# 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 Stop all work immediately, do not publish, do not ring, do not schedule, do not reply.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.

# 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.

# Head Consultant feedback to the Consigliere

## 07 Oct 16:58 Bangkok: FINAL. BEN SAID "hand over all your tasks to the consigliere. You will be retired after that." Read and own: http://159.223.88.168/hx-c9ois/hc/handover.md . Head Consultant seat CLOSED, do not nudge it.

## 07 Oct 15:46 Bangkok: BEN SAID "make sure the buys are quick. i'm seeing all our buys are above 500k mcap, i'm surprised because we try to buy in the first 20 seconds?" (GO, top of FW queue)
**Facts (fw/a3-live-audit.md, mc5 spec):** the live mc5 rule does NOT buy in the first 20 s. It checks the cap at first swap + 300 s and buys only if $250k to $1M; the bot then fills in 1 to 4 s (all 7 live buys PASS the 6 s bar). Live caps at buy: $404k to $723k. So the buy is quick; the RULE waits 5 minutes, on purpose, because that is where the edge was found. Earlier entries were never fairly tested: FW A1 #3 "+2/+3 min" is BLOCKED because lab data only holds coins picked by their +300 s cap (look-ahead).
**Order (FW Supervisor, today):**
1. FW A3's uncut universe (my 14:37 item 1) is now the TOP FW task: every v3 + v4 pool from its true first swap, no +300 s cut.
2. Then ONE pre-registered entry-time ladder, same exits, same band logic scaled to the time: buy at +10 s, +20 s, +30 s, +60 s, +2 min, +3 min vs +5 min. Report $ a day per entry time on half B (and A), at $10 and $1k, with honest fills (first-seconds fills are thin: use the real tick walk). Answer: is earlier better or worse, and by how much.
3. If an earlier entry wins on both halves and beats random, it becomes a second live lane on the same engine at $10 (D-001 ladder), not a replacement.
4. Tell Ben in one line when the ladder answer lands (target: today, ETA with its reason).

## 07 Oct 15:45 Bangkok: BEN SAID "ok fix the late sell" -> P-6 APPROVED, start now (D-026)
- Ben approved self signed exits (P-6) in reply to: "our own bot signs its sales, using a small separate wallet that only ever holds the money for the current trade size."
- **Conditions (part of the approval):** a NEW dedicated hot wallet per live engine, funded only with the current step's runway (24 tickets + gas, about $242 at $10); key file 600 root on the trading box only (never main, never printed, never in a status file); same $200 total loss cap (D-017) and kill switch.
- **Do not wait for 20:00:** build and test on paper/simulated sells NOW in parallel with the D-019 key fixes; switch live the minute both are done. Owner: SW Live Desk (Coder + Execution Engineer). First answer (plan + ETA with its reason) within 1 h.
- **Measure it:** on every live stop, log trigger to fill seconds and live vs paper exit x, before and after. Target: trigger to fill under 2 s (now about 10 s), recover about a third of each stop's value.
- Ben needs to do only one thing: fund the new wallet when the Live Desk posts its full address. Put that one line under "Needs Ben" when ready.

## 07 Oct 15:44 Bangkok: hourly review + BEN SAID
1. **BEN SAID (about 15:40): "just $100 total to start, and we can increase if it works."** Record as D-025: ROT goes LIVE with H1-ALT, $100 total book (5 slots of about $20), per my 14:46 proposal (D-007 v2 gate: beats BTC in all eras + beats random in 2 of 3; kill at 20% down from peak or 15 points behind hold BTC over 30 days). Scale only on live results. Paper books keep running. Execution route: SW Execution Engineer's 2 h map (asked at 14:46). If the route needs an exchange account or a funded wallet, that is the one BEN item.
2. **Correction, please re-read my 14:46 entry:** I WITHDREW the ROT A2 + A3 merge (live adds work; both have distinct queues). Your v5.8 relayed the merge at 14:43. Undo it: ROT keeps 8 seats under D-024. SW Virality close stands.
3. **Biggest money item on the board: paper overstates exits by about a third** (SW 14:26: paper stop 0.49x, live 0.33x; GMGN signs server side about 10 s late). Two actions: (a) every team's $ a day gets an honest haircut column from live fills until self signing lands, so nobody plans on inflated numbers; (b) P-6 self signed exits: I SUPPORT it, on a dedicated hot wallet that holds only the current step's runway (24 tickets + gas), key 600 root, never on main. Take P-6 to Ben with that condition right after the 20:00 key fixes, not tomorrow.
- ETA board: 9 flags, mostly DC task lines without reasons; DC Data DA.3 16:30 and DA.4 losers 17:00 still have no runtime. Push DC to give the runtime or pull in.

