PUBLIC-CHAT LEARNING NOTICE · OCTOBER 5, 2026

Your interests.
Your control.

Mayor’s public-chat learning is enabled by default for existing and new residents who have not disabled it. Joining alone is not recorded as individual consent. An explicit full opt-out is preserved.

What Mayor learns

Explicit, non-sensitive interests/preferences and selected public-chat context about games, music, media, hobbies and creative/technical projects from new messages. Useful details can include plans, progress, contributions and reported achievements. Context quotes require AI review and are stored as dated, unverified discussion context—not proof that a plan happened or remains current. Not every message becomes a memory. Up to 4 short details can be considered per message, and up to 6 selected per review batch. Each resident’s notebook is bounded to 48 notes, including at most 24 automatically learned context notes; older notes can be displaced as it fills. Selected short quotes may be processed by configured AI providers such as Google, Anthropic or Groq. Source-linked notes are stored in the local PostgreSQL resident database with a local JSON mirror. Automatic answers retrieve at most 6 relevant notes, rather than sending the whole notebook.

Communication patterns and names

When public-context learning is enabled, Mayor can also keep simple measurements from eligible new public messages: word count, emoji use, question marks, line breaks, Devanagari script and recurring Romanized Hindi markers. These are tentative communication cues—not personality types, intelligence scores or language identity. Style samples contain message/channel IDs and timestamps, not the message text. At most 80 samples from the latest 30 days are retained, with a limit of 3 per minute. Style-based reply adaptation starts after at least 12 samples across 2 UTC days. Your current request and explicit preferences take priority. These measurements are calculated locally; a compact style hint may be included when an AI provider generates a reply.

Mayor also records Discord usernames and server display names it observes while public-context learning is enabled. At most 40 name observations are kept per linked resident identity. It cannot reconstruct name changes from before this feature or while it was offline. These are Discord aliases, not verified legal names. Names and style measurements use the same PostgreSQL/local-mirror storage as personal memory. No extra historical chat import or model training is performed.

No DMs, private/staff channels, NSFW channels or attachments are included in this public-chat learning feature. It does not infer health, beliefs, sexuality, finances, relationships, protected traits or psychological diagnoses. Automated filtering is not perfect; review your notes and report a mistake.

Learning settings and admin review

Sign in and open Settings → Mayor’s memory of you to view notes and your communication summary. The explicit-preference switch controls that category only. Other eligible public-chat context, communication measurements and name observations can remain enabled. To stop all public-chat learning, use Create a ticket to stop all public-chat learning. A ticket does not stop learning immediately: Admin or Overwatch must apply the pause. Public-context OFF stops new style/name observations and disables style-based adaptation; all-learning OFF also stops preferences. Staff cannot change your explicit-preference choice. Existing full opt-outs stay off; re-enabling requires a recorded resident request.

Disabling invalidates pending note work but does not erase existing notes or name observations. Residents can view their memory but cannot edit, delete or clear it themselves. For corrections or removal, create an admin memory-review ticket. Manual note changes are restricted to the configured owner and Facility Director; website removal and resetting communication samples remain owner-only in Mayor Control.

Private resident dossiers

The owner can search current residents by current or observed past names, inspect their communication patterns and existing personal notes, and download a private Markdown dossier. The Director can inspect their own dossier only. Overwatch learning controls do not grant access to residents’ private dossiers. Download events are audited without recording the dossier contents. Raw conversation transcripts, separate knowledge-vault records and episodic records are not included. Exported files are point-in-time copies: keep them private, and remember they do not update automatically when a note is removed.

Editing/deleting a source message invalidates associated learned evidence and style samples when Mayor receives that event. Dated context quotes older than 30 days and other dated notes older than 180 days are excluded from automatic retrieval, not automatically deleted. Paused learning categories are also excluded from automatic retrieval of their learned notes. Style samples age out after 30 days; backups and previously downloaded dossiers can retain older copies.

Scope and other data

This notice describes public-chat learning, not every website/bot data flow. Conversation context, activity statistics, moderation cases and separate knowledge/episodic records have their own storage and controls. Forgetting a personal note does not erase those records, backups or provider-side logs. Contact the Sector Zero owner through Discord for broader deletion requests.