Real metadata, not invented reporting
Headlines, photographs and expandable abstracts come from sources and
publisher metadata when available. Open Graph is a publisher’s preview,
not fact verification. An abstract may be only a short feed description,
not a summary of the full reporting.
Original articles and discussion threads are distinct destinations.
Hacker News or Reddit conversations should have their own links,
separate from the reporting they discuss.
What retrieval cannot do
The server proxy retrieves public pages. It cannot execute a site’s
JavaScript, bypass paywalls, sign in to a publisher, or guarantee access.
Some sites block retrieval or provide incomplete previews. Failed
sources should show their status and offer retry without removing
the last good edition.
A publicly accessible image or Open Graph tag does not guarantee
hotlink or reuse rights. Images may be blocked, withdrawn or unavailable.
Credits should be retained where supplied; unrelated stock or generated
illustrations must not stand in as news photography.
An edition is an editorial choice
Default sources and political groupings are editorial selections,
not inferred objective truth or reliability scores. Left-leaning
sources are available but disabled initially. You can adjust
individual publications and add your own.
Dates, duplicates & offline editions
Syndicated articles, tracking redirects and inconsistent canonical
links can leave duplicates. RSS publication dates can be missing,
inaccurate or changed after an edit. Fetch time is not publication time.
If live retrieval fails, a dated archive may keep the page useful.
Archive or offline material must be identified as such, never
presented as breaking or freshly fetched news.
Your reading history
Cloud mode stores your preferences, custom sources and reading history
in your own Popchat storage. Guest mode keeps a local fallback on this
device and does not synchronize across devices.
Read stories stay in place while you browse and rank lower in later
unread-first editions. History is bounded: above 100,000 entries, the
oldest entries are pruned. Dispatch can prune only its own storage,
never unrelated user data. Sync failures should remain visible rather
than implying that unsaved changes reached the cloud.
Tools that would improve this
- A bounded batch metadata endpoint with timeouts, rate limits and partial-failure reporting.
- Canonical redirect resolution with provenance and safe URL handling.
- An image proxy with explicit rights, attribution and cache controls.
- A scheduled feed cache with freshness indicators and retained last-good responses.