Design Review Doc

A design review doc is a short dated artifact that pins down one design decision before any pixels ship. Six fields make it work. Decision states the call in one concrete present-tense sentence. Alternatives lists two to four options that were actually on the table. Constraints names the real limits sourced from engineering roadmaps, support data, legal, brand, or user metrics. Reasoning explains the tradeoffs with because statements and direct links to tests or research. Owner names the designer plus the reviewer who signed off. Outcome gets written two weeks to two months later with numbers, surprises, and what the team would change next time. Stripe has run these for years so institutional memory survives turnover. Linear calls them design RFCs and treats them as first-class as code. The doc shifts every critique from taste arguments about button radius to sharp debate about which constraint actually ruled. In the AI era it survives because models can generate polish in minutes but cannot defend the eighteen paths not taken.

It is not a case study. Case studies rewrite history so every choice looks brilliant and inevitable from the start. The review doc keeps the competing alternatives alive so a reviewer sees exactly what got killed and why. It is not meeting notes. Meeting notes capture who talked longest and which emoji got used in Slack. The review doc records the decision even if it was made in a thirty-second voice note. It is not a Figma file full of comments and redlines. Those invite endless bikeshedding on shadow softness. The review doc demands evidence and forces the conversation onto judgment that compounds across projects.

Concrete example from a 2025 checkout redesign at a fintech team that copied Stripe's format. Date: 2025-02-10. Project: Checkout v3. Decision: Stack email password and magic link options vertically with no social login row. Alternatives: 1. Inline social login buttons above the fields. 2. Two-screen flow that collects email first then password. 3. Magic link only with no password field at all. Constraints: Mobile-first at 360px minimum width, engineering cannot ship OAuth this quarter, support data shows 14 percent of tickets stem from password reset confusion, brand guidelines reject generic Continue with patterns. Reasoning: Vertical stack ships fastest and matches the direct brand voice. The two-screen version tested cleaner in 2024 yet added a full screen that produced a 38 percent drop-off on screen two. Magic link alone blocked password managers that 22 percent of paid users rely on. The binding constraint was engineering capacity plus the support ticket data. Owner: Maya Chen design lead. Reviewer: Jordan Park engineering lead. Outcome updated 2025-03-01: Conversion rose 6.1 percent versus v2. Password reset tickets dropped 9 percent. Magic link remains viable for v3.1 once OAuth ships in Q3 but we will add a hybrid option sooner after seeing real user data. That one page replaced a 45-minute critique and prevented the same sidebar conversation from resurfacing three months later.

Use a design review doc when the decision will be questioned six months or three years later when new team members join or when a hiring manager wants to see how you think. Drop the doc in the channel 24 hours before critique so reviewers arrive armed with specific challenges instead of vague opinions. Linear bakes them into every major surface launch. Vercel keeps them in the same repo as the engineering spec so the why and the what live together. Anthropic publishes research notes that function as decision logs at model scale. Brian Lovin runs his publicly on a minimal personal site with three-sentence entries that double as portfolio and proof of taste. Jordan Singer attaches similar diaries to every small product he ships. The pattern compounds. Ten honest entries teach you more about your own constraints than any retrospective meeting.

Skip the design review doc for trivial changes like updating a label or tweaking icon weight. Do not use it when leadership treats writing as busywork or when the culture rewards shipped pixels over defended decisions. Never write the entry after launch if you plan to invent strong alternatives that never existed or round up every metric. Senior reviewers detect that theater in under two minutes and the entire artifact loses credibility. Jordan Singer skips logging micro calls on his solo projects because the overhead would kill velocity but still documents the three calls per project that shaped final direction.

Senior work without a public or private trail of reviewed decisions reads as expensive junior work.