15 Questions to Ask Your AI Trading Coach Right Now

A strong AI trading-coach question names the evidence, comparison, and decision it should support. “How am I doing?” asks for a story. “Within this account and setup version, where did planned risk first diverge from actual execution?” asks for an inspectable answer.

Trader’s Second Brain Coach is powerful because it can work across the selected journal evidence instead of relying on a trader’s memory. The discipline is equally important: canonical metrics stay canonical, missing fields stay visible, and the model must not invent psychology, causality, future results, or numbers outside the evidence set.

The Evidence Contract Behind Every Useful Question

Before choosing a prompt, set the scope:

• Account: one account, a named group, personal versus evaluation, or all reconciled accounts.
• Window: exact dates and whether open trades, deposits, withdrawals, resets, or incomplete sessions are included.
• Strategy version: the setup and rule version that was actually active at the time.
• Unit: order, fill, trade, position, signal, session, or account-day.
• Outcome: net P/L, R multiple, expectancy, drawdown, compliance, or another explicitly defined measure.
• Evidence gaps: which missing tags, plans, screenshots, or source records should limit the verdict.

The AI-for-trading workflow explains why a narrow evidence question is more useful than asking a model to predict a market or certify a strategy.

Questions 1–3: Establish Whether the Data Can Answer

1. “For this account and date range, which imported records, fees, deposits, withdrawals, time zones, and strategy tags are missing or unresolved?”
   Use this first. A precise analysis over an incomplete ledger is still unreliable.
2. “Which trades cannot be assigned confidently to the selected setup version, and how would excluding them change the cohort size and coverage?”
   This separates a clean comparison from a mixed strategy history.
3. “Show the source rows and definitions behind every metric you will use; which values are canonical and which are derived?”
   This prevents a plausible explanation from silently becoming a new calculation.

Questions 4–6: Inspect Signal Selection

4. “Among eligible signals with complete records, which were taken, skipped, delayed, or added, split by setup and session?”
   Selection is part of performance. A trade-only ledger cannot prove what was skipped unless signals were recorded.
5. “Compare the selected setup with the same setup in the prior untouched window. Which context fields differ enough to make the cohorts non-comparable?”
   Ask for confounds before asking whether the edge changed.
6. “Which setup contributes the largest verified net edge after costs, and how concentrated is that result in a few trades, days, symbols, or regimes?”
   A ranked result needs coverage and concentration beside the headline metric.

Questions 7–9: Find the First Execution Divergence

7. “Where did planned and actual entry, size, stop, target, hold time, or order type first diverge for this setup?”
   The earliest verified divergence usually gives a more actionable review than the most dramatic loss.
8. “After fills, spread, commission, financing, and partial executions are reconciled, which execution differences remain material?”
   Do not diagnose behavior while costs or order state are unresolved.
9. “Which repeated manual overrides have contemporaneous reasons, and which appear only as unlabeled record changes?”
   Coach may describe the sequence; it should not invent the trader’s motive.

Use the setup-failure analysis when the review must separate data, strategy, execution, and rule-design faults.

Questions 10–12: Audit Risk and Account Constraints

10. “Which decisions used more planned loss or account-loss capacity than the frozen rule allowed, and what evidence supports each flag?”
    Use the exact account and rule version; do not apply a current rule to an older trade.
11. “Where did simultaneous or correlated positions create larger aggregate exposure than the trade-level risk labels show?”
    Separate the individual trade from the portfolio path.
12. “After each loss or win streak, did size, frequency, setup selection, or stop behavior change—and does the record support an association after context is matched?”
    This asks for observable behavior without diagnosing revenge, confidence, fear, or causality.

Questions 13–14: Test a Change Without Hindsight

13. “If this exact rule had been frozen before the window, which trades would it include or exclude, and what assumptions make the counterfactual uncertain?”
    Require the rule, data granularity, costs, and unresolved intrabar order sequence to be explicit.
14. “Which single controllable change addresses the earliest verified failure, and what metric, observation window, stop condition, and rollback would test it?”
    A narrow experiment is easier to audit than a wholesale strategy rewrite.

The question should never ask Coach to optimize repeatedly on the same outcomes and then present the winner as validation. Preserve an untouched or future window and predeclare what would count as failure.

Question 15: Produce a Traceable Review

“Create a report card for data integrity, signal selection, execution, net edge evidence, and risk containment. For every grade, cite the evidence, coverage, exceptions, and the next reversible test. Mark any unsupported dimension ‘not enough evidence.’”

This is stronger than requesting a motivational summary. The grade is a navigation layer into records, not a credential or forecast. Use the monthly review template to keep the evidence window and decision cadence consistent.