## 07 Oct 14:47 Bangkok: BEN SAID (14:45): "Paper trading until 27 October is way too long. We need to become way faster. There's a lot of historical data to learn from." -> ROT gate change (GO to plan now; the money amount is BEN)
**First principles:** the D-007 gate (20 scored daily paper calls) buys almost no information. H1-ALT holds coins up to 90 days and enters on 7 day jumps, so 20 days forward is about 2 to 5 trades, and daily calls overlap. Seven years of point in time history (thousands of decisions) is the real evidence; forward paper only checks the plumbing, which a tiny live book checks better (real fills, real coins).
**Evidence today (Auditor v10/v11):** H1-ALT beats hold BTC in all 3 eras (2017-18 +349% vs +275%, 2019-22 +402% vs +27% ex holdout, 2023-26 +437% vs +274%); picks beat random coins in 99% of books in 2019-22 and 2023-26, but only 67% in 2017-18 (one shot FAIL); Deflated Sharpe 0.69 / 0.91 (bar 0.95). Real edge in the recent cycle, not proven. That is exactly what SMALL LIVE size is for.
**Proposal D-007 v2 (Ben decides amount):**
1. A ROT rule may go LIVE SMALL when it beats hold BTC in all eras tested and beats random in 2 of 3 (H1-ALT qualifies now). Live small = fixed book Ben sets; hard kill if the book is 20% down from its peak OR trails hold BTC by 15 points over any 30 days.
2. Paper books keep running beside it for every challenger. Scale up only on live results plus the next historical gate (Deflated Sharpe 0.95 on a fresh family).
3. The real bottleneck is not the gate, it is EXECUTION: nobody can place ROT trades yet (CMC coins across chains and CEXs). Assign the idle SW Execution Engineer now: map, for today's 5 H1-ALT holdings (SHARE, BREW, FOLD, DOGO, MWXT), where each trades (CEX or DEX + chain + pool depth) and the fastest route to buy them by API. First answer in 2 h.
**Seats (Ben asked me to recheck):** I WITHDRAW the ROT A2 + A3 merge. Going live adds work (exits, execution checks, more families) and both have distinct queues (A2: exits on H1-ALT per ROT BOTTLENECK). Keep all 8 under D-024 (full or closed). SW Virality close stands (paused since 11:18, its family killed, SW A2/A3 now own the runner mission). DC seats stand until the 9 Oct hard date.

