Handling Conflicting Data Sources

Core

Preserve information provenance and handle uncertainty in multi-source synthesis · Difficulty 3/5

0%
conflictsdata-qualityattributiontemporal

When multiple credible sources contain contradictory information, agents must handle the conflict appropriately rather than arbitrarily choosing one.

Correct Approach

  • Preserve both values: Include both conflicting figures in the analysis
  • Annotate explicitly: Mark the conflict with source attribution
  • Defer reconciliation: Let the appropriate authority (coordinator or human) decide
  • Complete the task: Don't halt analysis -- continue processing with the conflict noted
  • Temporal Awareness

    Require publication/collection dates in structured outputs. What looks like a contradiction may actually be temporal change:

  • Source A (2023): Revenue $10M
  • Source B (2024): Revenue $15M
  • Without dates, these appear contradictory. With dates, they show growth.

    Anti-Patterns

  • Choosing one value: The agent lacks broader context for credibility judgments
  • Halting analysis: Wastes completed work and blocks the pipeline
  • Hiding the conflict: Downstream consumers can't assess reliability
  • Coverage Annotations

    When operating with incomplete data (e.g., some sources timed out), annotate which findings are well-supported vs. which areas have gaps. This enables graceful degradation with transparency.

    Key Takeaways

    • Preserve conflicting data with source attribution; don't choose one
    • Continue processing (don't halt) but annotate the conflict
    • Include publication dates to distinguish temporal changes from contradictions

    Test Yourself1 of 3

    The web search agent and document analysis agent return contradictory market size figures from equally credible sources — $4.2B from an industry report and $6.8B from a government database. How should the synthesis agent handle this?