@ecency votes with millions of Hive Power, nearly all of it delegated by its supporters. It returns 100% of the curation reward to them daily. Curators are volunteers deciding where that vote goes; the desk is the shared queue they work from.

A few curators are followed. When one of them votes a post, five Ecency accounts vote it in one transaction within seconds.

Conditions: the post is inside its payout window (the desk shows the scale factor in the last 12 h because the chain scales those votes linearly toward zero), none of the five has voted it, no hivewatchers downvote, the author is not on the ignore list, the bot is not paused and @ecency VP is above 55%. The desk shows not trailed with the reasons it can derive.

@ecency, @ecency.stats, @esteem.app, @good-karma and @esteemapp vote together, all at the same weight. @ecency's VP alone drives that weight; the budget is shared by all five.

The base weight is 9.3% for every front-end, scaled by the bot from @ecency's VP between 55% (zero) and 100% (full). At 82% that is about 5.6%; every 1% of VP moves the trail weight by about 0.21%.

This ramp is the bot's budget rule, not a chain rule: since HF28 the chain pays a vote of a given weight the same rshares at any VP and only refuses the vote when mana is short. Vote value does not depend on voting power. VP is a budget that refills 20% a day; a full vote costs 2% of it. About 179 votes a day at 5.6% is the sustainable pace.

Own posts of the curator and moderator accounts get a flat 10%: do not push them. Comments trail at the same weight.

Full curation weight inside 24 h of the post's creation, half from 24 to 72 h, one eighth after 72 h, scaled linearly toward zero in the last 12 h before payout (92% at 11 h left, half at 6 h, zero at the cashout second).

There is no early-vote penalty (the reverse auction is 0 since HF25) and order inside a window does not matter: two curators who both vote a post inside 24 h split the pool by the rshares they added, not by who was first. So never race the first minutes. Read the post, vote inside 24 h. Nearly all trail votes already land there; the badge on every row counts down to the boundary.

This is team etiquette, not a bot rule: the bot does not enforce it today. The desk shows author trailed N h ago at X% so you can choose the author's best post rather than vote every one.

The header shows @ecency VP, the implied trail weight, the sustainable votes per day at that weight and the mana spent today in vote-equivalents, trail and other separately. Other Ecency bots spend the same mana (the poster bot's self-votes and the flat 10% own-post votes are the other part), which is why the line counts mana and not trail votes.

More votes a day lower @ecency's VP, which makes the bot cast smaller trail votes for everyone; fewer votes let VP climb toward 100%, where regeneration is wasted. The sweet spot is spending what regenerates: 20% of mana a day. Keep the budget line green.

• Original writing or images with visible effort.
• Engagement: the author answers replies.
• On-topic for its community, readable formatting.
• First posts from new accounts (post #1 to #5 are marked; a comment goes further than a vote).
• Posts from every front-end, not only Ecency.

• Plagiarism: search a distinctive sentence, reverse image search.
• AI slop: generic structure, no personal detail, contradictions with the author's earlier posts (the drawer shows them).
• Recycled content, image-only posts, tag or community abuse.
• Reward farming clusters: the same accounts voting and recommending each other.

The abuse list warning is a list we display, not a verdict.

Formulaic is a boilerplate score from an old detector: today's AI text scores 0 to 10% and genuine people often score 80% or more, so it proves nothing. Style shift compares an author with their own past and alerts rarely. Images on Hive shows uploaded versus collected images. Replies per day shows conversation.

Never flag or comment on a number alone. Numbers prompt a read, they never decide.

When hivewatchers downvotes a post the five accounts unvote automatically. Do not re-vote; raise disputes with the team on Discord.

• No own posts through the trail, no daily friends.
• Comment on first posts.
• Mark honestly: your marks are the team hand-off, so a card you did not read is not reviewed.
• Flag instead of arguing; no public call-outs.
• Recommendations are on chain and cost Resource Credits; count networks, not accounts (other apps count as one network); dismiss a ring instead of arguing.

Anyone can recommend a post to the curators. The recommendation is written to the Hive chain with your posting key, so the record is public.

Every recommender carries a weight from the last 90 days. Recommendations that curators dismiss lower your weight; ones that get curated raise it. Everyone starts at x1.0 and stays there until five recommendations exist, so a new recommender is neutral rather than suspect. Ten or more recommendations at x1.2 or better earn a Trusted recommender chip next to the name.

Open the recommendation badge on any post to see who recommended it, why and how each of them has done before. Your own record sits in the recommend dialog under the reasons.

Weight changes the order of the recommended list only. It never hides a post: the raw counts are always shown.

Four weeks as a trial curator with desk access and untrailed votes. Trial curators mark, snooze and flag like everyone else but do not see other curators' note bodies or flag reasons until promoted (those notes name authors and allegations). Then a review of reviewed-per-hour, flag precision and overlap.

How many hours behind the queue is plus what got curated are public; reviewed-per-hour and marks per curator never are.