This project began as a straightforward utility: a Bluesky feed that would reliably surface a user’s posts made “on this day” in previous years.
That is still the goal I’m working towards.
In practice, the current Bluesky APIs make this difficult to achieve on demand for prolific users with thousands of posts - the very kind of user that would find this feed most interesting.
Rather than treating this purely as a technical failure, this project explores what happens when partial recall and delay are treated as part of the experience rather than an error.
- The feed looks for posts you made on this day in previous years (±24 hours).
- Post history is scanned in small, bounded slices to respect platform limits.
- if a user's full post history is not retrieved within the request limit, results are cached and progressively expanded over multiple visits.
As I struggled through various approaches like jumping cursors, targeted searches — it became clear that, given the current platform constraints, completeness and immediacy could not be guaranteed for everyone.
The most reliable solution I could find was progressive recall: caching partial results and gradually expanding them over time, allowing users to refresh or return and discover more of their history without exceeding platform limits.
From this, I became interested in the idea of memory itself, some things surface easily; others are buried deeper. This project became an artistic reflection of memory’s fragility.
- The core Cloudflare Worker logic for the feed generator
- Progressive scanning and caching logic
Infrastructure files (Wrangler config, KV bindings, environment variables) are intentionally omitted.