## 19 — Channel-aware Jev moderation and complete review logs

**Decision: keep as a user-requested major feature.** Check player-authored communication quickly with Jev before publishing it. If a message appears offensive or breaks the relevant channel's purpose, ask the writer whether they are sure they want to send it. If they confirm, send it, save the original message and decision, and draw an administrator's attention to it for review. The writer can instead cancel or edit the message. Jev supplies a fast assessment; a human administrator makes the final judgment.

### Channel-specific evaluation

Evaluate offensiveness in every player communication channel, including in-game speech, tells, group and public channels, forum posts, chronicles, and messages produced by a PG's script. Also evaluate whether the text fits the channel's purpose:

| Channel or space | Additional context check |
| --- | --- |
| In-game `say` | In-character roleplay is expected. Flag likely out-of-character or technical discussion, such as talking about software, for confirmation and review. |
| Other in-world channels | Apply the roleplay expectations assigned to each channel; staff can configure them explicitly. |
| Out-of-character forum or support | Technical and real-world discussion is allowed; check offensiveness and the space's own rules. |
| In-character chronicles and event threads | Check fit with the story or event context as well as offensiveness. |
| Private tells | Check offensiveness; whether a tell is in-character depends on its explicit channel or setting. |

The check should consider the original language, context, quoted material, and the channel's rules. Translations can assist review but must not replace the original. Jev should assess predefined concern categories and return uncertainty information; the game can render a localized explanation from the category using an authored template. Jev does not generate free-form explanations. Exact categories, thresholds, and channel settings need playtesting in the supported languages.

### Prompt, delivery, and review

For a likely concern, show a short, specific prompt before the message is sent: for example, “This looks out of character for `say`. Send it anyway?” The writer can revise, cancel, or confirm. A confirmed message is delivered under the normal channel rules and placed in an admin review queue with its reason and context. Admins can inspect the message, surrounding conversation permitted to their role, the Jev result, the writer's choice, and subsequent actions. They can mark a false alarm, resolve it, or take the applicable moderation action.

The flow must work in both web and Telnet, and it must handle a script's attempted message without an unavailable human confirmation. A server-hosted NPC-mode script cannot silently answer “yes” to its own moderation prompt; the owner must review or an agreed script-specific rule must handle the held message. External programmable clients follow the normal authenticated prompt and logging path. If Jev is unavailable, the system records that no assessment occurred and follows an explicit fallback policy while keeping the channel usable.

### Complete, controlled logs

Durably record all player communication and moderation events needed to reconstruct what happened: original text, channel, sender account and PG, audience, time, active control mode, any translation shown, Jev assessment, confirmation or cancellation, delivery result, admin review, and action taken. A canceled attempt is recorded as an attempt, not as a delivered message. Keep logs protected from casual access, searchable and properly viewable by authorized administrators, with an audit trail of who viewed or changed a case. Define retention, access levels, and export rules before launch; no log entry should disappear silently.

The review interface should let an authorized admin follow a conversation in order, filter by channel, PG, case status, and time, and distinguish original text from machine translations. Normal player-facing forum or chat history still obeys its own visibility rules; admin log access is an explicit staff permission.

### Decisions and open design

- Keep: **Jev pre-send evaluation, a writer confirmation prompt, delivery after confirmation, and admin attention for confirmed concerns**.
- Keep: **offensive-content evaluation everywhere; in-game `say` additionally expects roleplay and flags technical or out-of-character talk**.
- Keep: **durable, complete, searchable communication and moderation logs with controlled admin access and an access audit**.
- Still to choose: **thresholds, channel policy details, exact fallback when Jev is unavailable, case retention, and the treatment of held script messages**.