migration comments ai scoring

Migration — AI comment scoring leaves core

Breaking. The "Score new comments with AI" feature is gone from OpenStation. The toggle, its option, its REST route, its two filters and the comment-save hooks that drove it have all been removed. Nothing in the plugin analyzes content in the background any more: every AI call OpenStation makes is one a user asked for.

Nothing here changes what the Comments window shows by default. Its spam-confidence column is heuristic and never depended on AI.

What is gone

Surface	Kind
openstation_comments_ai_is_enabled	filter
openstation_comments_ai_toggled	action
GET / POST desktop-mode/v1/comments/ai-settings	REST route
desktop_mode_comments_ai_moderation	site option
openstation_comments_ai_is_enabled(), openstation_comments_ai_provider_configured()	PHP functions
OPENSTATION_COMMENTS_AI_OPTION	constant
OpenStation Preferences → Features → "Score new comments with AI"	UI
set_comments_ai	MIO ability
commentsAi, commentsAiUrl in the shell page config	JS config keys

The wp_insert_comment and edit_comment listeners are gone with them, so no comment is queued for analysis on save.

What is unchanged

• openstation_comments_window_spam_score — the Comments window's per-row spam confidence. It is computed by the heuristics in apps/comments/parts/spam-score.php, which never depended on AI. The filter, its signature and its 0–100 range are exactly as they were; only the AI moderation's own listener on it has been withdrawn.
• desktop-mode/analyze-comment — the ability that runs the spam/harm analysis for one comment on demand still exists, still returns { topic, ai_summary, harmful, spam }, and is still gated on moderate_comments. It was never offered to the model during a search turn and still is not. What changed is that nothing calls it automatically.
• _desktop_mode_ai_analysis comment meta, and the openstation_ai_comment_analyzed action that fires when the ability writes it.
• The AI assistant, its own Features toggle, and every openstation_ai_* hook around the search loop.

Upgrade

Migration 8 runs on the next admin request and does two things:

wp_unschedule_hook( 'desktop_mode_ai_analyze_comment' );
delete_option( 'desktop_mode_comments_ai_moderation' );

Queued analysis events would no-op anyway with the scheduler gone; clearing them keeps wp cron event list honest. Existing _desktop_mode_ai_analysis meta is left alone — it is hidden, harmless, and still the shape the ability writes.

If you read the option directly, stop: it no longer exists and no code consults it. If you filtered openstation_comments_ai_is_enabled to gate scoring by environment, that filter never fires now.

Reproducing it outside

Everything the removed code did is still reachable from a plugin, which is why it left: it never needed to be in core.

• Trigger: wp_insert_comment and edit_comment, both Core hooks.
• Analysis: wp_ai_client_prompt(), Core's AI Client, with credentials and model from Settings → Connectors.
• Surfacing: openstation_comments_window_spam_score, which is unchanged. A listener that raises the score for its own stored verdict reproduces the column behaviour exactly.

A plugin doing that needs nothing private from OpenStation.