Design a Food Review System That Pays Rewards Based on Review Quality

Quick Overview

A system design question about letting food delivery customers review the items they ordered and earn monetary rewards based on review quality. It tests eligibility checks, asynchronous quality scoring, an exactly-once reward ledger with limits and reversals, and defenses against reward farming.

Design a Food Review System That Pays Rewards Based on Review Quality

Company: DoorDash

Role: Software Engineer

Category: System Design

Difficulty: medium

Interview Round: Onsite

Design a system that lets customers of a food delivery platform write reviews of the food items in orders they received, and that pays each reviewer a monetary reward whose size depends on the quality of the review. The reported prompt is only this statement, so how quality is judged, how large rewards are, and the scale and payout rules are open: settle them with the interviewer before designing. ```hint Decouple scoring from posting Think about which checks must finish before a review is accepted and which can run afterwards, and what the reviewer sees in between. ``` ```hint Money changes the threat model Ask how someone would try to collect rewards without writing genuine reviews, and at which step of the flow each attempt can be caught. ``` ### Clarifying Questions - Who may review: only customers whose order was delivered, once per ordered item? Can a review be edited after its reward is paid? - What makes a review high quality: specific detail about the dish, photos, a rating consistent with the text, later helpfulness votes from other customers, a moderator's judgment? - How are rewards sized: fixed tiers or a formula, with per-customer limits and an overall budget? - Must a reward be paid immediately, or can it wait for a review period, and can it be taken back if the review later turns out to be fraudulent? - How many orders and reviews arrive per day, and how soon must a new review appear on the item's page? ### What a Strong Answer Covers - Requirements and scale agreed first, including eligibility, quality criteria and reward rules - A review submission path that verifies the order and item, is idempotent, and does not wait for quality scoring - An asynchronous quality-scoring pipeline that combines rules, models and later signals, with human review for borderline or high-value cases - A reward ledger that pays each review exactly once, enforces limits and budget, and supports reversals - Defenses against reward farming: duplicate or generated text, reused photos, linked accounts, edits after payout - A data model, a read path for item pages, scaling, failure handling and monitoring ### Follow-up Questions - A reviewer replaces a well-rewarded review with low-quality text after the payout. How does the system respond? - The quality model is retrained and its scores shift. How do you roll it out without paying very different rewards for similar reviews? - The reward budget runs out partway through the period. What happens to reviews submitted afterwards? - How would you detect many accounts posting near-identical or machine-generated reviews to farm rewards?

Overview: A system design question about letting food delivery customers review the items they ordered and earn monetary rewards based on review quality. It tests eligibility checks, asynchronous quality scoring, an exactly-once reward ledger with limits and reversals, and defenses against reward farming.

Read the full DoorDash Software Engineer interview experience this question came from

DoorDash logo
DoorDash
Nov 16, 2026
mediumSoftware EngineerOnsiteSystem Design
0
0

Design a system that lets customers of a food delivery platform write reviews of the food items in orders they received, and that pays each reviewer a monetary reward whose size depends on the quality of the review. The reported prompt is only this statement, so how quality is judged, how large rewards are, and the scale and payout rules are open: settle them with the interviewer before designing.

Clarifying Questions Guidance

  • Who may review: only customers whose order was delivered, once per ordered item? Can a review be edited after its reward is paid?
  • What makes a review high quality: specific detail about the dish, photos, a rating consistent with the text, later helpfulness votes from other customers, a moderator's judgment?
  • How are rewards sized: fixed tiers or a formula, with per-customer limits and an overall budget?
  • Must a reward be paid immediately, or can it wait for a review period, and can it be taken back if the review later turns out to be fraudulent?
  • How many orders and reviews arrive per day, and how soon must a new review appear on the item's page?

What a Strong Answer Covers Guidance

  • Requirements and scale agreed first, including eligibility, quality criteria and reward rules
  • A review submission path that verifies the order and item, is idempotent, and does not wait for quality scoring
  • An asynchronous quality-scoring pipeline that combines rules, models and later signals, with human review for borderline or high-value cases
  • A reward ledger that pays each review exactly once, enforces limits and budget, and supports reversals
  • Defenses against reward farming: duplicate or generated text, reused photos, linked accounts, edits after payout
  • A data model, a read path for item pages, scaling, failure handling and monitoring

Follow-up Questions Guidance

  • A reviewer replaces a well-rewarded review with low-quality text after the payout. How does the system respond?
  • The quality model is retrained and its scores shift. How do you roll it out without paying very different rewards for similar reviews?
  • The reward budget runs out partway through the period. What happens to reviews submitted afterwards?
  • How would you detect many accounts posting near-identical or machine-generated reviews to farm rewards?