Summarise weekly journal stats by setup, with a minimum sample before any conclusion

How it works

1. Open the journal sheet from your post-trade review, with computed columns win_rate, avg_win_R, avg_loss_R, expectancy_R and n already in place.
2. Copy the grouped stats (or export that range) and paste them into the agent with {{min_sample}}.
3. The agent drafts a short summary: best and worst setup by expectancy among groups that meet the minimum, always showing n.
4. Groups with n below the minimum are listed as "not enough data", with no ranking or conclusion.
5. You read the draft. The agent does not write anywhere; when you are happy with it, you paste it into your weekly-review tab yourself.

Copy-paste prompt

Summarise these pre-computed journal stats (pasted below) from {{journal_sheet_name}}.
0. If any individual trade rows or raw data appear in the paste (instead of only the pre-aggregated table with exactly the five computed columns), refuse to summarize and reply only "Please provide the pre-computed grouped stats table only."
1. Use the computed columns only (win_rate, avg_win_R, avg_loss_R, expectancy_R, n). Do not recalculate.
2. Report by setup_tag, regime and weekday. Always show n.
3. Any group with n < {{min_sample}} is "not enough data": no ranking, no conclusion.
4. Name my best and worst setup by expectancy among groups that meet the sample size, and suggest one question to look into (not a rule change).

Guardrails

• Minimum sample size before conclusions.
• Always show the sample size next to each statistic.
• Questions, not rule changes; rule changes go through the weekly review.
• The model never calculates win rate, average R, or expectancy; those come only from the sheet. If the pasted table is missing n or expectancy_R, stop.
• If any individual trade rows or raw data appear in the paste (instead of only the pre-aggregated table with exactly the five computed columns), refuse to summarize and reply only "Please provide the pre-computed grouped stats table only."
• The agent only drafts the summary. It does not write to Google Sheets and it never edits journal rows. After you accept the draft, you paste it into the weekly-review tab yourself.

What can go wrong

Ten trades make any setup look brilliant or terrible. Showing n next to every number, and hard-blocking conclusions below a threshold, stops overreaction. Two further risks: when data is pasted in, the model may ignore the "use computed columns only" instruction and try to recompute statistics, so keep the calculations in the sheet and tell it not to recalculate. And if a spreadsheet formula is wrong, the agent will present a confidently worded but wrong summary, so check your formulas before trusting the output.

Source and references