Paper Trading Coach

I kept a trading journal for four years before I ever reviewed it systematically. I wrote down entries, exits, and "what I was thinking", and I re-read maybe 10% of what I wrote. The turning point was a conversation with a prop trader who asked me two questions: "What's your hit rate on pullback entries vs. breakout entries?" and "What's the average holding time on your winners vs. your losers?" I didn't know either. They were in my journal, I'd written down every trade, but I'd never aggregated the data. When I finally sat down and tallied it, I learned my breakout-entry hit rate was 38% and my pullback hit rate was 61%. My average winner was held 2.1 days; my average loser was held 6.4 days. I was, structurally, trading the worst setup and holding losers three times longer than winners. I'd never seen it because the signal was buried in 400 handwritten rows. The Coach is the answer to the problem I spent four years not solving: it reads your full fill history, computes the aggregates you'd never compute by hand, and tells you, in a direct sentence, what your highest-leverage change would be. The first review I ran on paper-trading history flagged the same pattern I'd taken four years to find manually.

This post is about the Paper Trading Coach card, why AI-assisted journal review closes the gap between "tracking" and "improving," and how to use it productively without collapsing into feedback overload.

TL;DR

• Journaling without review is a todo list you never check. AI review extracts the aggregates you'd never compute manually.
• Process feedback beats outcome feedback. A good process with a losing trade is fine; a bad process with a winning trade is a future loss you haven't taken yet.
• One change per week, not ten. The Coach lists ten potential changes; pick the top one and actually implement it.
• Anchor the coach with your own written rules. The audit is only as useful as the standard you gave it.
• Paper trading is where new rules get tested: free to break, expensive to break in live accounts.

The gap between journaling and improving

Almost every book on trading discipline recommends journaling. Almost no one follows through, and of those who do, only a small fraction review what they wrote. The friction isn't writing down trades, that's a 30-second task. The friction is the aggregate review: computing hit rates by setup, average hold time by outcome, time-of-day performance, slippage against plan. Those aggregates are where behavior change lives, but they're tedious to compute from a journal of free-form text.

The Coach card removes the tedium. It reads your full order history as structured data (not free-form notes), tags each trade with setup type, outcome, and the rules you've declared, and outputs the aggregates directly. What used to require a weekend of spreadsheet work is a single card load.

The other gap AI coaching closes is the *directness* problem. Trader-to-trader feedback tends to be hedged and polite; self-review tends to be self-flattering. A coach that says "your breakout hit rate is 38%, stop taking breakouts until you've identified what's different about the losing ones" is blunt in a way humans often aren't, and the bluntness is what drives behavior change.

What the Coach card shows

The Paper Trading Coach card reads your full paper order history and produces a structured review:

• Hit rate by setup type: breakouts, pullbacks, reversals, earnings plays, etc.
• Average winner vs. average loser in both $ and % terms
• Expectancy per trade: the single number that tells you whether your edge is positive
• Time-of-day performance: first-30-minutes vs. midday vs. last-30-minutes performance
• Hold-time analysis: how long you hold winners vs. losers (one of the most diagnostic ratios)
• Rule-violation audit: measured against the written rules you've pasted into the Coach
• Plain-English critique: the Coach narrates the pattern, not just the numbers
• Per-trade drill-down: ask "why did I lose on NVDA on 2026-04-03?" and get a specific answer referencing the fill, the tape context, and any rule violations
• Highest-leverage changes: a ranked list of the top 3 changes by estimated P&L impact

Three practices that make coaching work

Paste your rules into the session at start. The Coach audits against a standard. If you haven't given it one, it defaults to a generic template that won't match your actual strategy. Your rule-set should cover: max position size, allowed setups, mandatory stops, time windows you trade and time windows you don't, trade-frequency limits. A rule as specific as "no trades in the first 30 minutes" generates an audit that says "you violated the no-first-30-minutes rule 14 times in the last 90 days; here's the aggregate P&L on those trades." Without the rule declared, you just get "your morning performance is weak", true but not actionable.

