Engagement integrity: anti-self-inflation and verified-stats reporting
Stops project owners from faking customer engagement to make their workspace look more active than it is. Important because Hollahoop will surface public marketing stats ("shipped 23 user requests in 90 days", embedded badges, public roadmap vote counts) - if owners can fabricate engagement, those signals become dishonest and burn the brand.
Detection signals layered, all logged to audit_events with fraud_score and reason: (1) team-vote separation - owners/admins/members of an org cannot have their votes counted toward public engagement totals on their own projects; their votes still register but render as "team vote" with a small badge. (2) device-fingerprint and IP clustering - many distinct accounts voting on the same posts from the same IP/device fingerprint within a short window get a soft flag and stop counting toward public stats until reviewed. (3) behavioural anomalies - sequential account creation timestamps, identical email patterns ([email protected] [email protected] [email protected]), votes within seconds of signup, fresh accounts that only ever vote on one project. (4) email-domain reputation checks (cross-references real-user verification card).
Public stats compute on verified engagement only by default; admin dashboard shows a delta between raw counts and verified counts so owners see "you have 47 votes, 32 verified, 15 flagged" with explainability per flag. Public widgets (shipped-requests badge, public roadmap counts) display verified-only numbers and link to a public methodology page so the integrity is provable.
Per-org tools: review queue for flagged engagement, manual override per item (admin can mark a flagged signal as legitimate, with audit log), and a quarterly integrity report. Hollahoop-staff-only override for cross-project egregious cases. Gated behind feature flag "integrity.verified_stats" so the rules can be tightened gradually without breaking honest workspaces. Out of scope: ML-based fraud models (rule-based v1 only), monetary penalties or auto-suspensions (always human-reviewed first).
Comments (0)
No comments yet.
Sign in to leave a comment.