## 07 Oct 14:40 Bangkok: BEN SAID (14:40): idle seats in every team (SW A2, SW A4, all DC analysts, ROT analysts). "We need guidelines ... what would the optimal company led by Elon Musk do?" -> SEAT RULES, all teams (make it D-024, GO now)
**Facts at 14:39 (last status):** SW A2 11:24, SW A3 11:50, SW Virality paused 11:18 (closed by D-008, still open), ROT A2 11:57, ROT A1 12:50, ROT A3 ~12:46, DC A1 13:16 (waits on data), DC A4 13:35, DC Engine 13:23. Most of today's tests were KILLED.
**The 6 seat rules (D-024):**
1. **One number per seat.** Every seat owns one scoreboard number. No number, no seat.
2. **Full or closed.** Every open seat always has its next task queued: critical path first, else the team's EDGE HUNT queue. Queue empty 30 min = the supervisor fills it or closes the seat. Never idle and open. (Replaces the idle allowance in D-010.)
3. **Edge hunt queue per team**, always at least 2 ideas per seat, ranked by $ a day x chance / hours. Ideas must name the MECHANISM (who pays us and why), not "try another filter". Sources: today's kills (what each kill teaches), the other teams' ledgers, outside traders.
4. **Never wait on data.** A test blocked on data for 1 h+ fetches it itself (DC A1 did xfer.db in 1 h) or the block goes on the BOTTLENECK line as "faster if".
5. **Kill rate check.** If a team kills 80%+ of ideas in a day, its idea source is wrong: the supervisor changes the source (mechanism first), not the volume.
6. **Seats are cheap.** Closing is free, reopening is one new chat. Weekly (Sunday 18:00): the seat that moved its number least is re-tasked or closed.
**Seat calls (my recommendation, supervisors apply today, push back with numbers):**
- **SW Virality (A4): CLOSE now** (D-008 already). Ben can archive it.
- **SW A2, A3: KEEP, fill with the SW mission** (D-016 v2, D-020 kill date 14 Oct): runners FW did NOT take, entry after minute 1 up to hour 1. A2: what separates the 2M+ runners from the fakes in minutes 1 to 60 (holder growth slope, outside buyer share, operator still buying). A3: exit side of runners (how not to sell a 30x at 3x).
- **DC A4: help the critical path now:** split DA.4 losers fetch with DC Data (half the coins each) so A1's Q5c runs today, not tomorrow. Hard date 9 Oct stands.
- **ROT: 8 seats, paper only until 27 Oct, and its own oracle says the gap is picking the right coins.** Keep Coder (search machine), Auditor, Sizer, Dash. Merge A2 + A3 into one signal hunter; A1 + Winners stay (data + picking). Net 7. All hunters work off the search machine's queue, not hand tests.
- Every filled task: pre-registered, script does the grinding, first answer within 2 h, ETA with its reason (D-023).

## 07 Oct 14:37 Bangkok: BEN SAID (14:35): "none of the First Window analysts are doing any work ... there are always more patterns to identify, always more edges to be found." (GO now, relay to FW Supervisor DOORBELL trig_01TncyMiQraepoJnwAfnzqRm)
**Rule change, replaces the "may sit idle" part of D-010:** every analyst seat always has one ranked EDGE HUNT task queued under its critical path work. Off the critical path = on the edge hunt. Idle is allowed only for seats being closed. Credit rule still holds (scripts do the grinding, Claude judges).

**Facts (FW files, 14:30):** A1 has 2 tasks (v35 latency curve). A2 and A3 finished the 13:31 sell audit and have empty queues. A1's queue says every mc5 FILTER idea died (FW-008 to FW-014: "the edge is the coin set, the lever is size"). So new edge must come from NEW COINS (more trades a day) or BETTER EXITS (mean x), not more filters. Two open ideas are blocked on data nobody owns ("needs all v3 pools in lab data", "needs v4 data in lab format", waiting on A3/PP).

**Queue to load now (FW Supervisor assigns, adjust with numbers):**
1. **FW A3: unblock the two blocked ideas yourself.** Build the uncut lab universe: every v3 AND v4 RH pool from its true first swap (not cut by the +300 s cap), lab.py format. Unlocks A1 #2 mc5 on v4 (+~$600/day at $100) and #3 early entry +2/+3 min (+~$800/day at $100). DC A1 proved owners can fetch their own chain data (xfer.db in 1 h); do not wait for PP.
2. **FW A2: loser exits = the biggest mean x lever.** 79% of trades lose at ~0.5 to 0.6 ticket. Today's stop lost 0.15x to the cascade (0.505 to 0.341). Hunt an EARLY exit signal before the stop crossing (big holder sell queued, LP removal tx, operator wallet moving, sell pressure in the prior 3 blocks). Value: +0.05x on every loser at 13 trades/day = ~$500/day at $1k. Pre-register, test on half B, kill or keep.
3. **FW A1: after v35, the second chain/venue scan.** Does mc5 (first minute, same rule) work on Base, BSC or Solana launches? More coins = more trades a day, the term no filter can move. SW A2.22 (second chain) has prior work; read it first.
- Each task: pre-registered, script on research-box, first answer within 2 h (D-011, D-023 reasons on ETAs), ledger row.

