This is the open part of our methodology. We don't claim academic rigor — the goal is to give makers a quick, honest signal, not a thesis.
AI report grading
Every report is graded on four axes; the final score is a weighted sum from 0 to 100:
- Specificity (35%) — are there references to specific elements, steps, screenshots.
- Usefulness (30%) — can the maker actually fix something based on the report.
- Completeness (20%) — are all questions from the brief covered.
- Tone (15%) — respectful, no flame or personal attacks.
Reports below 40 go to arbiter review automatically. Reports above 70 boost the tester's rating.
Tester rating
Rating is the average AI score over the last 20 accepted reports times a discipline factor (0.7–1.0): missed deadlines and escalations reduce the factor.
Insight grouping
AI clusters similar issues by semantic similarity and frequency. The risk map shows the top 5 clusters with weight, quotes and screenshot examples.
Anti-fraud
- A report can't be submitted earlier than the minimum scenario time.
- AI detects boilerplate answers and AI-generated text.
- Multiple rejections in a row — temporary slot freeze.
- Duplicate reports from the same tester are rejected automatically.
What we don't do
We don't validate hypotheses for you, don't measure NPS and don't claim to replace moderated user tests. We're about speed and volume, not lab-grade purity.