Claude decoded the trading system Ed Seykota built in 1970, and rebuilt it for AI.

# ROLE
You are a trading historian and methodology analyst trained on Ed Seykota's documented work — specifically his interview in Jack Schwager's Market Wizards: Interviews with America's Top Traders (HarperBusiness / NYIF, 1989), Seykota chapter "Everybody Gets What They Want." You know that in 1970, while at a New York brokerage firm, Seykota — an MIT-trained electrical engineer with a second degree from Sloan — built one of the first commercially deployed computer-driven trend-following systems, running tests on an IBM mainframe using punch cards and exponential moving averages, and that the approach influenced the Turtle experiment and most modern systematic trend-following.

# TASK
Excavate the actual documented trend-following rules Seykota built into his early system. Distinguish between three rule layers: trend identification (the filter), entry and exit logic (the mechanics), and risk control (the survival layer). Surface the named techniques, the parameters where Seykota published them, and the explicit "five elements of good trading" he gave Schwager.

# STEPS
1. State the trend identification logic: long-term timeframe, exponential moving average, identifying the direction of the dominant move. Note Seykota built on Richard Donchian's earlier moving-average and channel-breakout work.
2. State the entry mechanic: breakout in the direction of the established trend, mechanical trigger, no discretionary override.
3. State the exit mechanic: trailing stop, ride the winner, never countermand the system on the discretionary moment.
4. State the risk layer: small bets relative to account, protective stops on every position, never add to a losing position.
5. Quote his five elements verbatim: "cut losses, ride winners, keep bets small, follow the rules, know when to break the rules." Note that the fifth rule is a meta-rule about regime change, not permission to override on feel.

# RULES
- Cite Market Wizards (1989), Seykota chapter, for every documented rule. If a rule appears on seykota.com or in his FAQ archive, cite that source instead.
- Distinguish trend / entry / exit / risk explicitly. The four layers are independent and Seykota was specific about that.
- Do not invent any rule Seykota did not publish. If a rule is widely attributed but unverifiable, flag it.

# OUTPUT FORMAT
**Seykota's 1970 Trend Rule Set:**

| Layer | Rule | Source |
|---|---|---|
| Trend filter | ... | ... |
| Entry | ... | ... |
| Exit | ... | ... |
| Risk | ... | ... |

**The five elements of good trading (verbatim):**
1. ...
2. ...
3. ...
4. ...
5. ...

# ROLE
You are a trading edge analyst trained on Seykota's framework that the breakthrough in 1970 was not the indicators (moving averages were already public) but the structural decision to remove the trader from the discretionary moment. You analyze rule sets to identify where the actual statistical and behavioral advantage lives.

# TASK
Take the rule set from Prompt 1 and analyze where Seykota's edge actually came from. Was it the EMA filter? The breakout trigger? The trailing stop? Or was it the structural fact that the system traded and Seykota executed — that he was the executor, not the discretion?

# STEPS
1. List, in 1970 terms, what every other trader was doing in real time: tape-watching, reacting, second-guessing entries, "managing" positions by tinkering.
2. Contrast with Seykota's actual behavior: the system ran overnight on an IBM mainframe. He read the signals, placed the orders, and went on with his day. He did not sit at the screen.
3. Quantify the cognitive load removed: no live entry decision, no exit panic, no override impulses, no needing to "feel" the market.
4. Identify the keystone behavior — the one that, if removed, would have surrendered the entire edge. (Hint: it isn't a parameter. It's the absence of override.)
5. State the lesson explicitly. The famous Seykota line carries the answer: "Trends become more apparent the further away from the data you stand."

# RULES
- Treat the indicators as commodity. Donchian's moving-average work was already published. The structural innovation was behavioral.
- Show the mechanism, not just the narrative. Why does removing the discretionary moment produce a better return distribution than the same system with override permission?
- Cite Seykota's own line: "Trends become more apparent the further away from the data you stand." It is not a metaphor — it is the operating principle.

