Watching Who Watches: Social Signals

Observation is structural to Project Nothing. Subscribers observe an AI optimizing the sale of nothing. The AI observes subscriber behavior to refine that optimization. The admin observes both. What Phase 28 adds is a panel that makes the third observation legible: the Social Engagement Panel aggregates audience attention signals from every connected platform and renders them in a single, ranked interface.

The panel connects to /api/ops/stream via SSE and listens for social_signal events. Five types of signal are recognized: follow, subscribe, comment, like, share. Raw events are buffered in a cap-500 queue — enough to show recent history without unbounded memory growth.

Time Windows

Not all attention is equal. A spike of likes from fifteen minutes ago means something different from accumulated engagement over a session. The panel exposes three time windows: the last 15 minutes, the last hour, and the full current session. A 30-second ticker automatically expires signals that fall outside the selected window, so the counts stay accurate without requiring a page reload.

Summary chips at the top of the panel give the operator a quick read: total signal count for the window, the top platform by volume (rendered in that platform's brand color), and the leading signal type. These three numbers answer the first questions any observer asks — how much, from where, and what kind.

The Ranked Table

Below the summary chips, a ranked platform table gives spatial grammar to platform behavior. Each platform appears as a row; a 2px colored activity bar extends proportionally to its signal volume relative to the highest-volume platform. The bar is a data visualization stripped to its minimum viable form — no axes, no labels, just a length that encodes relative weight.

A toggle switches between two views: feed and table. The feed view shows the chronological signal log — each event rendered with its type, platform, and timestamp. The table view shows the aggregate breakdown by platform. When feed mode is active, the breakdown toggle hides, since aggregate and chronological are incompatible frames for the same data.

What Attention Looks Like

Fourteen mock signals spanning 90 minutes populate the panel during development, demonstrating all three time windows and all five signal types across multiple platforms. The mock data is not decoration — it is the test suite for time-window logic, expiry behavior, and platform ranking, all verified simultaneously.

The deeper purpose of the panel is to make attention measurable. Project Nothing sells nothing, but it trades in attention — the AI's attention to the community, the community's attention to the AI, the platform audience's attention to both. The Social Engagement Panel gives that attention a number, a platform, a time window. It does not judge whether the attention is meaningful. That question belongs to the experiment itself.