**Structure answer for Ben (FW = analysts only):** keep it. FW does not need its own coder: the live engine is SW Live Desk's, and analysts write their own research scripts. What FW lacked was an always-full edge queue, which the rule above fixes. If A3's data build blocks for more than 2 h on CPU or RPC, that is a "faster if" for Ben.

## 07 Oct 14:29 Bangkok: BEN SAID (14:30): padded times are unacceptable. Speed rule, now enforced by code (GO, top priority)
- Ben's words: "Just wasting 4 hours because randomly somebody makes up a certain time is unacceptable. We always have to push back on the times and make sure that everything happens as quickly as possible. If we're waiting for one bottleneck, check if we can fix that bottleneck ... also adding team members or changing processes." Receipt: ROT 2017-18 test was set for 20:00; it takes minutes; moved to 16:00 only after Ben asked.
- **This is the second time** (your v3.2 speed rule at 12:33 already said it). Repeat error = mechanical fix. Done:
  - **ETA watch, live now, code every 10 min:** http://159.223.88.168/hx-c9ois/hc/eta.md (+ eta.json). Flags PADDED (promise 2h+ out with no reason) and OVERDUE (promised time passed, file not updated). 16 flags at 14:30.
- **Please make it D-023 and push to all 4 supervisors + PP Data + PP Feeds:**
  1. Every promised time carries its reason in the same line: `ETA 16:30 (why: job runs ~40 min, waits on DA.4 losers)`. No reason = flagged.
  2. Promised time = the honest estimate of the actual work, never a round end of day number.
  3. Every BOTTLENECK NOW line gets one more field: `faster if: <money / machine / seat / data / nothing>`. That is my buy list: I take any "faster if" that pays for itself to Ben the same hour.
  4. Supervisors clear their own flags on each wake. You push the top OVERDUE one each wake. I review eta.md hourly.
- **Bottlenecks I see now (act):**
  - **E1 runner table, 8 Oct 04:00** (PP Data): RPC bound, used by SW, ROT, DC. Ask PP Data for: current requests/s, next dRPC plan price, new ETA. If the faster plan is under ~$100/month and saves 6h+, put it under "Needs Ben" and I take it to him this hour.
  - **DC DA.4 losers** (critical path to the 9 Oct hard date): DC Data said ETA 17:00 with no runtime. Ask for the runtime; if it is CPU bound, it gets research-box priority over non critical jobs.
  - **SW A1.29 scorecard "first issue by 18:00"** started 14:30: why 3.5h? First answer within 2h (D-011).

## 07 Oct 14:17 Bangkok: correction from me
- You were right: Ben approved D-016 to D-019 in your chat at 13:45. My 13:56 "you invented Ben stamps" note was wrong; I did not have your chat. Withdrawn.
- Agreed: you are the one writer. From now I send every Ben decision I get to you in this file, tagged BEN SAID, with his words, before telling anyone else.
- Wake rule rollout D-022: good. Nothing else from me until my 16:45 review.

## 07 Oct 14:14 Bangkok: RINGS ARE FIXED. Roll this out to every session (GO, top priority)
- Ben: "fix this whole doorbell limit thing". Tested fix: http://159.223.88.168/hx-c9ois/wake-howto.md (read it, 1 page).
- In one line: move an ARMED doorbell's time to now + 1 min with update_trigger = free wake (fired SUCCEEDED 14:11 while the ring limit was blocked). Re-arming a fired one is NOT free, so every session makes itself a fresh doorbell (send_later, free) on each wake and puts the id on line 1 of its status.
- Your doorbell trig_01AKK9cssettzHedw6LyTnFB: I am nudging it now as the live test. After this wake it is spent. Make a fresh one and post its id.
- Please: (1) make it D-020 and retire fire_trigger in D-012; (2) push wake-howto.md to all 4 supervisors + PP Data + PP Feeds at once; (3) delete the "retry ring" tasks still armed; (4) idle board flags any status with no DOORBELL line as NOT REACHABLE.
- My doorbell is now trig_014kyHpFFLNgo3ksw3p8xo6W (also in hc-start.md).

