What we allow, what we don't, and why.

Anyone can create a profile for any working professional or company. A profile can be created without the subject's consent, the same way a Glassdoor review, a media profile, or a public bio can. What cannot happen without consent:

• Rating anyone who is not a working professional — students, minors, private individuals who do not interact professionally.
• Rating non-professional behavior — romantic relationships, personal medical information, family life, protected characteristics.
• Posting content that identifies anyone other than the profile subject, including co-workers who were not intended to be rated.
• Posting private messages, internal documents, or information obtained under confidentiality.

Subjects can always claim their own profile by signing in at /me. Claiming a profile unlocks notifications, lets the subject add context, and surfaces the profile in the subject's network. Claiming does not give the subject control over ratings or comments, because it can't — those are anonymous by design.

Ratings do not become public signal on their own. They have to pass thresholds designed to prevent any single person — or small group — from defining someone's reputation.

• Per-type rater thresholds. People and professors unlock at twenty distinct anonymous raters; companies and universities at three. Below the threshold, aggregate scores are never shown — only the comment thread is visible. The per-type difference reflects that companies and universities already have institutional disclosure; people and professors do not and need more rater diversity before being summarized.
• Five people for a pattern alert. Cohort-level pattern detection requires at least five ex-colleagues from the same company who overlapped with the subject.
• 65% above the midpoint. A pattern only surfaces when at least 65% of that cohort rated the subject above fifty on the dimension in question. Below that it's noise.
• One vote per person per dimension. Ratings are not re-weighted, decayed, or boosted for credibility. The only weighting is the thresholds above.

We remove profiles, comments, and ratings when they violate the rules below. We do not remove them for being unpopular, harsh, or unflattering — disagreement is the product.

Profiles will be removed for:

• Fabricated identity — the profile is of a person or company that does not exist, or is an impersonation.
• Subjects outside the consent policy — minors, private individuals, or anyone not acting in a professional capacity.
• Doxxing — the profile contains non-public contact information, home address, family details, or similar.
• Subject requests that meet the consent policy above.

Comments will be removed for:

• Hate speech, harassment, or credible threats of violence.
• Sexually explicit content.
• Specific, falsifiable claims that are demonstrably false and amount to personal attack.
• Off-topic spam, advertising, or disclosure of private info.

Ratings themselves are never removed. They are numeric and anonymous — there is nothing to moderate. The aggregate is the truth. Anti-manipulation work (below) is a separate path.

If a profile of yours, a comment you wrote, or a decision about a profile you care about seems wrong, you can appeal. Every appeal is reviewed against the policies above by the moderation team — not by the person who made the original call.

Open an appeal by emailing safety@personagraf.com with the profile URL or comment permalink, what you're appealing, and any context you want us to consider. We respond within seven days. If an appeal is rejected, we explain which policy the decision was based on.

During an appeal, flagged comments remain visible. Removed profiles remain removed until the appeal resolves — we would rather be wrong on the side of being offline for a week than of leaving clear abuse up pending review.

Coordinated manipulation — brigading, dogpiling, a group agreeing offline to tank someone's score — is the most common attempt to weaponize a reputation platform. Our response is layered and, by design, does not require us to know who anyone is.

• Rate limits. A single anonymous identity cannot submit ratings for more than a small number of profiles in a short window. This removes the most obvious automation path.
• Burst detection. We monitor rating velocity per profile. When a profile receives a statistically unusual burst from new identities within a narrow time window, the burst enters review. During review, affected ratings still count in the stored totals but may be excluded from public aggregates until cleared.
• Cluster analysis. We look at structural signals — shared IP ranges, device fingerprints at the edge, account-age skew — to identify coordinated clusters. Clusters confirmed as coordinated attacks have their ratings rolled back. We never reveal the rating values of any specific identity, including as part of this process.
• Transparency. When we take action on a profile for manipulation reasons, the profile shows a small notice: "Some ratings on this profile are under review." We do not disclose which ones. The notice is removed when review closes.
• What we will not do. We will not reveal the identity or rating values of any individual rater to anyone, including the subject of a profile, law enforcement without a lawful order, or our own staff outside the strict bounds of this review process. The anonymity contract is structural, not a policy line.

For safety concerns, appeals, or pattern reports: safety@personagraf.com. Product questions and bug reports go to hi@personagraf.com. We aim to respond within seven days for safety, and within a reasonable window for everything else.