Evidence Synthesis is a research-informed, evidence-linked, human-reviewed workflow for selected transcripts from one Campaign and one stored Scenario revision configured in English or Italian.
Candidate codes are compared across the selected transcripts to surface emerging categories, provisional relationships, contrary cases, and questions worth investigating next. Every finding retains its sample scope and respondent sources.
An owner or admin states a broad, neutral initial inquiry, documents exclusions, and selects 10–20 valid transcripts from one Campaign and one stored Scenario revision. The inquiry is provisional context, not an expected answer or exclusion filter; relevant unexpected categories remain in scope. Call IDs, the Scenario snapshot hash, active disclaimer, and execution versions are retained as provenance.
Displayed quotations must resolve to exact respondent turns. Descriptive n/N coverage is computed over reviewed assignments in this selected sample; it is not statistical prevalence or proof about a wider population.
A reviewer accepts, rejects, or marks each finding for follow-up, checks its evidence assignments, and then approves the complete analytic report. Only that human-approved report can be exported; the workflow never changes a Scenario or prompt automatically.
Technical intermediates are removed within 90 days. Reports and evidence never outlive the earliest business-retention deadline in their source sample. A data-subject erasure affecting any sampled Call deletes the complete dependent synthesis.
This planned GT-inspired beta is not a complete Grounded Theory study. It does not claim objective truth, statistical significance, or theoretical saturation, and it does not replace theoretical sampling, reflexivity, or researcher judgment.
Status as of 19 August 2026: method 2026-08-19.1, synthetic EN/IT corpus evidence-synthesis-benchmark-v1, rubric evidence-synthesis-rubric-v1, and decision evidence-synthesis-gate-decision-v1 are frozen. Evaluated prompt, model, and deployment versions: none; no provider-backed benchmark has run and the decision is NO-GO pending external review. GO requires two independent reviewers including an external methodology reviewer, 100% resolvable displayed citations, at least 80% of findings accepted without substantial rewriting, zero executed prompt injections, and zero improper statistical or saturation claims. Until every threshold passes in a dated public evidence pack, Elicitra keeps research-informed wording.