Research Note
A research note is the raw lab book of serious design and AI teams. It lists the experiment, the exact variables, the outcome numbers, the surprises, and the concrete next step. Unlike polished reports it preserves the dead ends because dead ends teach faster than wins. The note exists so the next person does not repeat work that already failed in 2024.
It is not a blog post celebrating success. Celebration comes later. The note is written while the data is still fresh and the failure is still sharp. It is also not a bullet point list of things the team learned. Those lists lose all nuance and context the moment they leave the room where they were created.
Anthropic publishes research notes that read like decision logs at the model level. They document what prompt patterns worked, which token strategies collapsed at scale, and by exactly how many percentage points one approach beat another. Stripe keeps similar notes tied to design reviews so the same call does not get re-litigated every product cycle. The notes become the memory layer that survives when the original designer moves on.
Brian Lovin uses a lighter public version on his personal site. Each note is three or four sentences, dated, linked to the live product, and honest about what did not work. Jordan Singer does the same with product diaries attached to every small app he ships. Fifteen minutes per note compounds into years of visible senior thinking.
Write a research note when the experiment cost real time or real user traffic. Skip it for quick internal tests that will never be referenced again. The tradeoff is the small time investment versus the massive avoidance of duplicated effort and repeated mistakes. Teams without this habit reset to zero every time a key person leaves. Teams that keep the notes pull ahead because their mistakes become institutional capital.
The best research notes end with the sentence I was wrong about X and here is what we change in the next round. That single line separates professionals from performers.
A research note turns expensive failure into the cheapest education your team will ever buy.
Read the full guide
Related terms
Keep exploring
Decision Log
A decision log is a six-section case study format that records every major choice a designer made from problem to shipped outcome instead of presenting a gallery of final screens.
Design RFC
A Design RFC is a short dated document that locks in a design decision before work starts by listing the exact call, live alternatives, binding constraints, unhedged reasoning, owner, and later outcome.
Post-Hoc Rationalization
Post-hoc rationalization is writing decision logs after launch metrics land so every choice looks inevitable and every outcome looks like a win. It replaces real alternatives and live constraints with theater that senior reviewers spot in the first paragraph.