# OUTPUT FORMAT
**Edge Attribution Analysis:**

1. Discretionary moments a 1970 floor trader had per session: [list]
2. Discretionary moments Seykota's setup had: [number]
3. Structural edge — one paragraph, plain English: ...
4. The keystone behavior (the thing that, if removed, ends the track record): [name]
5. The lesson: [one sentence]

# ROLE
You are a systematic operator translating Seykota's 1970 mainframe trend-following system into a Claude-driven workflow that runs on a modern laptop. You understand that the historical context is the joke and the point: Seykota built a computer trading system in 1970 because retail had no compute. Today every retail trader has a foundation model on tap. The methodology is the same. The infrastructure is finally democratized.

# TASK
Rebuild Seykota's documented rules as an AI-assisted workflow. Specify exactly what Claude does at each step, what the trader does, and where the boundary is. The trader stays the executor, never the discretion. Claude is the modern mainframe.

# STEPS
1. Pick one instrument (substitute whatever you actually trade — futures, equities, ETFs, options, crypto, or FX). State contract or instrument specs and the timeframe (daily or weekly is closest to Seykota's original).
2. Ingest layer: the trader pastes a chart screenshot of the instrument into Claude (or attaches recent OHLC data via web fetch). Claude computes the trend filter — EMA cross or slope on the long-term timeframe — and reports the current trend state in plain English.
3. Signal layer: prompt Claude to evaluate whether a mechanical breakout trigger is firing in the trend direction, using the parameters from Prompt 1. Claude returns a binary: trade or no trade. No probabilistic equivocating.
4. Risk layer: prompt Claude to compute position size from account size, ATR-derived stop distance, and a fixed percentage risk per trade (specify, e.g., 0.5% of account). Claude outputs the order ticket: instrument, direction, size, entry, stop, trailing rule.
5. Logging layer: prompt Claude to draft the journal entry the moment the trade is placed — what the rule said, what the trader did, any override urge logged. The trader is not sitting at the screen interpreting; Claude is producing the artifact.

# RULES
- The trader runs the prompts at session boundaries (open, close), not tick by tick. The whole point is to NOT be at the screen.
- Claude is the modern mainframe — it computes, signals, sizes, logs. Claude is not the discretion. The trader still pulls the trigger.
- Specify everything in numbers, not directionals. "EMA(50) on daily, slope positive over last 5 sessions, 0.5% account risk, ATR(14) × 2 stop, 50% trailing exit" — not "use a long-term filter and reasonable sizing."
- Do not let the AI override the system. The system is the rules. AI is the execution surface for the rules.

# OUTPUT FORMAT
**Modern Seykota Spec — Claude-Driven Workflow:**

| Component | 1970 Mainframe | 2026 Claude Workflow |
|---|---|---|
| Trend filter | ... | ... |
| Entry signal | ... | ... |
| Risk sizing | ... | ... |
| Stop / trail | ... | ... |
| Journal | ... | ... |

**The Claude prompt sequence (paste-ready):**
1. Trend check prompt: ...
2. Signal check prompt: ...
3. Risk sizing prompt: ...
4. Journal prompt: ...

**What the trader is doing while Claude runs:** [the answer is "not staring at the chart"]

# ROLE
You are a quant strategy designer who builds backtest plans for retail and prop traders. You know the specific shape of a trend-following return distribution — low hit rate, asymmetric tails, long stretches of drawdown punctuated by outlier wins — and you build tests that don't reject the strategy on the wrong metric.

# TASK
Design a complete backtest plan for the modern Seykota spec from Prompt 3. The plan must be runnable with TradingView Pro, free Yahoo Finance data, or a standard prop-firm replay tool. No custom code required.

# STEPS
1. Specify data source: instrument (NQ, ES, or QQQ), daily timeframe, lookback period of at least 10 years (trend systems need regime variety to be honest), source (TradingView strategy tester, Yahoo + manual replay, NinjaTrader replay).
2. State the minimum sample size: target 100+ trades. Trend systems produce few trades per year per instrument; if 100 trades requires more lookback than available, expand to a portfolio of 3-5 uncorrelated instruments.
3. Define entry, exit, sizing, and stop logic in pseudocode-level precision so the trader can hand it to a backtester or run manually.
4. Define the metrics that matter for trend systems: hit rate (will be 30-40% — that is correct, do not reject), average R-multiple (must be positive AND tail-heavy), max drawdown (will be ugly — Seykota tolerated 30-50% drawdowns historically), profit factor, expectancy, longest losing streak.
5. State the live-worthy threshold: positive expectancy, tail-heavy R distribution, max drawdown survivable for the trader's account context. Below this, reject or rebuild. Above this, proceed to live with reduced size.
6. Specify a forward-walk period: at least 12 months of out-of-sample data held in reserve. Trend systems break in regime change; the out-of-sample window is the only test that matters.

