# First Window ELITE PLAN (FW Supervisor, 7 Oct 2026, 11:45 Bangkok)

## Read this first
- Ben's order (7 Oct 11:30 Bangkok): take the first window to elite level, run it the way Elon Musk would, fully solve it and print money. Any team members and resources allowed.
- This page is the operating system for FW from now. It sits on top of the learning playbook (http://159.223.88.168/hx-c9ois/fw/learning-playbook.md), which stays in force for how ideas are tested.
- One number runs the team: **live $ per day, net of every cost.** Every task must say which term of the equation below it moves, and by how much.

## 1. The physics (the equation that must be true)
**Live $ a day = trades a day x ticket size x (mean x - 1 - costs)**

| Term | Now (live) | Proven ceiling (paper, honest fills) | Gap | Lever |
|---|---|---|---|---|
| Ticket size | $10 | $1k still 1.69x mean on half B (real depth walked) | **100x** | Ramp. Biggest lever by far. |
| Trades a day | about 7 (5 tries in 18 h: 4 closed, 1 failed) | about 13 a day mc5 half B (146 in 11 days) | 2x to 3x | Zero missed trades, then more books. |
| Mean x | 7.62x on 4 trades (one 30x runner) | 1.84x at $100, 1.69x at $1k, half B | small n | Exits and entries, through lab.py only. |
| Costs | fill/model 1.0000 on all 8 legs | 1% fee + gas each way modelled | none seen | Keep proving at every new size. |

- At $10 the whole edge is worth about $100 a day. At $1k the same edge is about +$9,195 a day on paper (half B, lab.py). **Everything else we do is worth less than getting size right, safely.**
- The edge is fat tailed: 79% of trades lose (mean loser about 0.5 to 0.6 ticket), a few runners pay for all. So 10 live trades can prove that FILLS match paper, never that the MEAN is right. The mean is proven on paper (n 146, half B, random 95th, half C forward). The ramp must test fills, not luck.

## 2. Musk's five steps, applied
1. **Question every requirement.** The "10 clean trades at $10 before $100" gate was set to prove execution. Execution is already proven on 8 of 8 legs (fill/model 1.0000, buy delay 1 to 4 s, stop SELL_SENT 1.4 s). The only leg not yet seen live is the pull exit. A $10 ticket cannot teach us anything about $1k depth. So the gate becomes **fidelity per size step**, below.
2. **Delete.** No new test on a dead family (ledger KILLED rows: entry checkpoint, cap band, stop level, trails, dead coin exit, hold past 4 h, drip exit, X pulse, fomo pulse, bot router share, late bloomers, re-entry). No status rewrite when nothing changed. One ledger for all windows: http://159.223.88.168/hx-c9ois/fw/lab/ledger.csv .
3. **Simplify.** One harness (lab.py), one live book, one scoreboard, one ramp ladder. Numbers from lab.py only.
4. **Speed up the loop.** An idea goes from card to verdict in one wake (target under 60 min). Overnight, scripts on research-box search the queue's parameter grids; Claude only judges the top result.
5. **Automate.** Watching, reconciling, post-mortems, decay checks and the scoreboard are server scripts. Claude wakes only for a judgment.

## 3. The ramp ladder
- **FINAL (Ben via Consigliere v3.0, 7 Oct about 12:30 Bangkok): ladder $10, $50, $100, $200, $300; 2 bug-free days in a row per size; days count only on the merged FW + SW engine after its 1 day shadow. Lines below are older and kept for the bars only.**

### Older version (Ben's rule, CHANGED 7 Oct 12:30 Bangkok: 2 bug-free days in a row at each size before the next)
- **Ben decided: stay at $10 now.** Each size runs until 2 bug-free days in a row (no key bug, fills in line with paper); then Ben flips the next size. A found bug gets fixed and logged; the bug-free count starts after the fix. Ladder: $10, $100, $300, $1k.
- Days run from LIVE_ON time (17:26 Bangkok). $10 day 1 = 6 Oct 17:26 to 7 Oct 17:26. Day 1 had bugs fixed today (pull kill, late pools), so it does not count. Earliest $100 = 9 Oct 17:26 Bangkok, if days 7 to 8 Oct and 8 to 9 Oct are bug free.
- **A successful day (Analyst 2 grades it every day at 17:30 Bangkok):** (1) bot up all day, kill switch not fired; (2) every paper mc5 trade was a live attempt, or has a named, fixed reason (A3 recon); (3) every buy gap under 3% (5% at $10 because of gas), every stop slip within paper priced at the live sell block; (4) live x within 10% of its paper twin on every closed trade.
- Profit on the day is shown, but it is NOT a pass bar: 79% of trades lose and most days have no runner, so a red day with perfect fills is still a successful day. A day that fails any bar resets the count to 0.
- **Kill switch at every size:** drawdown over 20 tickets, or 2 sells in a row more than 10% under paper, stops new buys at once.
- Wallet needed at each size = 24 tickets + gas: $10 $242, $100 about $2,450, $300 about $7,300, $1k about $24,200. Above $1k only after a $3k depth test.
- Two windows hold the same coins (CrawlScan pool 0xa6caf95634d4bcffdfb535eec889c9c928ff13d5 was bought by FW and SW 3 min apart). Before $300, one combined exposure cap across FW and SW.

## 4. Team (lean; no new Claude seats needed now)
| Role | Owns | Moves which term |
|---|---|---|
| Supervisor | this plan, ramp asks to Ben, daily review 09:00 Bangkok | all |
| Analyst 1 | idea queue + lab.py tests + overnight grid search scripts | mean x, trades a day |
| Analyst 2 | red team + ramp gate grading per step + combined exposure cap | size (safely) |
| Analyst 3 | lab.py, live audit, **trade count reconciliation (every paper mc5 trade must be a live trade or a named reason)**, scoreboard | trades a day, costs |
| Price Predict | the live bot: size steps, per-size slippage caps, cross-window exposure cap, wallet need on the board | size, costs |

## 5. One scoreboard (built by A3, server cron, phone and desktop)
- http://159.223.88.168/hx-c9ois/fw/lab/scoreboard.html : live $ a day (last 24 h, 7 d), trades a day live vs paper (misses named), mean x live vs paper twin, fill gaps, ramp step and gate count, wallet runway, open ideas by stage.
- Live P&L for Ben stays at http://159.223.88.168/hx-c9ois/fw/pnl.html .

## 6. What Ben's hands are needed for
- Nothing now. Stay at $10. When Analyst 2 posts 2 bug-free days at a size, I tell Ben, and he flips the next size and tops up the wallet 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286 for it.

## 7. Honesty bars (never traded for speed)
- Pre-register every test. Half A picks, half B judges, random 95th, top 3 removed, n of at least 30. Live is graded against paper priced at the live sell block.
- A wrong call gets corrected in the ledger with the number that misled us.

Version: v3 (ramp = 2 bug-free days per size, Ben 12:30), 7 Oct 2026 11:45 Bangkok, FW Supervisor. Latest: http://159.223.88.168/hx-c9ois/fw/elite-plan.md