One reaction kind, author-only aggregate, no public counts

Reactions in social media exist to drive algorithmic feeds and to broadcast popularity. We have no algorithm and we refuse to broadcast popularity. So what would a reaction be for?

Three legitimate purposes survive that filter:

1. Author feedback. Quiet acknowledgement that something landed, like a "received" signal in a long letter. Author-only visibility of aggregates removes social-proof manipulation of other readers.
2. Reactor's own filter. The reactor's own past reactions constitute a private filter ("show me what I appreciated").
3. Substrate connection. Already supported via references_atom_id (re-share / quote as new atom).

What we explicitly refuse:

• Public counts. No "12 people liked this" on any card.
• Multi-axis feedback (useful / clarifying / valuable / didn't land). Tempting because semantic, but each axis still asks the reactor to score the atom — that's engagement-metric psychology in semantic clothing. Start with one bit. Add nuance only if the data shows a clear gap.
• Counts visible to the reactor. Even showing "you've appreciated 47 atoms" gamifies it.