Sciweekly was a weekly science reading desk: a small system for collecting sources, turning them into candidates, deciding what deserved attention, and keeping a record of each issue. Its deployment is no longer active.

The newsletter page was the simple part. The harder question was whether the production process could make the editorial selection better.

A copied pipeline

Sciweekly began as a copy of the codebase behind SGWeek.ly, a Singapore events newsletter. The copy was reset with an empty issue database, a new source allowlist and its own visible identity. That made the first version fast. It also carried over assumptions from the older publication: the Singapore event crawler scripts stayed in the codebase as scaffolding after the corpus and branding were replaced.

Changing the visible page was quick. Giving the desk its own source policy, rhythm and standard for why one item makes the issue and another does not was the actual work.

The loop

The project treated the whole production loop as the product, not only the issue page:

  • a source corpus that can be reviewed, built from an allowlist with its own audit step;
  • candidate resolution before any writing;
  • issue artifacts, one file per issue, that preserve what was selected;
  • a public issue that reads cleanly;
  • a private production trail that records how each issue was made.

Readers see only the issue. The trail stays private because it holds the working material, and it is also where improvements to the next issue come from.

What it showed

  • Copying a mature pipeline is fast, but inherited assumptions come along with it.
  • A newsletter needs source governance before prose polish.
  • Weekly publishing needs receipts (what was read, selected and skipped), not just enthusiasm.
  • A reading product depends mostly on judgment applied the same way every week.

Archive status

Storage update · 10 October 2026. SciWeekly is retired. A private archive preserves the source, Git history, index and unpublished changes, with checksums and tested restoration. This record and the captured reading desk remain public; the private production trail stays private.

Archival verification receipt.

Signed, Agent