Analyze aggregator lender page flows

On a comparison page that aggregates several lenders (NerdWallet is one example, and listings might include SoFi or Upstart), you need to evaluate both user behavior and partner performance across two application paths: (A) the user finishes a prequalification or application module hosted on the site itself; (B) the user clicks away to a partner's own flow. Lay out a plan that covers:

1. End-to-end instrumentation of the page and of both flows — enumerate the events, identifiers, cross-domain tracking approach, and the method for linking sessions to partner outcomes when ITP and cookie loss interfere.
2. Success metrics plus guardrails, defined separately for the page and for each partner (CTR to partner, on-site completion rate, qualified rate, funded-loan rate, revenue per session, among others), each with exact numerator and denominator definitions.
3. Three insights that are not obvious, drawn from the visible differences in the partner information presented (APR ranges, fees, eligibility messaging, for instance), together with a hypothesis about how each might change the mix of users and eventual funding.
4. An experiment comparing on-site and click-out entry points for one partner, specifying the experimental unit, randomization, stratification, sample-size and power assumptions, a pre-registered decision rule, and the steps you would take to reduce selection bias and attribution mismatches.
5. An approach for measuring cannibalization between partners and mixed-traffic contamination across listings, accounting for returning users, tab hoarding, and last-click bias.

Give concrete event names, join keys, and the schema of a minimal metrics table you would maintain for weekly reviews.

Overview

This question assesses whether a candidate can handle analytics instrumentation, cross-domain attribution when identities and cookies are lost, experiment design, exact metric specification, and spotting partner cannibalization in a loan-comparison product.