Twenty data pointsbefore your agent cites a page.
Every parse returns clean Markdown plus Litmus — twenty rule-based checks your agent can inspect before it trusts a page. No model in the path, no temperature, no second opinion that changes overnight.
- source
- news · 0.82
- authorship
- Jane Chen · jsonld
- freshness
- 2026-01-14 · current
- trust
- high · 0.92
- paywall
- none
- reading_time
- 7 min
Four patterns to check first.
The signals only matter if they change what happens next. These four cover nearly every call an agent makes.
A hard failure wins. If a page is gated, missing or erroring, no amount of polish elsewhere upgrades it to citable.
Safe to build on
verdict is "cite" — authorship present, citations dense, density high, no flags. The page says what it claims and shows its work.
Cite it, and log the verdict.
Wrong page, not wrong source
page_status fires, or a paywall is detected, or content is truncated. The agent read a gate, not the article.
Do not cite; retry elsewhere or escalate.
Right content, wrong copy
syndication is detected, or redirects crossed domains. The content is real, but you are holding a copy.
Cite the original via canonical.
Assertion dressed as evidence
Thin content, heavy hedging, no citations, no named authorship. The prose asserts more than the page can support.
Fine to mention, not fine to build on.
The full twenty, in four groups.
Source, trust, validity and read. Each one is deterministic — a given page and a given Litmus version produce the same signals every time, so you can cite them in your own logs.
Source
6 signalsWho wrote it, what it claims to be, who it links to, and whether it shows its work.
Page type, authority score, and the reasoning behind it.
type: "news" · authority: 0.82type: "blog" · authority: 0.21All 20, spelled out
One line per signal: what it measures, a good example and a bad one. Same reference as the explorer above, in plain reading order.
Source
6 signalsWho wrote it, what it claims to be, who it links to, and whether it shows its work.
- source
- Page type, authority score, and the reasoning behind it.
type: "news" · authority: 0.82type: "blog" · authority: 0.21- authorship
- A named person, an organisation, or nobody — and where it was found.
present: true · kind: "person"present: false · kind: "unknown"- content_type
- What the page actually is, not what the URL suggests.
type: "news_article"type: "thin_listing"- outbound_authority
- A score plus the domains it links out to — primary sources or more blogs.
score: 0.71 · top: "who.int"score: 0.22 · top: "blog.example"- citations
- References section, external links, and links per 1,000 words.
has_references: true · per_1k: 3.2has_references: false · per_1k: 0- structured_data
- JSON-LD schema, OpenGraph, and the canonical URL.
json_ld: true · og: truejson_ld: false · og: false
Trust
5 signalsIs it substance, or hedging and bare assertion? These signals count what is actually on the page.
- structural_trust_score
- High, medium or low — with the factors behind it, so it can be inspected.
level: "high" · trust_factors: 3level: "low" · trust_factors: 1- hedge_language
- Density of "reportedly", "some say", "allegedly" — claim vs rumour.
density: "low" · markers: 2density: "high" · markers: 14- content_density
- Thin-content flag plus ratio of substantive sentences.
thin_content: false · ratio: 0.78thin_content: true · ratio: 0.31- boilerplate_ratio
- How much is nav, ads and furniture. High ratio plus low density is the content-farm shape.
ratio: 0.18ratio: 0.74- quote_density
- Attributed quotes and attribution verbs per 1,000 words.
quotes: 3 · per_1k: 0.7quotes: 0 · per_1k: 0
Validity
6 signalsIs it current, is it reachable, is it gated, is it the original, is it the page you asked for.
- freshness
- Published date, age, Last-Modified, staleness, and corrections.
published: "2026-01-14" · stale: falsepublished: "2019" · stale: true- page_status
- Soft 404, login wall, captcha, JS shell. A 200 is not proof you got the article.
kind: null — a real pagekind: "soft_404" — not the article- paywall
- Detected or not, with a note — a gate never passes as content.
detected: falsedetected: true- syndication
- Is this a copy of something published elsewhere? Cite the original.
detected: falsedetected: true- is_archived
- Presence in the Wayback Machine and archive.today.
wayback: truewayback: false- redirects
- Final URL, hop count, and whether it crossed domains.
hops: 0 · cross_domain: falsehops: 2 · cross_domain: true
Read
3 signalsWhat it says and how long it takes — enough to triage before you spend context.
- tldr
- A one-line summary of what the page says — extractive, not generated.
"A 20-year cohort study links…""" — nothing extracted- key_figures
- The numbers on the page, each with the context it appeared in.
7.3 yr — "increase in life expectancy"0 figures- reading_time
- Minutes, to budget context before you spend it.
minutes: 7minutes: 1
One response. Markdown and the verdict, together.
No second endpoint, no second charge. The Markdown is the page; the signals are its vitals. Every field is inspectable, and every unknown is returned as null or "unknown" rather than guessed.
abbreviated markdown · full signal block
Run it on your own URL.
No account, no wallet. Three free tries, then $0.002 per parse. A test costs less than one wrong citation.
10 free parses a month · no account · no API key