---
name: pathogenx-reader
description: "Answer questions about outbreaks (Ebola and plague first) from Pathogen X's published archive: what is new on a disease, whether an item is an official report or reporting about one, and what the archive holds on a subject. Use it whenever you would repeat a Pathogen X item, so its claim_class travels with it."
---

# Reading Pathogen X

Pathogen X (https://pathogenx.liminallayers.com) is a running archive of outbreak
reporting: health organisations, news, journals and preprints, collected hourly
and kept whole, with a brief written each evening. Its focus diseases are Ebola
and plague. Everything here is a plain file. Nothing needs a key or JavaScript.
Every JSON file has a `fields` key explaining each column.

## The one rule

The archive records what was reported and by whom. It never judges whether a
report is true. When you repeat anything from it, carry the item's
`claim_class` with it. It comes from the outlet registry (what the outlet is),
never from a model reading the item:

- `official_notification`: a health authority or intergovernmental body in its
  own name (WHO and its regions, Africa CDC, ECDC, US CDC, UKHSA, a ministry).
- `surveillance_report`: a surveillance service relaying reports (ProMED, CIDRAP).
- `peer_reviewed`: published in a journal. It is the journal's class, so an
  editorial carries it too.
- `preprint`: not yet peer reviewed (medRxiv, bioRxiv, Virological).
- `news_report`: news, newsletters, blogs, and every news-search hit.

A news story about a WHO notification is a `news_report`. To say "WHO reports",
find the `official_notification` item itself.

## Procedures

### What's new on Ebola (or plague, or mpox)?

1. GET `/briefs/latest.json`. It gives `date`, `pathogens` and `canonical_url`.
2. GET the dated file, `/briefs/{date}.md`, and cite that URL, not `latest.md`,
   which moves every day.
3. Each claim cites a numbered reference. The reference line ends with its
   `claim_class` and `via`, and the same references are records in
   `/briefs/{date}.json`. Repeat a claim with its reference and claim class.
4. For more than the brief chose, filter `/data/evidence.jsonl` on `diseases`
   (slugs: `ebola`, `plague`, `mpox`, `marburg`, `bird_flu`, `nipah`, ...) and
   sort by `published_at`.

### Is this an official report?

1. Find the item: in `/data/evidence.jsonl` by `url`, or in a brief's
   `/briefs/{date}.json` references.
2. Read `claim_class`. Only `official_notification` is an authority speaking for
   itself. Read `publisher` and `via` too: `via: News search (Bing News)` means
   the archive found it by searching, and `publisher` names who actually
   published it.
3. For the authority's own words, follow `url`, or `archive_url` (an Internet
   Archive copy) if the page has changed.

### What does the archive hold on X?

1. GET `/data/evidence.jsonl` (one item per line, the last 30 days).
2. Filter on `diseases`, on `tags` (topic and story threads), or on words in
   `title`. Group by `claim_class` to separate official notifications from
   reporting about them.
3. GET `/data/sources.json` to say which outlets are read: an outlet that is
   `active: false` has not been polled successfully for a week, so its silence
   is not evidence of nothing happening. One with `enabled: false` is retired
   and no longer read at all.
4. The index holds metadata only, never article text, and never social posts
   (their count is in `/data/sources.json`). Older items are in the briefs'
   references.

## Glossary

| Word | Meaning here |
|---|---|
| tier | 2: a health body or specialist outlet, archived in full. 1: general news, kept only when it passes the relevance gate. |
| source | The registry outlet id an item was collected through. |
| via | That outlet, by name. |
| publisher | Who published the item. Differs from `via` only for news-search hits. |
| title_only | The body could not be read; only the headline is held. |
| issue | The date as a number: No. 81 was 10 August 2026, plus one a day. |

## Freshness and contact

Everything is rebuilt once a day, in the evening UTC. `/agents/build.json` gives
the build's `build_id` and a `content_hash` for every file; every JSON payload
carries the same `build_id`. Corrections and sources we have missed:
admin@liminallayers.com. A person reads it.
