We are currently building something exciting and will be launching soon. Registration is temporarily closed. Launching soon — registration temporarily closed.

EVALUATION CHECKLIST

Evaluate Sentiqra in 30 minutes.

Nine metrics. Your own data. No sales call required. The same measurements our solutions engineers use during pilots.

Before replacing an existing analyzer - LLM-based, hosted moderation API, or in-house classifier - run both systems against the same labeled sample of your production content. Adopt this checklist, adapt it, or ignore it - but don't skip the parallel evaluation.

The nine metrics

  1. Top-1 hit rate
  2. Top-3 hit rate
  3. False-positive rate
  4. False-negative rate
  5. Per-language performance
  6. Threshold behavior
  7. Latency: p50, p95, p99
  8. Cost per decision
  9. Reproducibility check

The full checklist - how to build the labeled set, per-metric measurement detail, and interpretation guidance - is available to customers. Request access to get started.

Does preserving secondary signals improve your workflow?

Compare a highest-ranked-group-only policy with a policy that can use all returned qualifying groups. Use the same recorded response for both so you measure the policy difference, not a change in analysis. Keep child labels attached to their parent groups and evaluate against your existing process on the same labeled cases.

Include multi-concern messages, ordinary single-concern messages, and cases where nothing should qualify. Freeze the analyzer configuration, output profile, routing rules, and success criteria before evaluating.

  • Relevant secondary concerns surfaced and correctly routed.
  • Additional unnecessary alerts, duplicate tasks, and inappropriate escalations.
  • Reviewer workload and time to the correct owner.
  • Total operational cost, including review and integration work.

This is a customer-workflow evaluation, not the published Top-3 hit-rate metric. Finding an expected label among three results does not establish that every returned label is correct or that business outcomes improved. Reduced output profiles may omit signals; evaluate the profile you intend to deploy.

Keep your evaluation dataset in your own environment. Use the authenticated API for custom-text analysis under the published processing terms; do not send datasets through contact forms or email.

Evaluate on your own cases → Discuss your workflow →

Free tier: 5,000 runs per month

Enough to run this evaluation on a substantial labeled set. No card required. Configuration IDs and receipts are on every plan, including Free.

Request access Evaluation playbook