Turn Vague Prompts Into Evidence Questions

Audit the Answer Before Acting

A useful response should survive a five-part audit. First, confirm the scope displayed by the product matches the account, window, and setup you intended. Second, open the cited records and verify that the labels, costs, and order states mean what the answer says they mean. Third, separate observation from inference: “size increased after losses” may be visible; “the trader was angry” is not established by the sequence. Fourth, check alternative explanations such as a strategy release, instrument change, volatility shift, missing signals, or an account-rule reset. Fifth, write the smallest decision that the evidence actually supports.

Numbers need the same discipline. Ask whether the metric is raw or normalized, gross or net, trade- or day-weighted, and whether deposits, withdrawals, fees, partial fills, and currencies are reconciled. Inspect the denominator and coverage dates. If only twelve of forty records have a setup tag, a precise setup ranking should remain qualified even when the arithmetic on those twelve is correct.

For a proposed change, require a compact experiment card:

• the exact rule and strategy version;
• the eligible future decisions and excluded cases;
• one primary process metric plus downside and cost checks;
• the minimum evidence conditions—not a universal trade-count promise;
• the stop, rollback, and next review date.

Do not let the answer rewrite history. Retrospective labels may be useful for cleanup, but a setup, emotion, or plan tag added after seeing the outcome is not contemporaneous evidence. Keep it marked as reconstructed and test the rule prospectively.

Archive the answer with the evidence-set identifier and product version. If a later import repair, currency conversion, or account reconciliation changes the canonical rows, re-run the question and compare revisions. The old answer was valid only for its recorded inputs; it should not survive silently after the underlying ledger changes.

Avoid Question Fatigue

Fifteen questions are a library, not one chat session. Start with data quality, choose the one decision currently at risk, inspect the evidence rows, and write one action or one reason to wait. Asking every question at once produces competing conclusions and no operating change.

Keep a decision log with the question, evidence set, answer date, limitations, selected action, and recheck window. Do not repeatedly ask until a preferred answer appears. A recheck should describe what narrowed, widened, or stayed unchanged under the same definitions.

A useful cadence has three layers. After the session, ask only whether the record is complete and whether a declared rule changed. At the weekly review, investigate one repeated divergence inside comparable setups. At the monthly review, test whether the intervention survived a fresh window and whether its apparent effect remains after costs, account state, regime, and strategy version are separated.

Keep rejected questions in the record too. “Why am I afraid?”, “What will work next month?”, and “Which setup guarantees profit?” exceed what fills and notes can prove. A refusal is not weakness when it prevents the model from inventing psychology, causality, or future performance. Rephrase the question around an observable sequence, or collect the missing evidence before acting.

Why TSB Coach Is Strong When the Evidence Is Bounded

Coach can move quickly across account, setup, risk, execution, cost, and journal context because the relevant records live in one evidence system. It can surface the first divergence, rank visible drags, compare matched cohorts, and turn a broad review into a focused next test. That is the product’s strength: disciplined interrogation of the trader’s own history at a scale that is painful to review manually.

The guardrails make the conclusion more useful, not weaker. Canonical values are rendered from the selected evidence set rather than recalculated in prose. Missing evidence produces qualification or refusal. Psychology, causality, market regime, strategy quality, and future performance are not inferred unless the canonical set supports the exact statement. The full Coach capability guide explains that evidence architecture.

TSB recognizes 332 exact import profiles and has normalized 600K+ imported trades. These figures describe import coverage and imported volume—not users, the sample behind an answer, proof of edge, or promised improvement.

TSB is our product. We disclose that ownership because this guide is specifically about using its AI Coach.

Methodology Note

• Product source: question scopes and conclusion boundaries were checked against `coach.php`, `AICoachEvidence`, `AICoachConclusionPolicy`, and `AICoachRouter` on September 10, 2026.
• Removed claims: invented monthly loss savings, universal “optimal” trade counts, automatic no-edge labels, guaranteed transformation, fixed adoption decay, and predictions were not retained.
• Evidence boundary: Coach explains and compares canonical evidence; it does not replace source reconciliation or manufacture missing intent.
• Decision boundary: an answer proposes a review or test, not personalized financial advice or permission to risk money.

For our evidence and correction process, see the editorial methodology.

Final Verdict: Ask for Evidence You Can Open

The best Coach answer points back to the exact records, definitions, limitations, and decision it supports. Scope one question, inspect the cited evidence, choose one reversible action, and recheck under frozen definitions.

That turns AI from a confident narrator into what TSB Coach is designed to be: a fast, rigorous second brain for the trader’s own evidence.