There is no profile-free code quality

Performance ↔ Readability. optimised code stops reading cleanly - unrolling, caching tricks, and branch-avoidance all trade clarity for cycles.

Performance ↔ Memory. caching buys speed with memory; streaming buys memory with speed. You pick one direction per hot path.

Robustness ↔ Quick-iteration economy. defence is ceremony: validation, failure handling, and guards are exactly what a one-off script strips to stay fast to write.

Security ↔ API ergonomics. fail-closed friction fights ergonomic defaults - the safest interface is rarely the smoothest one.

Auditability ↔ Performance / Maintainability. audit trails add cost and coupling: every traced decision is extra work on the path and extra structure to maintain.

Concurrency ↔ Readability / Economy. correct concurrent code is verbose - guards, bounds, and ownership discipline don't compress.

Resource cost ↔ Observability. telemetry costs compute: every log line and metric is capability bought with efficiency.

Strict precedence, most honest signal first:

1. Your ranked priorities (the 1st-to-5th picker on the dashboard) - the most precise signal there is: you say exactly which dimensions matter, in order, and the whole report reads against that.
2. Your declared profile - you name one of the ten; the report holds the code to that bar.
3. The auto-read - a confidence-gated heuristic over the measured shape. It only names a profile when the code is differentiallystrong on that profile’s priorities, it reports its own confidence (weak / moderate / strong), and when the dimensions read as balanced it abstains and says so rather than force a label. The auto-read describes what the code is; only you can say what it is for.