Quality SLA
Compiles first try, or your credit's back
If your AI-generated Power BI visual fails to build, or scores below 14/18 on our 6-axis quality grader, we refund the credit automatically — within 5 minutes, no ticket required.
Live · last 30 days
Collecting baseline
We turned on the Tier-2 grader at the end of April 2026. Our rolling 30-day rate publishes here once we have ≥50 graded generations in the window. Until then, the auto-refund SLA is still active — every low-graded visual is refunded within 5 minutes. Currently 2 graded generations and counting.
Most AI tools claim “works most of the time.” We measure it, publish the number, and refund you when we miss. The grader runs on every successful build, scoring the visual against your original prompt + the spec the AI inferred, across six independent axes (0–3 each, summed 0–18). The cutoff for a refund is 14— below that, at least two of the six dimensions are materially wrong and the visual probably isn't usable as-is.
The 6-axis rubric
Visual type match
Does the rendered chart match the requested chart family? A request for a sankey shouldn't return a stacked bar.
Field roles
Are the dataRoles in capabilities.json named and typed sensibly for the prompt? Measure vs Grouping picked correctly.
Encoding
Axes, scales, and mark choices appropriate for the data shape. Categorical on the right axis, no log scales on counts, etc.
Labels
Axis labels, value labels, and titles present where they aid comprehension. Not over-labelled, not bare.
Legend
Legend present iff there are multiple series, and legible. Single-series visuals don't need a legend.
Color logic
Categorical vs sequential palette chosen appropriately. No clashing combinations. Brand colours respected when specified.
How the refund works
1. Build fails. Validator catches a TypeScript error, capabilities mismatch, or pbiviz packaging error. Credit refunded immediately, before the user is even told the build failed (you see the failure in your history with credit refunded next to it).
2. Build succeeds, grade < 14. The judge scores within ~90 seconds of build success. If the total is below 14, an auto-refund posts to your credit balance. No action needed from you.
3. Build succeeds, grade ≥ 14.Credit consumed normally. If the visual still isn't quite right, click Refine in the workspace and iterate — refinements are a new charge, but the original credit stays consumed.
FAQ
What counts as a 'failed' visual?
Two things: (1) the .pbiviz fails to build (TypeScript error, broken capabilities.json, etc.), and (2) the visual builds but scores below 14/18 on our 6-axis grader. Either path triggers an automatic credit refund within 5 minutes — no support ticket required.
Why 14/18 specifically?
14 is the threshold below which the visual probably needs a refinement to be useful. Above 14, the visual matches the prompt across most axes and only minor polish is missing. Below 14, at least two axes are materially wrong — usually wrong chart family, broken field bindings, or illegible labels. Our internal data shows 14 is the inflection point where users click Refine within the first 24 hours.
Who decides the score?
An LLM judge (Anthropic Claude Sonnet 4.5) scores every successful build against the original prompt + spec across the 6 axes, on a 0-3 scale per axis. The judge sees the prompt, the classified intent, the generated capabilities.json, and excerpts of visual.ts + settings.ts. The model is locked to one snapshot per quarter so scores are comparable across time.
What happens if the judge is wrong?
If you think a refunded visual was actually fine — or a non-refunded one wasn't — email support@vizforge.ai with the VF-XXXX-XXXX reference ID. We override individual scores by hand. The judge model is also retrained quarterly against human-labelled disagreements, so systematic biases narrow over time.
Does a refund mean I lose the visual?
No. The .pbiviz is yours to download even if the credit was refunded — refunds are a credit-balance event, not a delete-the-output event. If a low-graded visual is actually useful for your case, you keep it; we just don't charge you for it.
Can I see my visuals' grades?
Yes — every completed generation in /workspace shows its Tier-2 grade in the iteration row when graded. Grades populate within ~90 seconds of build success.
Try VizForge — risk-free
5 free credits on signup. Auto-refund covers paid generations.
Generate your first visual →