The arithmetic for one report
Trump defends Jensen Huang and Pete Hegseth at surprise news briefing with Dana White
91Few flagged expressions
What was read
430 words in the outlet's own voice, plus 14 in the headline. 24% of the text is in quotes, with 2 listed words inside quotes. 13 attribution markers (such as "said" or "according to") found.
Listed words inside quotes (not scored; quoting is reporting, but which quotes to use is an editorial choice): very
, very
.
Scare quotes (a phrase of one to three words in quotation marks with no speaker in the same sentence): smear.
. Since method 3.1.0 their words count as the outlet's own.
Words found
- Word choice:
rigged
attributed to a speaker, not counted Contentious label
thought the results had been rigged . He then repeated his claim of - Raises the temperature:
massive
attributed to a speaker, not counted Raises the temperature
Secretary Pete Hegseth and the massive Republican losses at the ballot box - Attribution phrases:
people familiar
2 points Unnamed source
People familiar with the matter told NBC News
Layer scores
Rates are counted points per 1,000 words of the story's own text. For a story longer than 200 words, the rate used is the average of the whole-story rate and the rate in its densest 200-word stretch, so neutral filler cannot dilute a loaded passage away. Score = 100 × (loaded end − rate used) ÷ loaded end, kept between 0 and 100. Headline words count only in the headline layer.
| Layer | Points | Words | Whole-story rate | Densest 200 words | Rate used | Scale | Score |
|---|---|---|---|---|---|---|---|
| Word choice | 0 | 430 | 0.0 | 0.0 (words 1 to 200) | 0.0 | 0 to 40 | 100 |
| Temperature | 0 | 430 | 0.0 | 0.0 (words 1 to 200) | 0.0 | 0 to 40 | 100 |
| Flagged attribution phrases | 2 | 430 | 4.7 | 10.0 (words 51 to 250) | 7.3 | 0 to 20 | 63 |
| Headline wording | 0 | 25 | 0.0 | headline only | 0.0 | 0 to 240 | 100 |
Densest stretch: Flagged attribution phrases, words 51 to 250 of 430, with people familiar
.
Shown, not scored: the headline's rate minus the rate of the story's first 80 words is 0.0 points per 1,000 words. Experimental: 3 of 3 names and figures in the headline also appear in the text read.
Overall
Weighted geometric mean of the layers that are established, with 10 added to each layer first and taken away at the end: (Word choice 100 + 10)0.4 × (Temperature 100 + 10)0.3 × (Flagged attribution phrases 63 + 10)0.2 × (Headline wording 100 + 10)0.1 − 10 = 91. Adding 10 means one layer at 0 lowers the total steeply without collapsing it. A weak layer pulls the total down.
Lowered the score most: Flagged attribution phrases.
Name-swap test: with party, politician and institution names swapped, this report's score moves by 0.0 points.
What a reader could check
| People named as sources | 0 |
|---|---|
| Unnamed sources | 1 (100% of all sources found) |
| Links to primary records (government, court, filings, PDFs) | 0 |
| Documents cited in the text (a report, bill, ruling, filing) | 1 |
| Checkable specifics (numbers and dates) | 1 |
| Data shown (a table or chart) | no |
| Correction or clarification noted | no |
What a reader could check, counted by published patterns. Shown beside the score, never part of it. How each count works
Checked against the official record
No vote count or bill number found in the text read.
Record of this measurement
Method 3.1.0, word-list fingerprint 807a17189402a416, read by text extractor version 4. Fingerprint of the text read: 82ad9f2b08bc1dc1. Internet Archive copy nearest the time it was read. The text read was not kept (read before method 3.1.0, or from the feed only), so a later rule change can be applied only while the report is still in the outlet's feed.