How to read a result
The analyzer returns six fields, and every one of them is narrower than it looks. This page says what each measures, what it refuses to measure, and the thresholds it applies before it will say anything at all.
What comes back
- Framing strength
- None, slight, moderate, heavy, or uncertain.
- Direction
- Toward or against a named subject in the story, or non-directional.
- Confidence
- 0.00 to 1.00. How consistently the signals repeated.
- Quoted excerpts
- The journalist's own sentences, each tagged framing, language, sourcing, or attribution.
- Article type
- News, opinion, or analysis.
- Other coverage
- Articles on the same story from other outlets, never the one you submitted.
There is no left, centre, or right
Direction names the subject the framing points at, not a political side. A result reads “toward Honeywell”, or “against the plaintiff”, or “non-directional framing”. It never reads “left-leaning”, because that is not a value the analyzer can return.
That is a deliberate limit, not a missing feature. Placing an article on a political spectrum requires a fixed idea of where the centre sits, and one piece of text gives no way to establish that. Naming the subject the language favours is something you can check against the article in front of you. Placing it on a spectrum is not.
Nothing here rates an outlet. The unit is one article, and the same publication returns different results on different stories.
Confidence measures consistency, not truth
This is the number most often read backwards. It describes how consistently the signals repeated across the text. It is not a measure of how strongly framed the article is, and not a measure of how certain anyone should be.
A well-written opinion column can return high confidence precisely because its structure is deliberately consistent, and that is not a finding against it. A low number is not a clean article either. It is an article the analysis is less sure about, and saying so is more useful than rounding to a confident answer.
What it does when the evidence is thin
- 0.40
- Below this the result is none or uncertain
- 0.85
- Above this is rare
Four rules run before a result is produced, and all four push toward saying nothing.
Minimum impact
Signals that are minor, isolated, or purely stylistic set framing strength to none rather than being reported.
Weak evidence
Where the evidence is thin or ambiguous, the result is none or uncertain, and confidence is held below 0.40.
High values are rare
Above 0.85 is reserved for framing that is clear, repeated, and explicit in the text.
No excerpt, no finding
When framing strength is none, the excerpt list is empty. There is no path that returns a finding with nothing to show for it.
The analyzer runs at temperature zero rather than sampling. That is what makes a second run comparable to the first.
Where the analysis has limits
Messy input
Very short passages, failed extraction, navigation text, or blocked pages can weaken the read significantly.
Rhetorical edge cases
Satire, irony, or unusual style can resemble framing signals even when the intent is clearly different.
Scope of the result
The result reflects patterns in the submitted text, not a universal judgment on the outlet, the author, or the topic.
Reader judgment
The analysis is one data point. It works best as a prompt to slow down and inspect, not as a substitute for forming your own view.
Run one and read it against this page
The fields make more sense with a result in front of you.