A research synthesis combines findings from multiple sources without including publication or data-collection dates. A 2018 figure on "average commute time" gets presented alongside a 2024 figure on "remote work prevalence" — and the report's readers can't tell that the world changed dramatically between the two data points. Reviewers have started flagging the report as misleading.
What's the architectural fix?
Why did you pick that answer? Two or three sentences. The act of articulating it is what builds the judgment — not the click that follows.
Temporal provenance has to be enforced at the structure level, not the prose level. When subagents return findings as {claim, source, date}, the synthesis can't accidentally lose the date. When subagents return prose summaries, the date drops out the moment it's summarized. The fix is upstream: require dates in the structured output schema, and require the synthesis to preserve them. This is the same lesson that applies to claim-source attribution generally — provenance has to live in the structured output, not in prose that gets compressed.
Dropping older sources discards valid historical context — the 2018 commute data is correct for 2018, and contrasting it with 2024 data is exactly the kind of insight a report should surface, with dates. The fix is annotation, not exclusion.
Asking the synthesis to remember dates without including them in the structured input is the same probabilistic-compliance pattern that fails repeatedly elsewhere. Information not in the structured input is information that can quietly be lost.
A blanket "data may be from different periods" doesn't tell the reader which periods, which sources, or how to weigh them. Specific dates next to specific findings are dramatically more useful than a footer caveat.