# Calibrated fresh-chat comparison

Both products have identical descriptions for signup_started (form entry), sign_up (submission), and signup_completed (explicit completion), grounded in the generator's executable state transitions. No expected counts, user-level diagnosis or answer key was added. Cartos's company-specific browser fixture has the same definitions and retains all simulation limitations.

Repeat the original four questions verbatim, with the one-day window specified in question 1. Cartos receives only the source prefix Use Mixpanel for this. If a product widens the onboarding window, request the same one-day window and preserve the original answer. Saved prior context remains available; this is not blinded. Preserve the pre-correction run separately.

After observing native Mixpanel's onboarding response, the clarification was expanded to: “Keep the original signup-start cohort and the same one-day window.” Its original check began at signup_completed over widened dates and returned eight entrants. The wording change restores the requested cohort as well as the window, adds no event-name or count hint, and applies to Cartos if it changes scope too. This amendment was made after observation and is logged; it was not preregistered.