Free, transparent, and publication-ready cross-cultural adaptation workflow.

Independent translations surface different readings of the same item, and that divergence is what the reconciliation step examines. If translators could see one another’s drafts, the versions would converge and the comparison would lose its meaning. Equivalence withholds each translator’s work in the API and in database row-level security until every translator has submitted, so independence is a property of the system rather than a promise (Beaton et al., 2000).

Back-translation tests whether the reconciled version carries the original meaning on its own. A back-translator who has studied the source instrument can reproduce it from memory rather than from the translation, which empties the check. The role matrix therefore refuses to assign a forward translator as the back-translator. Back-translation itself is optional under some frameworks and does not appear in TRAPD, whose team review and adjudication steps do the equivalent work (Beaton et al., 2000; Epstein et al., 2015; Harkness, 2003).

The rationale records why one wording won over another. It is the part of the process reviewers and committees ask to see, and it is what makes the reconciled version an auditable decision rather than an unexplained outcome. The requirement is enforced in the interface and again on the server.

Each stage consumes the output of the one before it. Rating an item before it is reconciled, or reconciling before all translations are in, produces evidence about a version that no longer exists. The gates keep the audit trail coherent, and every gate explains its condition and shows who the workflow is waiting on.

Experts rate each item’s relevance on a 4-point scale. I-CVI is the proportion of experts rating an item 3 or 4, with 0.78 the conventional floor for panels of six to ten (Lynn, 1986). S-CVI/Ave is the mean I-CVI across items, with 0.90 the accepted scale-level threshold (Polit & Beck, 2006). Modified kappa corrects I-CVI for chance agreement. Equivalence computes all three live as ratings arrive.

Gates, blinding counts, and the content validity indices are computed per language pair. Folding two target languages into one project would mix their denominators. When you adapt the same instrument for a second language, duplicating the project reuses the source items with a fresh workflow.

Some combinations remove an independence the method assumes. A forward translator cannot serve as the back-translator, because the back-translator must remain naive to the source (Beaton et al., 2000). Under Beaton/ISPOR and the ITC guidelines a translator cannot rate their own work on the expert panel, while TRAPD and the dual-panel method seat translators in review by design, so the disabled set changes with the framework you selected (Harkness, 2003; McKenna & Doward, 2005; International Test Commission, 2018). Combinations that are workable but weaker are permitted with a recorded rationale.

The certificate records the stage sequence with timestamps, the team roles, the content validity results, and a SHA-256 hash of the final items. The hash proves that the version you cite is byte-for-byte the version that completed the process; any alteration after sign-off produces a different hash. It does not certify translation quality, which remains the panel’s judgment.

The evidence the workflow produces depends on human translators working independently. Machine suggestions would anchor the versions to a shared starting point and make the independence claim in your methods section false. This is a design decision in favor of the property the method needs.