Streaming Surface Audit

The streaming surface audit is a seven question checklist you run before any AI output interface ships. It catches surfaces that look functional but feel broken. The audit exists because most teams test for model accuracy and forget to test how the output actually feels to a human watching it arrive. Three failures or more and the surface ships broken regardless of model quality.

The questions are concrete. Does anything appear within two hundred milliseconds. Is the cadence smoothed or raw jitter. Does structure lock in without reflow. Is the stop button visible from token one. Can the user edit or reject mid stream. Does completion hand off to a real next step. Is the final output interactive or just a frozen transcript.

This audit is not a generic UX review. It is not optional polish. Teams treat it like an afterthought then wonder why users prefer competitors. The checklist forces the same rigor on output that most teams already apply to input.

Cursor would pass all seven. First feedback is instant. Cadence is smooth. Structure commits early. Stop button is obvious. Edits stay live. Completion leaves an owned diff. The artifact is fully interactive. Claude passes six easily with strong rhythm and artifact materialization. ChatGPT passes four at best. The gap is visible in daily use.

Run the audit on prototypes, betas, and production surfaces. Print it. Make it part of the handoff. The teams that treat streaming surfaces as full components run something like this before every release.

Use the streaming surface audit when your product ships any feature where users watch AI generate content longer than a few seconds. It is mandatory for AI native tools that want to feel premium instead of prototype.

Skip the formal audit for internal experiments or pure research interfaces where only accuracy matters. The checklist adds process that can slow velocity when the surface is disposable. The tradeoff matters. On customer facing products the audit prevents shipping broken feelings. On throwaway tools it becomes ceremony with no payoff.

Seven checks. First feedback, cadence, structural commitment, stop, edit, completion, post stream. Three failures and the surface ships broken no matter how good the model is.

Run the audit or ship a console log with a model attached and call it a product.