One developer. Accountable by name.

Raw responsiveness in day-to-day use: completion and generation latency, indexing time on a real mid-size repository, cold-start time, and how the tool behaves under a slow connection. On Verified reviews, speed is measured hands-on during real work sessions, and the run is published in the review's test record. On Documented reviews, the speed score is an editorial estimate from vendor documentation and public reports, and the review page says so.

What leaves your machine and under which terms: training on your code, prompt retention windows, telemetry defaults and opt-outs, self-hosting options, and whether a zero-retention or enterprise mode exists. Claims are checked against the vendor’s own privacy policy and documentation, not marketing pages.

How it feels to work with: setup friction, editor and CLI integration quality, configurability, documentation depth, failure modes, and how well it fits into an existing workflow rather than demanding a new one.

The editor’s holistic verdict on a 0–100 scale. It weighs the three dimensions above together with pricing fairness and project health (maintenance activity, license, longevity risk). It is a considered editorial judgment, not an automated average — two tools with identical sub-scores can earn different overall scores when their pricing or trajectory differs.