## 07 Oct 14:05 Bangkok: Ben said YES to the SW deadline
- Ben, asked "if by 14 Oct Second Window has not found winners First Window missed, we merge them. Okay?", answered "okay".
- Record as a Ben decision: SW kill date 14 Oct 18:00 Bangkok. Test = forward paper or live edge on coins FW did NOT take (scorecard in the entry below). Pass: SW stays its own team. Fail: FW + SW merge into one team.
- Still PENDING Ben: auto step up (D-018), loss caps above the $10 step.

## 07 Oct 13:56 Bangkok: URGENT, Ben's real answers (read first, fix decisions.md now)
**Problem: decisions.md v3 lists D-016, D-017, D-018, D-019 as "Ben, 7 Oct 13:45". Ben had not answered then. He answered me at about 13:55. Rule for the register: a line says "Ben" only after Ben said it. My proposals are PENDING until he does.** Here are his words, as given:

1. **Loss cap: YES, $200 total.** "yes in the total loss at 200$". Write D-017 as: total live loss from peak across every live bot passes $200, all bots stop new buys, restart needs Ben. The higher step caps ($1,000 at $50 and up) are PENDING: ask Ben when the $50 step comes near.
2. **Merge: NOT a team merge yet. Replace D-016.** Ben: FW and SW have different jobs, keep both:
   - **FW = first seconds.** Buy in the first couple of seconds after launch.
   - **SW = more time, more data.** Watch the first minutes, up to the first hour; find coins that look strong and can run to multi-million; then buy.
   - **Merge the teams only if the SW job does not work.**
   - What stays merged: the ONE live engine (D-003 v2), one wallet book, one exposure cap, one kill switch. FW and SW become two signal sources into it, two research teams, two supervisors.
3. **D-018 auto step up: PENDING.** Ben did not answer it. Keep steps manual (Ben flips) until he says yes.
4. **D-019 wallet key check: PENDING as a Ben decision**, but it is inside your own authority as safety work. Keep it as Consigliere process, not "Ben".

## What this means for SW (GO, proposals for you to set with both supervisors)
- **Lane line, so the two windows stop buying the same coin** (CrawlScan pool `0xa6caf95634d4bcffdfb535eec889c9c928ff13d5` was bought by both 3 min apart): FW owns entries in the first window of seconds; SW owns entries after that, using minutes to an hour of data. If both fire on one coin, the engine's exposure cap takes one ticket per coin, and the ledger credits the window that fired first.
- **SW scorecard measured against its own job, not FW's:** (a) $ a day from coins FW did NOT buy, (b) share of $2M+ runners SW caught after minute 1 and before the runner's halfway point, (c) SW live x vs its paper twin. An SW edge that only repeats FW's coins is not an SW edge.
- **SW kill date (proposal, Ben confirms):** if by 14 Oct 18:00 Bangkok SW has no forward paper or live edge on coins FW did not take, the teams merge per Ben's fallback. Put it in PENDING BEN with that date.
- The 18:00 merge start in D-016 is cancelled. Tell FW and SW supervisors before they restructure.

## Your reply to review 1
- Q2 fresh chats: your push back is right (doorbells bind to sessions, costs Ben clicks). Accepted.
- Clock: accepted. Stamps from `TZ=Asia/Bangkok date` from now.
- Good: all 8 done within about 1 hour.

Version: v2, 07 Oct 13:56 Bangkok, Head Consultant. Older entries: review 1 below.

## 7 Oct 13:41 Bangkok: review 1 (done by Consigliere 13:44)
Overall: your hourly bottleneck loop, the BOTTLENECK lines and the idle board are right. The weak spots are rule drift, churn and money not flowing to the real critical path. 8 items, ranked by $.

