Find what comes before your rule breaks with a pre-session state log

How it works

1. Before each session, add four ratings from 1 to 5 to a sheet (sleep, stress, mood, focus), plus an optional note such as "after a big loss yesterday". Keep it under 20 seconds or you'll stop doing it.
2. Record rule breaks and checklist skips per session in your journal, for example from the pre-trade checklist and the exit-rule check.
3. Each week, run a local script that joins the two by date and computes the rule-break rate for each rating band, with the number of sessions next to every figure.
4. Paste only the script output into the agent with the prompt below. The agent explains only what the script shows. Below your minimum sample (for example 30 sessions per band) it says "not enough data yet" for that band and still reports any band that meets the minimum.
5. When a pattern holds, the agent states it plainly as an association. You may use the awareness to strengthen your pre-trade checklist on those days.
6. Paste the agent's note into your journal without changing the figures. Nothing is added to your journal until you review it and confirm.
7. Once a month, paste the latest script output and ask whether earlier patterns still hold or have faded.

Copy-paste prompt

You analyse my trading behaviour, not the market. You do not give trading advice and you never place orders.

Inputs: the output of {{analysis_script}}, built from {{state_sheet}} (date, sleep, stress, mood, focus, note) and {{journal_sheet}} (date, rule_breaks, checklist_skips). Each journal row is one session: a date with a rule_breaks count and a checklist_skips count, joined to the state log on date. I have already run the script and paste {{script_output}}, a table with columns rating, band (low 1-2 / middle 3 / high 4-5), rule_break_rate, checklist_skip_rate, n. rule_break_rate is the share of sessions with at least one rule break. Use only those figures and never recompute them. Do not use any other numbers.
1. Use only the script output for numbers, with n shown next to every figure. Do not recompute or invent figures.
2. For each rating, report the rule-break rate for low (1–2), middle (3) and high (4–5) days.
3. If a band has fewer than {{min_sessions}} sessions, write "not enough data yet" for that band and skip it. Still report every other band that meets the minimum.
4. Describe at most two patterns in plain words, and say clearly that they are associations, not causes.
5. Do not comment on profit or loss, any security or any market. Do not give medical or psychological advice.
6. If I say this is the monthly check, say whether each previously noted pattern still appears in the latest sample or has faded. If I have not given a prior pattern, skip this.
7. Present the summary as a draft for me to review. Do not write to my journal until I confirm.

Never suggest, imply, or give examples of any trading rule, position-size adjustment, checklist change, or action. Stick exclusively to neutral statistical reporting. Explicitly label every pattern as "This is a correlation only and may be entirely spurious."

Guardrails

• All numbers come from code, with the sample size shown. Below the minimum sample, nothing is concluded.
• Patterns are described as associations, not causes.
• The outcome measured is your own process (rule breaks, skipped checks), not daily profit and loss.
• Ratings stay in your own sheet and are only shared with the model you choose. This is not medical advice; if low ratings persist, talk to someone qualified.
• Output is forbidden from recommending or exemplifying any change to position size, trading schedule, or rules; you must supply any such interpretation yourself.

What can go wrong

Comparing ratings with daily profit and loss mostly finds noise, because one big day dominates. Comparing them with rule breaks and skipped checks, which you control, gives steadier results. The other failure is rating after the session, when the outcome colours the answers. It only works if you rate before you start.

Source and references