Data collection principles
- Only publicly reported media coverage, verified NGO materials, and voluntary user signals are collected.
- Non-public information, internal materials, or data obtained through unlawful means (hacking, eavesdropping) is not used.
- Source data is preserved immutably and is not modified after the fact.
Classification and scoring
- Scores are computed per algorithm version, with change history kept permanently at /methodology/changelog.
- AI auto-classifications may be wrong; corrections are accepted via the correction channel.
- Risk Watch displays only levels 0 and 1 in public surfaces; level 2 and above are not shown.
Correction request rights
- Celebrities and their legal representatives may request data corrections or takedown.
- Media outlets may request the suppression of their own coverage or dispute the credibility evaluation of their outlet.
- General users may report factual errors, misclassifications, or inappropriate cheer messages.
Processing time (SLA)
| priority | Completion SLA | Examples |
|---|---|---|
| urgent | 24 hours | Takedown request from celebrity legal representative |
| high | 48 hours | Media representative request; press-noted issue |
| normal | 72 hours | Factual error or misclassification request |
| low | 7 days | Label or romanization corrections |
Rejection reasons
- When the facts match a verified source (FACTUAL_VERIFIED_NO_CHANGE).
- When the dispute concerns the algorithm itself (ALGORITHM_NOT_DISPUTABLE — change proposals go to /methodology/feedback).
- Third-party disputes (THIRD_PARTY_DISPUTE).
- Pending legal proceedings (LEGAL_PROCEEDINGS_PENDING).
All eight standard rejection codes are listed on /correction-policy.
Fairness of decisions
- Every correction decision is recorded in the operations log.
- Quarterly random-sample audits are conducted jointly by external counsel and the operations team.
- The company does not provide preferential or discriminatory treatment to any celebrity, outlet, or B2B customer.