What's the difference between deployment patterns and pricing tiers?
Deployment patterns describe how you integrate Looper (inline gating, scheduled monitoring, or continuous sampling).
Pricing tiers are based on volume and features. You can use any deployment pattern with any pricing tier.
Why is multi-variant analysis recommended over single-variant?
Our experiments proved that single-variant baselines fail to detect drift—they showed "negative drift" even as models
degraded from 80% to 50% correctness. Our proprietary multi-variant configuration is 113% more
sensitive and provides clean, monotonic degradation signals.
What are the comprehensive deep-dives in Business tier?
Weekly comprehensive analysis using our expanded proprietary variant set. This provides maximum failure mode coverage
and helps identify specific reasoning weaknesses. Best for research, pre-production testing, or regulated industries.
How does drift detection work?
Looper measures reasoning stability over time. When models degrade (due to updates, data drift, or context shifts),
their reasoning becomes less consistent before accuracy drops. By tracking stability scores, we detect degradation
5-10 days earlier than traditional accuracy-based monitoring.
Can I sample requests to reduce costs?
Absolutely! Pattern 3 (Continuous Sampling) is designed for this. Sample 5-20% of production traffic to reduce costs
by 80-95% while still catching drift within 1-2 days. You can also use adaptive sampling—start at 10%, increase to 100%
when instability is detected.
What's the ROI of drift detection?
Our cost-benefit analysis shows typical ROI of 100-1000× on the first prevented incident. For example:
Professional tier costs $199/month but prevents incidents that typically cost $50K-$500K in lost revenue,
support costs, and brand damage. See our detailed case studies in the documentation.
Do you store our prompts and data?
Privacy-first by default. We store stability scores, risk bands, and metadata—but you control whether raw prompts are stored
via the STORE_PROMPTS flag. For production, set it to false to avoid storing any customer data.
The LRI (Looper Reliability Index) tracks model performance using only hashed prompts and aggregated metrics.
What happens if I exceed my plan's request limit?
We'll notify you at 80% and 100% of your limit. After that, you can upgrade your plan or pay overage at
$0.002 per additional request (Business tier) or contact sales for Enterprise pricing.