Netsplit logic
Netsplit compares direct evidence for a treatment contrast with indirect evidence implied by the rest of the network. A simplified inconsistency statistic is z = (direct - indirect) / sqrt(SE_direct^2 + SE_indirect^2), with a p-value based on a normal approximation when assumptions are met.
Coventra runs netsplit-style diagnostics where the network has closed loops and direct evidence supports the comparison. Tree-like networks do not support local inconsistency testing.
Netheat
A netheat plot visualizes which comparisons contribute to inconsistency in the network. It is a diagnostic map, not a single pass/fail test.
Coventra returns not-applicable or unavailable module status when no closed loop exists, when netheat is not implemented for the selected exact binary method, or when the R package cannot estimate a stable diagnostic.
Contribution and confidence signals
Contribution matrices describe how much each direct comparison contributes to network estimates. They help reviewers trace whether an important estimate depends mostly on indirect, sparse, or high-risk evidence.
Coventra's confidence diagnostics are support signals. They do not replace CINeMA, GRADE, or expert judgment. Certainty domains such as within-study bias, reporting bias, indirectness, imprecision, heterogeneity, and incoherence still require human review.
Reporting checklist
- State whether closed loops exist.
- Report netsplit results only for estimable comparisons.
- Report unavailable diagnostics explicitly rather than silently omitting them.
- Use contribution matrices to explain which direct evidence drives key estimates.
- Do not call automated confidence signals a complete certainty assessment.