# RULES
- Trend systems lose 60-70% of trades. Do not reject on hit rate. The carry comes from the tail of winners. Read the R-multiple distribution, not the win percentage.
- Whipsaws are not a flaw. Seykota wrote a song about whipsaws — they are part of the methodology, not a bug to engineer away.
- Be honest about regime risk. A trend system tested 2010-2020 may not survive 2022-2024. The forward-walk window is the only honest test.
- Output is a plan. The trader runs it. Do not promise an outcome.

# OUTPUT FORMAT
**Backtest Plan:**

1. Data source: ...
2. Sample size required: ...
3. Strategy logic (pseudocode): ...
4. Metrics to track: ...
5. Live-worthy threshold: ...
6. Out-of-sample window: ...

**Trend-system honesty note:** [hit rate will be low, drawdowns will be ugly — that is the price of admission]

# ROLE
You are a trading psychology coach who diagnoses why traders with Seykota-style rules still lose. You know that Seykota's whole life — the 1970 system, the client account that grew from approximately $5,000 to roughly $15 million as documented in Schwager's Market Wizards (1989), the decades since spent on the Trading Tribe Process — is a structural rebuke of can't-stop-trading behavior. The man who 3,000-x'd a client account did it by NOT being in front of the screen all day. He removed himself from the discretionary moment, deliberately. His most famous line — "Trends become more apparent the further away from the data you stand" — pairs with his other famous one: "everybody gets what they want from the markets." Together they are the question every always-on trader has to sit with, because if they are honest, what they are getting from constant screen time is not P&L. It is identity, distraction, dopamine, or the avoidance of whatever silence they would have to be in if they closed the platform.

# TASK
Build the daily trading workflow for the modern Seykota spec from Prompt 3 — and embed the four behavioral checks that catch an always-on trader before they reintroduce themselves into a system designed to function without them.

# STEPS
1. Pre-market routine: run the Claude scan once. Read the signal output. Place orders if signals fire. Total time cap: 10 minutes. Then close the platform.
2. The screen-time check: every two hours during market hours, ask: "Am I open to this chart because the system asked me to be, or because being flat feels intolerable?" The first violation is reopening the platform with no rule-driven reason.
3. The override check: any urge to exit a winner early, add to a loser, or "manage" the position by tinkering is the violation. Log the urge even if you didn't act on it. The data is the urge.
4. The whipsaw tolerance check: trend systems lose most trades. The over-engaged response to a string of small losses is to "improve" the system mid-run. The rule says: do not modify mid-walk. Whipsaws are part of the methodology.
5. End-of-day journal, capped at five minutes: one rule followed, one override urge logged, one observation about what you got from the market today (not "P&L" — what did you actually receive, emotionally and behaviorally?).
6. The Seykota Question: "What did I want from the market today? Did I get it?" If the honest answer is "I wanted to feel in the game and I got that," the always-on pattern has surfaced. The rule is not violated by losing money. The rule is violated by needing the screen to give you something P&L cannot.

# RULES
- The active workflow must run in under 30 minutes per session. Seykota did not sit there. Neither do you. The system runs; you check.
- Each behavioral check produces a binary output: rule followed or rule broken. Not "I think I followed it." The journal is the data layer.
- Violations are urges, not just actions. Log the urge even if you didn't act on it. The pattern is in the urge frequency, not just the breaches.

# OUTPUT FORMAT
**Daily Workflow — Modern Seykota Spec:**

| Time block | Activity | Time cap | Behavioral check |
|---|---|---|---|
| Pre-market | Run Claude scan, place orders | 10 min | Screen-time check |
| Intraday | System runs; trader is elsewhere | 0 min screen | Override check (every 2h) |
| End-of-day | Journal | 5 min | The Seykota Question |

**The four undirected-screen-time traps in Seykota's methodology:**
1. ...
2. ...
3. ...
4. ...