Trust the working, not the branding.

Three separate layers

Editorial rating, audience reaction and documented evidence never collapse into one mysterious score. Leaving a group can become a sourced event. It does not automatically make somebody less trustworthy.

Perception curves

A perception curve is published only where contemporary audience reaction has been researched separately, event by event. Today that is two profiles. Every other researched creator shows their dated events with no score at all. Where a curve exists it begins at 50 on the date the channel was created. That is a common comparison baseline, not an estimate of instant trust. Green segments rise and red segments fall, and the absolute height stays comparable across creators, so a recovery from 18 to 25 is still below average. The line holds its previous level between outcomes. On a researched date it begins a reaction slope lasting no more than 45 days, then holds the settled level. Same-date clusters divide that window in sequence, so backlash and a response can show as a fall followed by a partial recovery. Nothing slopes backwards through the quiet years before an event. Documented events without a researched audience outcome appear as hollow circles on the curve at the score that existed on their date. They do not move the line. A pale confidence ribbon surrounds each estimate, tightening as the reaction evidence strengthens. None of this is polling.

Until 28 July 2026 the other profiles carried a curve generated from the wording of each event. It could not tell subject matter from conduct, so a creator was penalised for documenting a scam, a death or an assault exactly as if they had committed one. Those scores were withdrawn rather than corrected in place, because a wrong number about a real person should come down first and be rebuilt afterwards.

The paused public-participation model

If participation opens after the beta, a visitor will be able to propose an event and check somebody else’s. Every proposal will answer two separate questions, with both counts visible. None of these controls is active today.

Did it happen? The current design requires at least 8 confirmations for a sourced event and 15 for an unsourced one. Raw numbers are not enough: agreement must also clear a statistical floor, so a narrow split never passes however many people arrive. These thresholds have not been tested with public traffic.

Was it good or bad? The proposed direction uses positive, negative and neutral votes. Neutral counts, so a shrug can hold the line still. Movement starts slowly and no single event can move a creator more than 14 points. This formula remains inactive during the beta.

Where editors would fit. The design caps an editor at halving or adding half to an event’s effect, with a standing adjustment of at most eight points. An editor cannot set a score or reverse the public’s direction. Rejections for fabricated, defamatory or duplicate material require a written audit reason.

Why it is paused. Open voting can be organised against. A coordinated group can push an event through, and a devoted fanbase can lift someone they like. Fingerprint limits raise the cost but do not solve coordinated campaigns. Public participation stays off until velocity and network-level abuse checks are ready.

Public trust history

This estimates how trusted a creator appeared in dated public discussion. It does not decide whether they are honest. Silence creates a gap, not an automatic recovery, and events only define windows for measuring contemporary reaction.

Evidence charts

The current category charts measure source coverage, independent sourcing, documented events, public corrections and editorial confidence. That record stays separate from public trust. The gap between reputation and documentation is often the useful bit.

AI research desk

Automated systems collect permitted public material, match it to creators, remove duplicates, cluster related events and prepare short drafts for an editor. They do not write public reviews, assign an editorial rating or decide whether an allegation is true.

• Routine channel metadata can update automatically.
• High-impact allegations, legal claims and sharp score movements enter human review.
• Summaries retain source links, dates, opposing accounts and creator responses.

Publishing thresholds

Editorial ratings require a published reason. Public-trust periods have separate source, sample and confidence thresholds. Audience ratings remain blank throughout the read-only beta. If reviews open later, a creator will need at least 25 approved ratings before an audience number appears.

Corrections and appeals

Creators and viewers can send responses and sourced corrections to corrections@creatorsonrecord.com. Privacy requests go to privacy@creatorsonrecord.com. Material methodology, policy and evidence changes are recorded in the public change log.