mason v0.3.0

Minor Changes

• #11 e51775c Thanks @antstanley! - Streamplace video replaces Steam. The wall now carries atproto livestreams
  from stream.place: archived streams from the people you follow (a 90-day
  window, since an hours-long stream stays worth watching long after a skeet
  about it would have expired), and anyone who is live right now.

  A live stream is the only brick with a deadline, so it is the only one that
  jumps the queue: it opens the wall, wears a LIVE badge and a viewer count, and
  never ages out while it is running. Everything else is unchanged; video is
  still click-to-play, and still never autoplays.

  Steam is gone entirely. Its storefront API served no CORS headers, so trailers
  never worked in the browser build at all; Streamplace is CORS-open, which means
  the no-server build now reads exactly what the native one reads.

  A wall that actually arrives. Chasing the live-stream work turned up three
  things that were quietly starving the first page, and they are fixed here:

  • The follow graph was fetched to completion before a single post was, and
    follows page 100 at a time with each request blocking the next. Someone with
    2000 follows waited ten seconds for a list nobody asked to see, and their
    first wall came back empty. The wall is now built from a head start of
    300 follows (the cohort only samples 100 authors anyway) while the rest of
    the graph is fetched behind their back for next time.
  • standard.site documents refetched their publication record once per
    document, so one blogger cost 25 sequential requests. A blogger has one blog:
    it is fetched once now. The repo fan-out went from 21s to under 4s.
  • Posts and repo reads shared a task per author, so an author's posts waited on
    plc.directory and two PDS reads before being admitted. They are fanned out
    separately now, and the per-author brick cap is per KIND, so a prolific
    poster's own blog is no longer turned away by a quota their skeets ate.

  Infinite scroll that keeps going. The wall could stop dead with a cursor
  still in its hand. IntersectionObserver fires on a change of intersection,
  and a page that came back short did not grow the wall enough to push the
  sentinel back out of its prefetch margin, so no second event ever arrived and
  the scroll ended there. The wall now pulls rather than waits to be told: it
  keeps laying while its bottom is within reach. And the AppView burst is raised
  from 40 to 100 (the 10/s sustained ceiling is untouched), so a cold cohort goes
  out at once instead of dripping, and a reader can no longer out-scroll their
  own wall.

• #9 05fa72e Thanks @antstanley! - An atmosphere client picker, and a wall that is actually fresh.

  Open posts in the client you use. bsky.app, mu.social and blacksky.community share a URL structure, so the picker in the header rewrites the host and nothing else. Blog and stream links are left exactly as they are, because they are not Bluesky posts. The choice persists.

  A stronger recency bias. Posts and Bluesky videos now live for 72 hours, blogs for 14 days, and nothing older is admitted to the wall at all: a hard window, not a soft preference, because decay alone leaves week-old content eligible and on a quiet follow graph it surfaces. Half-lives are steeper to match (posts 12h, blogs 3d).

  A wall could belong to one person. First paint gated on the number of bricks in the pool, and a single prolific account returns thirty of them, so the wall could open before anyone else's feed arrived. It now gates on distinct authors, no author may hold more than four bricks, and when the diversity rule truly cannot be honoured the mixer falls back to the least represented author rather than re-picking the loudest.