dotplot-analyze-product

Analyze this product

The statistics come from the dotplot MCP server, never from you. Your job is finding the data, choosing well, and explaining what it means.

1. Find the event data

Call analyze with no arguments first if you don't already know where the data is — it returns the procedure. In short:

• Is DOTPLOT_DB_URL set? Is there a connection string in .env? Is a Postgres/Supabase MCP already connected?
• Read the schema. You're looking for tables recording things users did — orders, sessions, posts, messages, subscriptions. An events table is nice but most early products don't have one, and that is fine.
• Turn those tables into events with load_from_db, one SELECT per action joined by UNION ALL:

SELECT user_id, created_at::date AS date, 'purchase' AS event FROM orders
UNION ALL
SELECT user_id, added_at::date, 'add_to_wishlist' FROM wishlist_items

If there's no database and nothing is tracked, say so plainly and switch to /dotplot-add-tracking. Don't invent a report.

2. Choose the value event yourself when you can

analyze picks the most-repeated non-vanity action, and says why. But you have read the codebase, so you know things the code cannot: that purchase is value and view_item is browsing, even though they look identical in the numbers. When you know better, pass value_event explicitly.

Check the result's value_event.why and others_available. If the choice looks wrong, call again — it's cheap.

3. Report it like a person

Open the report, then say the headline out loud. Lead with what to do, not with the numbers.

Three things to get right:

• The aha moment is a correlation. Say so. "Users who did X became regulars" can mean X causes retention, or that engaged users do everything. The honest recommendation is to put it in onboarding and A/B test it — not to declare it.
• Small numbers deserve hedging. The report drops weeks with under five users, but a rate over 8 users is still thin. Mention the denominator when it's small.
• Match their language. Pass lang="ko"/"ja"/"en"; for anything else, translate get_report_strings and use generate_report with lang="custom".