A conclusion synthesised from several sources cannot be attributed to any single one and must be marked separately

What it is

“Taken together, three reports show employment in this region is recovering.” Each report covers a different sector and a different definition; none issued “employment in this region is recovering.” Hanging all three on that sentence looks sourced; it is an unclaimed synthesis the model performed. Synthesis cannot be pinned to a single source requires that a conclusion inferred from several sources be marked as synthesis / model inference, not attributed to any item on the list.

Sentence-grain correspondence answers “which sentence to which paper.” A synthesis sentence has no original in any paper; correspondence itself should not occur.

Why it happens

People read a superscript as “this sentence is what that paper said.” Once a synthesis sentence hangs on a specific paper, readers go looking for the original wording, and when they cannot find it they assume they misread, rather than noticing an extra layer of inference. Generators often perform undeclared induction, comparison, and causation in summary tasks; the bearer of those operations is the system, not the material being summarised.

Multi-source hanging also mints a false consensus: three papers named looks like independent replication, when it was one synthesis. The feel of consensus comes from list length, not from whether the three agree on the conclusion.

Studying it

Construct three materials, none containing the target conclusion; the model output contains it and hangs all three. Compare hanging as ordinary citations versus marking “synthesised inference; the following were inputs.” Dependent variables: whether people think a given paper said the sentence, whether they think the three agree, attribution when the original cannot be found after opening (I misread / the system synthesised). Independent variables: list length, distance of the conclusion from each paper.

Attribution is the key. Blaming oneself for not finding the original is single-source hanging doing the misleading.

Where it stops holding

A genuine review article that itself wrote “taken together, prior work…” can be hung on that article, because the proposition lives there. Extractive listing of user-specified materials is not synthesis. A conclusion derived from axioms has another responsibility chain; do not reuse the literature-synthesis mark. This entry does not treat halo, nor one-click opening.

Applying it

• Distinguish in the output: verbatim from a paper / paraphrase of a paper / synthesis from several inputs. Give synthesis its own mark; do not make it click to a DOI.
• Under the synthesis mark, the input materials may be listed, with “none of the following directly states this conclusion.”
• If the distinction cannot be made, do not emit synthesis sentences; emit only extractive claims that can be singly hung.
• Check: point at the synthesis sentence and ask “which paper said this.” If the answer is some item on the list, and that item does not contain the sentence, single-source hanging is already posing as a source.

Related

• Same group: L3.08.1 A real citation that does not support the claim is a more hidden failure than no citation · L3.08.2 A source must open for checking in one click; an uncheckable mark only adds an appearance of credibility · L3.08.3 Attribution grain should be the sentence or paragraph; a list at the end cannot locate correspondence · L3.08.5 The presence of attribution raises trust in the whole piece, including parts that were not attributed
• Nearby: L3.02 Source Attribution · L3.03 Hallucination and the Fact-Checking Burden
• Search terms: synthesis attribution · model inference mark · cannot cite a synthesis