Vibrr Research: Measuring AI Coding Platforms
Most writing about AI coding platforms repeats vendor documentation or reputation. Vibrr's research section is for what can be checked: what a source documents, what we have observed, and what a controlled, reproducible experiment has shown. As of the last update no experiment has been executed, so this section publishes methodology and infrastructure, not results. That distinction is the point.
The four evidence levels
| Level | Meaning | Example |
|---|---|---|
| Documented | A primary source says so | Cursor's documentation states rules do not impact Cursor Tab |
| Observed | Vibrr saw it happen, without a controlled experiment | None published yet |
| Validated | Reproduced across independent runs in a versioned experiment | None published yet |
| Hypothesis | Plausible but untested | Authorization enforced only in the interface, in prompt-first builders — listed on the Lovable guide as untested |
What has been measured so far
Nothing yet. The benchmark infrastructure exists — a compiler that produces identical base specifications for every platform, a run-record format that captures what is needed to reproduce a result, and a summary function that refuses to conclude anything from too few or non-comparable runs. No experiment has been executed against a real platform, so there are no results, no rankings and no conclusions. Any page that appears to state one is a mistake worth reporting.
See the benchmark index for the first designed experiment and its status, and the methodology for the rules every experiment must follow.
Why publish an empty results page
Because the alternative is worse. A benchmark page with plausible-looking numbers that nobody measured would be more convincing and entirely misleading. Publishing the design first lets readers judge the method before there is a result to be swayed by.