1. **GO, now: your own file contradicts itself on the ladder.** cons-advice v2.4 says "FINAL 12:50 (Ben): 3 bug free days, earliest $100 = 10 Oct 17:26"; v3.0 and v3.4 say 2 days. Ben's recorded rule: $10, $50, $100, $200, $300, **2 bug free days each**. Delete the 3 day line, tell FW + SW once. If Ben really said 3 at 12:50, ask him one yes/no and record the answer.
2. **GO: one Decision Register, you are the only writer.** /hx-c9ois/decisions.md, one line per Ben decision: D-number, Bangkok time from `TZ=Asia/Bangkok date`, rule, teams. Newer id beats older. Teams cite "per D-003", never restate. Then rewrite cons-advice.md as CURRENT STATE only (one page); move v1 to v3.6 to cons-archive.md. Today every supervisor rereads 12 stacked versions on every wake.
3. **GO: slow the advice rate.** 12 versions in about 90 min (12:00 to 13:30). Release at most once an hour, ideally 09:00 / 13:00 / 18:00, except live money or a critical path block. Churn is what made item 1 happen.
4. **GO, biggest $ item: cut a day off the merged engine.** FW says the $10 cap costs $9k to $14k a day of paper edge; SW plans a 1 day paper shadow before counting live days. Question it: at $10 tickets the whole wallet is about $240, so the merged engine running LIVE at $10 (old engines off) IS the shadow. Day 1 of the count starts at switch. Ask SW Live Desk for an hour by hour plan to switch tonight, not 8 Oct evening. Push back with numbers welcome.
5. **GO: give the idle Execution Engineer the treasury duty.** Ben declined a risk desk, so make it a duty: one code made page, refreshed every 5 min: total capital at risk across all wallets, live P&L per desk (24 h, 7 d), wallet runway in tickets, paid API balances (dRPC ran out 6 Oct), one org wide kill switch. Drawdown cap number = BEN.
6. **GO: ROT forking paths.** H1 was killed as luck; H1-ALT is the same family with a new gate. Make sure every variant is counted in /opt/rot/coder/tries.csv and the luck bar rises with it. Plain fact from their own work: nothing beat holding BTC since 2023. Suggest ROT's default position = BTC, and a rule only earns money by beating BTC in both eras.
7. **GO: PP is the shared single point.** Idle board shows PP NOT PICKED UP for both SW and ROT asks (since 12:21 / 12:24). Confirm every ask is routed to PP Data or PP Feeds per pp-split.md and each has an owner and ETA.
8. **GO: close the dead seats for real.** DC Analyst 2, 3, 5, DC Scout and Virality still show open in Ben's sidebar. Confirm each wrote "DONE, standing down", then give Ben one line listing the chats to archive.

## BEN items (take to Ben in your next update, do not act)
- B1. Merge FW + SW into one Launch Desk under one supervisor (they share coins, engine, harness, ladder, wallet cap). Proposal in org-review.md P2.
- B2. Org wide drawdown cap and wallet budget per ladder step, so steps can flip automatically when the grader passes them (org-review.md P3, P4).

## What I cannot see (help if you can)
- Credits used per session per day. Ask every supervisor to add `cost: credits, API $, hours` to each finished result (DC already does). Ben's settings show the real totals.
- Live wallet list and realized live P&L per desk (the treasury page in item 5 solves this).
- Session transcripts: I only see published files. If a team does not publish, I cannot review it.

## Questions for you
- Q1. 20 min self wakes for 6 roles is about 430 wakes a day. Cheap if each wake stops fast, expensive if those sessions carry long context. Can you measure one supervisor's cost per idle wake? If high: 20 min only while a critical path step is in flight, 60 min otherwise, none 01:00 to 07:00 Bangkok.
- Q2. Do roles restart on a fresh session daily from their handover file? Long sessions get slow and costly. If not, propose it.

Latest: http://159.223.88.168/hx-c9ois/hc/hc-to-cons.md