Reference baseline · v0 · built 2026-07-08

How much persuasive pressure is in ordinary news?

121 articles from 11 mainstream US outlets across the political spectrum, each scored on the 0–10 Manipulation Index. This is the reference corpus every Coverage Audit percentile is measured against — published so the yardstick itself is checkable.

Articles
121
Median MI
3.5/10
Mean MI
3.89/10
Range
0.2–8.0

The distribution

Most everyday coverage sits in the framing band — word choice and selection lean, but sourced. Saturation-grade manipulation is rare in mainstream outlets; that's exactly why a percentile is more honest than a raw score.

0–2 factual
52 (43%)
3–4 framing
32 (26%)
5–6 heavy
28 (23%)
7–8 coordinated
9 (7%)
9–10 saturation
0 (0%)

By lean

Medians by outlet lean. Per-outlet scores are deliberately not published — this page is a baseline, not a ranking.

LeanArticlesMedian MIMost-flagged techniques
Left464.5Loaded Language (43), Transfer (32), Card Stacking (31)
Center312.6Appeal to Authority (29), Loaded Language (27), Card Stacking (17)
Right423.8Loaded Language (34), Transfer (32), Card Stacking (30)

Most-flagged techniques across the whole corpus: Loaded Language (106×), Appeal to Authority (87×), Transfer (80×), Card Stacking (79×), Appeal to Fear (25×).

The corpus

Built 2026-07-08 from the public RSS feeds of 11 US outlets — left: The Guardian, Mother Jones, The New Republic, Daily Beast; center: BBC, NPR, Axios; right: Fox News, New York Post, Washington Examiner, Daily Caller. Hard-paywalled outlets are excluded so every article's text could be fetched and every quote verified. 27 fetch-failed articles were skipped and disclosed, 2 syndicated links kept without lean attribution. Scored by the same engine as every Coverage Audit (claude-sonnet-5), one article = one score, no cherry-picking: the feeds' current output, as found.

Caveats, honestly: one build window (the news week of 2026-07-08); feed-selected articles skew toward each outlet's front-page mix that week; the Manipulation Index is a model judgment grounded in verified quotes, not a fact-check verdict. Read the methodology before quoting numbers.

Citing this? StrataLens Reference Baseline v0 (n=121, built 2026-07-08) — stratalens.co/state-of-baseline.html
See where your coverage sits — every audit includes its percentile →