synthesize-meta-analytic-evidence

synthesize-meta-analytic-evidence

Purpose

Design or execute quantitative evidence synthesis while preserving study-level quality, heterogeneity, bias, and sensitivity logic.

When to use / not applicable

Use when multiple studies or comparable effect estimates must be combined. Not applicable when evidence is a single case or no comparable outcome can be defined.

Input contract

mode_contracts:
  pairwise: &meta_input
    required: [study_records, outcome_definition, effect_measure]
    optional: [comparison_network, subgroup_plan, prior_quality_assessments]
    constraints: [study_level_provenance_required]
  network: *meta_input
  cumulative: *meta_input
  heterogeneity: *meta_input
  bias: *meta_input

Execution protocol

Do not perform called SOP operations inline; each loaded SOP owns its contract and thresholds.

1. You MUST load skill formulate-meta-analysis-scope to define outcome, effect measure, and inclusion boundary. You MUST load skill design-study-inclusion to define study-level inclusion and quality fields.
2. You MUST load skill plan-effect-size to plan effect-size calculation. You MUST load skill design-data-extraction to extract condition, sample, uncertainty, and provenance. You MUST load skill audit-study-validity to retain study-validity judgments.
3. Select one mode: pairwise, network, cumulative, heterogeneity, or bias.
4. You MUST load skill assess-sensitivity to run sensitivity checks and synthesize estimates with uncertainty and exclusions.

If the synthesis exposes a mechanism-level explanatory gap, consider formulate-hypotheses as the next tactic.

Mode branches

• pairwise: pool directly comparable effect sizes, report sampling uncertainty and study-quality judgments with sensitivity analyses, and stop only after at least 80% of every declared pairwise floor is met.
• network: compare N>=3 methods using direct and indirect evidence. You MUST load skill construct-evidence-network to construct the direct and indirect evidence network.
• cumulative: update the estimate as studies arrive; retain order and stopping state. You MUST load skill update-cumulative-evidence to perform the ordered cumulative updates.
• heterogeneity: investigate between-study variation; preserve I2 interpretation bands. You MUST load skill analyze-heterogeneity to analyze between-study variation.
• bias: test publication, selection, and small-study bias; do not treat absence of evidence as absence of bias. You MUST load skill assess-publication-bias to assess publication, selection, and small-study bias.

Output contract

mode_contracts:
  pairwise:
    produces: [question, inclusion_criteria, studies_included, effect_size_type, model, heterogeneity_plan, sensitivity_plan, bias_assessment_plan, reporting]
    delta_fields: [findings, evidence_updates, uncertainties, decisions, open_questions]
  network:
    produces: [question, network_geometry, inclusion_criteria, studies_included, effect_size_type, model, transitivity_assessment, ranking_method, heterogeneity_plan, inconsistency_plan, sensitivity_plan, bias_assessment_plan, reporting]
    delta_fields: [findings, evidence_updates, uncertainties, decisions, open_questions]
  cumulative:
    produces: [question, temporal_scope, inclusion_criteria, studies_included, chronological_order, effect_size_type, model, temporal_analyses, time_lag_bias, quality_trend, reporting]
    delta_fields: [findings, evidence_updates, uncertainties, decisions, open_questions]
  heterogeneity:
    produces: [question, heterogeneity_metrics, moderator_candidates, investigation_plan, a_priori_hypotheses, multiple_testing, reporting]
    delta_fields: [findings, evidence_updates, uncertainties, decisions, open_questions]
  bias:
    produces: [question, bias_domains, grey_literature_search, grade_assessment, sensitivity_plan, reporting]
    delta_fields: [findings, evidence_updates, uncertainties, decisions, open_questions]

Thresholds and quality gates

• Evidence-acquisition gates use relative coverage, not fixed study/web counts. Declare the eligible evidence universe (or current evidence pool), then record numerator, denominator, batch increment, stopping reason, and source references.
• Pairwise/network/cumulative/heterogeneity/bias modes stop when marginal information gain falls below the declared, justified floor across consecutive batches; report coverage ratio, independent-source ratio, and saturation state.
• effect-size-extraction and quality-assessment-protocol must cover the declared eligible records at the justified relative floor; do not substitute an unreasoned absolute count.
• If k >= 10, pre-specified subgroup/meta-regression investigation is required where applicable.
• I2 interpretation bands are retained: 0-40% low, 30-60% moderate, 50-90% substantial, 75-100% considerable; overlaps are source wording and must not be silently normalized.

Failure and counterexamples

Reject synthesis when study identity, outcome definition, or effect measure is missing. Flag disconnected network, incomparable conditions, unplanned subgrouping, and bias tests with insufficient studies.

Provenance map

meta-analysis, pairwise-synthesis, network-comparison, cumulative-tracking, heterogeneity-investigation, bias-detection, effect-size-extraction, quality-assessment-protocol, evidence-synthesis-planning; all source gates kept, repeated provider details compressed.

Legacy context checkpoint / Delta notes

Append study set, extraction table, estimate, uncertainty, heterogeneity/bias decisions, and unresolved comparability questions.

Preserved source criteria ledger

Context checkpoint / Delta notes

Append study set, extraction table, estimate, uncertainty, heterogeneity/bias decisions, and unresolved comparability questions.