# ELITE MODE: how the ROT team runs from now (v1.1, 7 Oct 2026, 12:55 Bangkok)

## Read this first (every role, every wake)
- Ben's order: run this the way Elon Musk would. Understand it to the bottom, then dominate fast. Goal: solve rotations and print money.
- This file OVERRULES older files where they clash. speedrun.md and learning-loop.md still apply.

## 1. Know the physics limit first (the "best possible" number)
- Musk starts from the limit set by physics, then asks why we are far from it.
- Our limit = the PERFECT HINDSIGHT ROTATION: every week, the money sits in the coin (with real liquidity) that went up the most that week. Also the "realistic perfect": top 5 coins of each month, held from the low to the high.
- Coder builds this in the time machine (the oracle). Every rule is then scored as CAPTURE % = our return / oracle return, per month.
- The gap tells us where the money is lost: wrong place, late entry, early exit, too small, or too slow. That is the to do list, ranked by dollars lost.

## 2. Musk's 5 step algorithm, applied
1. QUESTION every requirement. Do we need sectors and baskets at all, or just the strongest liquid coins? Do we need a macro light, or does BTC trend alone do it? Each requirement must prove it adds money in the time machine, or it goes.
2. DELETE. Anything that does not move capture % gets cut: dashboard blocks, signals, roles, reports. If we are not adding back 10% of what we delete, we are not deleting enough.
3. SIMPLIFY. One engine (time machine), one score (capture %), one page (map.html), one bus.
4. ACCELERATE. Idea to verdict in under 24 h. Many variants in parallel.
5. AUTOMATE, last. Only after a rule is proven: it trades on paper by itself, then live by itself.

## 3. The auto research machine (the factory that builds the product)
- Coder builds a search loop on research-box: it takes building blocks (entry, exit, size, universe, filter), makes thousands of combinations, runs them through the time machine overnight, and ranks them by capture % on the training years only.
- The top 10 go to the Auditor's luck test and fragility test, then to the locked holdout (one shot). Survivors go to paper.
- Claude roles design building blocks and read results. Machines do the grinding. This is the 100x speed step.
- Guard: the more variants we try, the higher the luck bar. The Auditor's 5 gate luck bar is LAW: http://159.223.88.168/hx-c9ois/rot/auditor-status.md (count every try in /opt/rot/coder/tries.csv, deflated Sharpe 0.95+, whole search shuffle, per rule shuffle, one shot holdout).
- v1.1 (12:55, Auditor push back ACCEPTED): search in FAMILIES, not blind knob turning. Each family starts from a reason (a mechanism), is registered before it runs, and has at most 200 variants. 7 years is about 2 cycles; thousands of blind variants mostly find luck.

## 4. Fly the rocket early
- Paper is not reality. As soon as one rule passes the Auditor and the holdout, it trades LIVE with a small fixed ticket (Ben decides the amount and the wallet). Real fills, real slippage, real behavior teach what no backtest can.
- Live and paper run side by side. Any gap over 2% per trade gets a cause within a day.

## 5. Hardcore cadence
- Every role owns ONE number on the scoreboard and reports it each finished task:
  - A1: share of history covered, dead coins included.
  - A2: lead signals tested and killed or passed.
  - A3: v3 decay explained in %.
  - Winners: separators found that beat controls in both eras.
  - Sizer: capture % added by mechanics alone.
  - Coder: time machine speed (rules per hour) and oracle gap.
  - Auditor: rules passed and killed.
  - Dash: page shows the champion call and capture % live.
- Hard dates:
  - Day 2 (9 Oct): oracle built, capture % of basket v3 known.
  - Day 4 (11 Oct): auto research running overnight.
  - Day 7 (14 Oct): champion v1 beats basket v3 by 2x on capture % in BOTH eras and passes the holdout, or we say plainly why not.
  - Day 10 (17 Oct): first live trade with a small ticket, if Ben says yes.
- Weekly: the role that moved its number least gets new orders or is merged.

## 6. Resources
- Compute: if the auto research loop is slower than 1,000 rule runs per night on research-box, the Coder asks for a bigger research box (32 CPU, about $250 to $400 a month). Ask with numbers.
- Data: if survivor free history is missing for a chain or era, A1 asks for the cheapest paid source that fills it, with price.

Version: v1.1, 7 Oct 2026 12:55 Bangkok. Latest: http://159.223.88.168/hx-c9ois/rot/elite.md

- 07 Oct 12:29: live gate replaced (Ben via Consigliere v3.0): live small only after 20 scored paper calls that beat holding BTC (earliest about 27 Oct).