Live telemetry
The engine
PRISM runs itself. This page is the honest version of that claim — the last cycle, source by source.
- Corpus
- 0
- Sources
- 0
- Reactions
- 0
- Cycle
- 1.74s
01
Source health
Next refraction in 30m. Refresh happens on a cron tick, or on the first visit after the content goes stale.
02
One cycle, six steps
No database server, no queue, no cloud bill. A single PHP process, a flat-file corpus and an append-only counter log.
-
01
Pull
Every source is fetched in parallel with a hard timeout. A dead endpoint is skipped, never fatal.
-
02
Normalise
RSS, Atom and five JSON shapes collapse into one record: title, canonical URL, summary, author, timestamp.
-
03
Dedupe
URLs are stripped of tracking junk and hashed; near-identical headlines across sources fold into one story.
-
04
Classify
A keyword model assigns a band — intelligence, science, space, security, engineering, design, hardware, culture.
-
05
Rank
Source authority + reader reaction, divided by age to the power of 1.45. Nothing stays on top by inertia.
-
06
Age out
Past the retention window stories leave the corpus, counters compact, and the cycle starts again.