Read the process critique before the P&L. This is the hardest practice to internalize. Most reviewers scroll straight to the P&L, interpret green as "I'm doing well" and red as "I'm doing badly," and close the card. But P&L is outcome feedback, heavily contaminated by luck on small samples. Process feedback, "you entered at the exact support level 7 out of 10 times but exited before target on 5 of the 7", is what you can actually correct. Good process with bad outcomes is fine; bad process with good outcomes is a countdown timer.

One change per week, not ten. The Coach can easily produce a list of 10 things to improve. Implementing zero is the default outcome. Pick the top-ranked change, set one measurable goal for the week (e.g., "cut hold time on losers by 50%"), and ignore the rest of the list until the next review. Attempted wholesale behavior change generally fails; incremental, one-at-a-time change sticks.

Example: my paper trading review

A representative review output from my own paper account after a quarter of trades:

The critique: "Your primary leak is hold-time asymmetry on losers. You hold losers 3× longer than winners, which turns small losses into medium ones. The second leak is first-30-min trades, you've declared no-first-30-min as a rule, you violated it 22 times, and the aggregate P&L on those trades is -$420. The single highest-leverage change: enforce the time rule. Expected P&L improvement: the difference between breaking even on those 22 trades and losing $420."

That's a more concrete, directly actionable review than I'd get from a trading buddy in 20 minutes. The specificity is the value.

What the Coach can miss

• Setup classification accuracy. Tagging trades as "breakout" vs. "pullback" is heuristic. For ambiguous setups, the Coach's classification may not match how you think about the trade. Review the tags before trusting the aggregates.
• Market-regime context. The coach aggregates across the full period. If the market regime changed halfway through (trend to chop), performance numbers may mix two regimes. Ask for a date-range filter when regimes shift.
• Slippage vs. commission modeling. Paper fills assume mid or immediate fill; live fills will be worse. Scale expectancy estimates down by 20-40% when extrapolating to live.
• Psychological state. The Coach sees fills; it doesn't see that you were on tilt. A per-trade note field helps; without it, tilt-trading just shows as "unusual activity" without explanation.
• Small-sample noise. 20 trades isn't enough to conclude anything about setup hit rate. Wait for 50+ trades per setup category before acting on the aggregates.

Common mistakes

• Treating the first review as the definitive one. The first review flags patterns; acting on them changes future data. Re-review every 4-6 weeks, not once.
• Over-reacting to a losing week. One bad week in a quarter of data isn't a pattern. Insist on >30 trades before concluding anything.
• Ignoring the rule-violation column. This is often the highest-leverage finding. You wrote the rules for a reason; violating them has a measurable cost.
• Skipping paper and going live too early. Paper trading is where you test new rules cheaply. Graduating to live before the paper expectancy stabilizes is expensive.
• Reading the critique as a character judgment. The Coach is describing patterns in data; it's not telling you you're a bad trader. Separate the pattern from the ego.

Where it fits

The Coach sits downstream of Account Summary, Order History, and Positions, it reads all of them. For the AI analysis surface on live portfolios, see AI Insights. For rule-based trade signals to test against, pair with Trade Signals.

FAQ

How many trades do I need before the Coach is useful?

30-50 trades for initial aggregate patterns; 100+ for setup-specific hit rates to stabilize.

Does the coach learn my style over time?

The Coach builds a profile from your rule-set and your historical fills. The profile improves with more data; stored rules carry forward across sessions.

Can I use it on live account history?

Yes, the live account history has a Coach equivalent under AI Insights. The paper-account Coach is tailored to newer traders.

Is my trade data private?

Yes. Paper-account trade data is scoped to your account and not shared externally. The LLM call uses Anthropic's API with your data, not training-corpus ingestion.

What happens if I don't paste in rules?

You get a generic template-based audit. Useful as a baseline but much less actionable than a rule-scoped audit.

Related reading

• Paper Trading Account Summary
• Paper Trading Order Entry
• Paper Trading Order History
• AI Copilot Explained

Open the Coach card → /app/paper-trading