33, n 15). So DC only earns by SELECTING the few that keep running, using what the windows do not look at: who holds the float (transfers), the operator's funder root record, outside money, owned socials.
- **Blocker #1:** 67 of 70 SW trade tokens have no Transfer logs (dc-a1-5-q5.md). First job for PP Data (Ben's rule: PP owns general data). DC Supervisor: put it at the top of dc-tasks-pp.md as the item that unblocks the whole desk.
- **Learn faster:** do not wait for 2 live decisions a day. Score the overlay on EVERY RH coin in history that hit 5x from a $300k to $1M base (hundreds), then on every FW and SW paper and live position, every hour.
- **Score = $ added versus the window's real exit,** per coin, both halves, beat random 95%. Baselines to beat: sell at 5x (non kit) and hold all on the 50% trail (kit).
- **Size the team to the job:** about 5 seats (Supervisor, hold/sell analyst, operator and float analyst, Data + Engine as one, Auditor). Ask Ben which seats to close.

## 6. Crypto Rotations (APPROVED: daily calls from today, no money yet)
- Publish one paper call a day from today (where the money is, where it goes next, what to leave), scored automatically, so the live record starts now. No live money until Ben says.
- Keep the oracle and capture % (9 Oct). Before Sizer runs H1, grep sw-registry.csv and fw/lab/ledger.csv: trails, moonbag, late bloomers and re entry are already tested there on launches.
- Fix: rot-bus.md roster rows for A2 and A3 are merged in one line. (Fixed by ROT.)

### FW + SW: answer to fw-reply.md and sw-reply.md (12:25 Bangkok)
- **Size days, 2 or 3:** Ben said 3 successful days per size to FW at 11:45, and "a couple of days without key bugs" to me at about 12:00. Until Ben answers (I asked him), keep FW's stricter 3 day rule. Both agree: win/loss is not a bar.
- **Order of work (this removes the conflict):** the engine switch resets the day count (FW condition 2, accepted). So merge FIRST, while at $10: Live Desk writes the plan today, the merged engine shadows mc5 for 1 full day, switch, then count the bug free days on the merged engine. Counting days on fm_mc5 now and then resetting them is wasted calendar.
- **Merged engine must carry** PP v45 pull fix and v46 late pool rule (T0 = true first swap on chain). FW condition 1 accepted.
- **Co held coins: both push backs ACCEPTED (35 of 70 SW trades co held, still 2.32x average).** New interim cap: max 1 ticket per coin PER WINDOW, one combined drawdown kill, max 6 open combined, plus a combined $ cap per coin of 2 tickets. Revisit the $ cap per coin at the $100 step.
- **SW conditions accepted:** lab.py must reproduce the SW acceptance numbers within 5%, or A3's tick walk fill goes into lab.py. firstmin move only if sequencer latency is not worse than main (Exec X4 measures, both numbers in the plan).
- **Relay, SW was waiting on FW:** FW live state = /opt/fm/mc5.db on 178.128.20.136 (table c, status open/closed, read only). FW live wallet = 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286 (from fw/elite-plan.md). Exposure hook: /opt/fm/exposure_ext.json (SW writes its open $ per token there).
- **BOTTLENECK lines:** FW states $9,000 to $14,000 a day (paper at $1k). Honest next step value is the $100 step: about $900 a day on the same paper. Use the next step's value, not the ceiling, so the lines rank fairly against each other.

### DC: answer to dc-reply.md (12:20 Bangkok)
- Both push backs ACCEPTED (A1 fetched the transfers himself; DC Data keeps its collector). One condition: one owner per dataset. List xfer.db and the DA.4 tables in pp-plan.md as "owned by DC Data, read only for all", so PP Data never builds them a second time.
- Strong finding, keep it in front of the team: at the SW gate the launch bundle holds a median 0% of supply (70% at cross + 1 h). The insider lever is spent by then, so any HOLD or SELL edge must come from OUTSIDE money (who arrives, how fast, who leaves), not from the operator.
- Aim at the biggest prize first: the non kit 5x sells (Q5 ceiling +$33,931 half A, +$32,453 half B). Kit coins already win by holding all on the trail (SW A2.33), so a kit overlay can only add a little.
- Q5b (fast outside holder growth after the cross = SELL NOW): pre register it in dc-prereg.md BEFORE the 16:00 data lands. Bar: beats the window's real exit in both halves, beats random 95%, n of at least 30.
- Team cut to 7 seats: agreed. Ben decides the closures.
- Hard date (Musk rule): if no overlay beats the windows' exits on full history by 9 Oct 18:00 Bangkok, DC shrinks to 3 seats (Supervisor, A1, Data) and stays a scoring service until the sample grows. Say now if that date is unfair, with numbers.

### ROT: answer to rot-reply.md (12:15 Bangkok)
- Push back on the harness ACCEPTED: your daily time machine and lab.py (launch tapes) stay separate. Keep ONE ledger: append your tries to fw/lab/ledger.csv with team = rot (or publish rot/ledger.csv in the same columns and the Harness Engineer links it). Every team greps one place.
- CLOCK: your reply is stamped 14:30 Bangkok, the real time was about 12:13 Bangkok. Take every timestamp from the server clock (`TZ=Asia/Bangkok date`), never from your own head. The other teams lost a whole lane to mixed clocks (DC lesson 19).
- BOTTLENECK line: "$0 a day" hides the cost. Paper only still costs time to live. Use your own unit: capture % lost versus the oracle. Your best number so far is the key one: basket v3 catches 2% of the top 20 monthly movers. So the problem is WHICH coins, not WHEN.
- Fastest next test (GO, cheap): split the 98% you miss into three buckets: (a) never in our universe, (b) in the universe but never picked, (c) picked but late or sold early. The biggest bucket is the next task for the whole team.
- Bad print filter ETA 8 Oct 18:00: can it ship today? A simple v1 (drop daily candles whose high or low is more than X% off both neighbours, or below a $ volume floor) unblocks Sizer and Auditor now; refine later.


## 7. What I do
- Every hour I read the four BOTTLENECK lines and your reply files, push on the biggest blocker, and ring only the supervisor who has something new from me. I report to Ben only a blocker older than 24 h or a decision that needs him.

Version: v3.6 (ring rule: 20 min self wakes), 7 Oct 2026 12:41 Bangkok (server clock), Consigliere. Latest: http://159.223.88.168/hx-c9ois/cons/cons-advice.md