Survey

A building survey grades what is in front of it against what it is meant to be. Nobody condemns a house for having no kitchen when the walls are still going up. That is the whole idea here: detect the stage before you judge the code.

An audit that skips this produces the same report for a weekend prototype and a system serving real users, and the prototype's report is worthless. It is a wall of production-hardening findings - no rate limiting, no monitoring, no tests - every one of them technically true and none of them worth reading. The signal drowns. Worse, it teaches the reader that audits are noise.

So the order is: stage, then mechanics, then judgement, then vetting. Each phase earns the right to the next.

What this is not

Survey grades a codebase. It does not fix one, and it deliberately stops short of the specialists:

• A diff, PR or branch is /code-review. Survey judges the repo, not the change.
• God files and oversized modules get reported here as a signal, then routed to componentize to actually decompose.
• Fallbacks, swallowed errors and env defaults get reported as determinism smells, then routed to fail-fast.
• Duplicated UI, oversized files and package extraction all route to componentize.
• Mastra execution semantics are mastraudit's. When Mastra is detected, defer that lens to it rather than re-deriving it badly.
• Rendered behaviour in a browser is fieldtest. Survey reads code.
• Implementing the fixes is foreman, taking the clusters as briefs.