Summaries shed the caveats before they shed the claim

When an assessment gets condensed for a busier reader, the claim survives and the qualifiers die. "We judge" becomes "is." Named dissenting agencies become "most analysts." The confidence box gets cut for space. Nothing about the evidence changed; the sentence just got shorter. But every dropped word was doing epistemic work, so the summary reads as more certain than the document it summarizes. Stack a few layers of this and a hedge at the bottom is a fact at the top.

The October 2002 Iraq WMD estimate ran the experiment in public. The classified estimate said "We judge that Iraq has continued its weapons of mass destruction programs," carried named agency dissents, and included a box admitting low confidence in its scariest judgments, including whether Saddam would hand weapons to al Qaeda. The unclassified white paper, released days later, opened with "Iraq has continued its weapons of mass destruction programs." Full stop. The Senate post-mortem later said the plain thing: removing the qualifiers turned assessments into statements of fact.

The mechanism is compression, not necessarily malice, though malice rides it happily. Caveats look like padding to an editor and are load-bearing to an epistemologist. It is the sibling disease to [vague probability words](Vague probability words let everyone silently disagree while appearing to agree): one blurs uncertainty inside a sentence, the other amputates it between drafts.

Source claim: Each layer of summarization preserves the judgment and discards its qualifiers, so apparent confidence rises as an assessment travels toward the reader while the evidence underneath stays constant.