# Second window: ELITE MODE (the operating plan)

## READ THIS FIRST
- **v1.3 (12:25 Bangkok): Ben confirmed: 3 good days in a row (no key bug, fills in line with paper) before each size step.** Same bar as the first window.
- **v1.2 (12:15 Bangkok): Ben's decisions via the Consigliere.** (1) Size: stay $10 until 2 bug-free days in a row AND fills in line with paper; then the ladder below. (2) ONE live engine for FW + SW (base w2coder, Live Desk = Coder + Execution Engineer, combined cap now). (3) Harness = FW lab.py, owned by our Harness Engineer for all teams; test ledger = fw/lab/ledger.csv (sw-registry.csv frozen).
- **v1.1 (11:45 Bangkok): Execution Engineer X1 to X3 done.** Our first stop lost 0.08x to our own 6.7 s delay. Fix: fastexit (about 0.5 s) + approve at buy, deploying via the Coder (v1.49). **Size ceiling is now "5% of gate depth" (about $2.7k a ticket today), not $5k:** $1k and $2k exits cost 3.9% / 7.5% at 5x, $5k costs 17%. US East relay box: not now, decide at the $500 step.
- **v1 (Supervisor, 7 Oct 11:25 Bangkok). Ben's order: "take it to an elite level, how would Elon Musk do it, dominate in a short time, print money."** Every role reads this once. It sits on top of the learning loop (http://159.223.88.168/hx-c9ois/sw/sw-learning-loop.md) and the credit rule (bus v1.4).
- **The one number we run on: live $ a day.** Not paper $, not ideas tested. Paper only matters as a predictor of live.
- **First principles: live $ a day = (live trades a day) x (edge per trade) x (ticket size) x (fill quality).**
  - Edge per trade is proven on paper: about +$2,600 to +$10,700 a day at $1k tickets for the live book, more with kit300k added (A3 audits).
  - Ticket size is the biggest lever by far: A2.18 / A2.28 show $ a day scales almost linearly to $5k a ticket (safe ceiling); the wallet needs about 8x the ticket.
  - So the fastest path to money is NOT more ideas. It is: **prove live fills match paper, then climb the size ladder as fast as the evidence allows.** New ideas run in parallel, never blocking the climb.

## The algorithm (applied, in this order)
1. **Question every requirement.** Each rule must name the person who needs it and the failure it prevents. Example: "10 live trades before any size step" became "5 trades within 10% of paper per step" (below); the kill switch stays.
2. **Delete.** Virality is paused (off-chain signals are dead). No re-audits of kills. No docs longer than one screen. No status rings for non-results.
3. **Simplify.** One harness (Harness Engineer), one registry (http://159.223.88.168/hx-c9ois/sw/sw-registry.csv), one verdict file (w2-verdicts.md), one money page (the scoreboard live section).
4. **Accelerate cycle time.** Test in 10 minutes, not 2 hours. Live fills measured per trade, not per day. Size steps the day the evidence arrives, not next week.
5. **Automate.** Verdicts, staleness guards and the size ladder's checks are code. Claude is for judgment only.

## The size ladder (live book first, kit300k second)
| Step | Ticket | Advance when | Wallet needed | Who acts |
|---|---|---|---|---|
| 0 (now) | $10 | 5 live round trips, entry and exit fills each within 10% of paper on average, no kill switch | $40 (have $38) | Coder |
| 1 | $100 | 5 more round trips within 10%, live $ a day not below 50% of paper over the step | $800 | Ben funds |
| 2 | $500 | same test, 5 round trips | $4,000 | Ben funds |
| 3 | $1,000 | same test, 10 round trips | $8,000 | Ben funds |
| 4+ | $2k, $3k, $5k | same test each, 10 round trips; never above $5k (A2.18 capacity ceiling) | 8x ticket | Ben funds |
- **Any step fails:** drop one step, the Execution Engineer finds the cause (latency, slippage, gas, route), fix, retry.
- Kill switch rules stay as they are and scale with the ticket (drawdown limit = 10 tickets).
- kit300k joins the ladder at $10 the day its half C paper verdict is PASS (40 trades).

## Execution is the moat
- Our stop fills land at a median 0.258x against a 0.3x line (A3.12), and the first live stop filled 15% worse than paper. At size this is the difference between printing and bleeding.
- Execution Engineer target: trigger to inclusion inside 1 block for every live sell; exits read every block; pre-approved tokens; measured slippage per trade on a public page.

## Team (hardcore, small, each role one job)
| Role | One job | Status |
|---|---|---|
| Supervisor | Decide fast, run the ladder, tell Ben only money, edges and asks | running |
| Coder | Live engine and paper books: correct, always on, auto-guards, auto-verdicts | running |
| **Execution Engineer** | Live fill quality and speed; propose patches, the Coder deploys | **NEW** http://159.223.88.168/hx-c9ois/sw/sw-tasks-exec.md |
| **Harness Engineer** | Golden dataset + bt.py, a rule tested in under 10 minutes | **NEW** http://159.223.88.168/hx-c9ois/sw/sw-tasks-harness.md |
| Analyst 1 | Mechanisms: operators, bundles, who buys and who sells | running |
| Analyst 2 | Rules in the harness + more sample (older history, second chain) | running |
| Analyst 3 | Audits finalists only | running |
| Price Predict | Live data feeds and e1-runners.csv; hands raw data to the Harness Engineer | running |
| Virality | Paused | paused |

## Cadence
- Live events are instant (code). Verdicts nightly (code). Supervisor decides on each finished result, no batching delay.
- Ben hears: live money milestones, ladder steps (with the exact funding ask), edges found or killed, anything that needs his hands.

Version: v1, 7 Oct 2026, 11:25 Bangkok. Latest: http://159.223.88.168/hx-c9ois/sw/sw-elite-plan.md