See what a DIBER experiment can tell you.
This demo starts with the results, not the JSON. Switch scenarios to explore KPIs, deltas, stability, transitions, logical relationships, methodological integrity, input exposure, and the model/prompt matrix.
Note: all values on this page are synthetic demonstration data. They are not real benchmarks for any model or provider.
—
—
—
Executive summary
The candidate’s key metrics and the delta versus the baseline.
Baseline vs Candidate
The comparison shows absolute values and deltas for every relevant metric.
Coverage × Quality
Each point represents a run/configuration.
Error profile
Not just how much performance is lost, but why.
Repeated-run stability
Mean and dispersion tell different stories. Here, DIBER makes the variability across repeated runs of the same configuration visible.
Point-by-point transitions
Which ground-truth points actually changed between the baseline and the candidate.
| Point | Importance | Baseline | Candidate | Transition |
|---|
Logical relations
Semantic dependencies reveal when an apparently correct conclusion rests on a problematic premise.
Methodology & input exposure
DIBER makes explicit what changed between configurations and which context was actually available to the model.
Model / Prompt Matrix
A compact view of the tested combinations, useful for separating the effect of the model from the effect of the prompt.
Behavioral profile
When a test includes behavioral signals, DIBER can aggregate them without introducing a second stochastic judge into the calculation.
Raw report
The demo is visual-first, while the complete deterministic payload remains fully inspectable.