EIP-6: Penalty for Positive Reviews Downvoted by Recipient

Summary

When the recipient of a positive review downvotes that review, this strongly indicates it is likely spam, slop, or inaccurate. In this case we will penalize the author's Ethos score. Penalties escalate with repeat offenses.

Motivation

As the author of a review, there is little risk in being wrong, writing slop, or incorrect information about the target. Ethos is currently facing issues with positive reviews written by Grok that are inaccurate (e.g., ZacXBT getting reviews as if he were ZachXBT), sloppy (generic reviews without substance), and inauthentic (written without actual knowledge about the person).

Recipients of these reviews are often frustrated, as this sentiment is now reflected on their profile regardless of its accuracy or likelihood of being spam.

This EIP adds risk to the equation where writing reviews that are inauthentic, even when they positively impact the target, carries credibility score downside. We believe this will help neutralize excessive AI usage in writing reviews that are inauthentic.

• Combat spam reviews that clutter profiles with irrelevant or promotional content.
• Preserve user agency over their profile presentation without enabling suppression of legitimate criticism.
• Create meaningful deterrent for serial spam reviewers through escalating penalties.

Specification

For the sake of this specification, we’ll refer to the act of the recipient downvoting a positive review for themselves as “marking it as spam,” although in the actual mechanism design it is simply downvoting the review.

1. Eligibility

• Only positive reviews (sentiment = "positive") qualify.
• Negative and neutral reviews cannot be penalized under this mechanism.
• Only the review recipient can enforce the penalty.

2. Penalty Calculation

• Each author maintains a Spam Penalty Count (SPC) tracking confirmed downvoted positive reviews.
• Penalty follows the Fibonacci sequence: F(n) where n = SPC + 1.

Spam Review #	Fibonacci F(n)	Incremental Points	Cumulative Points
1st	0	0	0
2nd	1	1	1
3rd	1	1	2
4th	2	2	4
5th	3	3	7
6th	5	5	12
…	…	…	…

3. Penalty Categorization

• Score deductions are attributed to the Vote Impact category (name subject to change).

4. Review State Management

• Spam-marked reviews maintain their original metadata but gain a spam_status field.
• Authors can view their spam-marked reviews and restore them as non-spam (by removing their downvote); any associated penalty is then removed.
• No appeal mechanism is provided in the initial implementation.

Considerations & Safeguards

• False-positive risk: legitimate positive reviews marked as spam. Mitigation: penalties start at 0, giving authors the benefit of doubt on the first offense.
• Brigade defense: authors choose to leave positive reviews for recipients; if recipients coordinate to penalize an author, that is an acceptable outcome of leaving too many positive reviews for people you don't actually trust.
• Reputation washing: this mechanism cannot be used to suppress negative reviews to falsely increase one's score.

Future Considerations

This EIP focuses solely on mechanism design within the protocol and scoring system. However, we can apply the same mechanisms in the front-end of the product to hide these reviews from user profiles if the recipient downvotes them.

We may also choose to change the downvote button label to “Mark as spam” to make it easier to understand; however, enforcement will continue to be driven exclusively by the voting system.