Log every setup you pass on, so your weekly review sees both halves

How it works

1. Add a skip sheet to your journal with fixed columns: date and time, setup tag, the grade you gave the setup in the moment (your own written grading scale), a reason code, and one free-text line. Reason codes come from your rules (outside my written criteria, risk limit reached, calendar rule) plus two that have no rule behind them: hesitated after a loss, and didn't trust it.
2. Write the row within a few minutes of passing, before you know how the setup played out. A row written later gets a late flag and is left out of the weekly comparison.
3. Keep the security out of the row the agent reads. The review is about your decision, not the name.
4. Each weekend, paste or export the week's skip rows together with your taken trades (setup tag, grade, rule check) and run the review prompt.
5. The agent counts skips by reason code and by grade, separates rule-based skips from unexplained ones, and lists every unexplained skip of a top-graded setup with its row number and your own note. It does not judge in hindsight whether a skip was right.
6. You decide what follows: a checklist line for the hour after a loss, a playbook note, or nothing. If you ever want outcomes for skipped setups, measure them later with your own tested script, never by asking the model to guess.

Copy-paste prompt

You review the setups I chose not to take. You do not give trading advice, you never say what I should have
taken or should take next, and you never estimate how a skipped setup would have turned out.

Inputs: {{skip_rows}} (date, time, setup_tag, my_grade, reason_code, note, late_flag) and {{taken_rows}}
(date, setup_tag, my_grade, rule_check). My reason codes and what they mean: {{reason_codes}}. Week: {{week}}.

1. Ignore rows with late_flag set, and say how many you ignored.
2. Count skips by reason code and by my grade. Split them into "rule-based" (a code tied to one of my written
   rules) and "no rule" (every other code).
3. List every "no rule" skip where my grade was one of {{top_grades}}, by row number and date, quoting my note.
4. Show taken versus skipped counts for each grade, so I can see whether I take my best-graded setups less often
   than my weaker ones.
5. If the week has fewer than {{min_rows}} skip rows, write "too few rows to read anything into" after step 2 and stop.
No praise, no reassurance, no price talk and no names of securities.

Guardrails

• The agent never estimates what a skipped setup would have made. It works from counts and your own notes only.
• Rows written after you have seen the outcome carry a late flag and are excluded, so hindsight can't rewrite the log.
• Security names stay out of the rows the agent reads; the review is about decisions.
• The agent returns a summary for you to paste into your journal. Nothing is written without your confirmation.
• The agent has no broker access and cannot place orders.

What can go wrong

The log only works if you write the row at the moment you pass. Filled in on Sunday from memory, it turns into a list of the setups that later worked, and the review then tells you to trust yourself more, which is usually the wrong lesson. The second trap is grade inflation: if every skipped setup becomes an A in hindsight, keep your grading scale written down and compare grade counts against your taken trades. And don't ask the model how a skip "would have done". It will give you a confident number with nothing behind it.

Source and references