Plan Transcript and Translation Quality Checks

Use only supplied evidence. Cite source labels and locations. Separate observations, source claims, calculations and proposals. Keep missing evidence unknown and preserve conflicts. Never invent values, quotations, completed checks, permissions, authority or sign-off. Apply only documented applicable obligations. Recommendations do not authorize access, reuse or publication. Descriptive findings do not establish causal impact.

Use only material supplied in this conversation. Ask me for safe transcript/schema extracts, recording and translation availability, languages, source/version references, transcription conventions, documented use boundaries and reviewer capacity. Do not request sensitive recordings or identifying text before permitted use is established. If essential sources are absent, give a conditional QA plan and focused requests, not source-specific findings.

Plan pre-coding transcript and translation quality checks for the actual materials and permitted uses. Do not conduct thematic analysis or invent participant findings.

1. Inventory source recording/text identifiers and versions, languages/dialects, transcription conventions, translation stages, safe speaker labels, time/line references and supplied reviewer roles/skills. Identify whether source audio, original-language text, translated text and relevant contextual notes are actually available. Ask for missing substantive material; a file name is not its contents. Check documented consent and access scope separately for recording, transcription, translation, AI processing and quotation. Do not assume participation permits any of these uses. Use safe metadata or excerpts while sensitive access remains unresolved. Request audio only if this environment supports it and the user confirms this use is permitted; otherwise plan an authorized local comparison and state that audio fidelity cannot be checked here.
2. Define a risk-based review scope and rationale using material length, languages, recording quality, sensitivity, speaker overlap, key passages and reviewer capacity where known. Separate proposed complete review from targeted or sampled checks; identify how material will be selected and what triggers broader review. Do not impose an arbitrary universal percentage, error tolerance or independent-review requirement. Include language-competent review when the task needs it, with roles proposed for confirmation if absent. A sample cannot certify unexamined passages or the entire corpus.
3. Specify checks for omissions/additions, meaning, negation/polarity, quantities, attribution, turn order, speaker labels, timestamps and overlap. Preserve source ambiguity, hedging, disagreement and interviewer versus participant statements. Distinguish an actual source error from a stylistic preference. Do not add words from context, turn an uncertain statement into certainty, assign unidentified speech to a person, or treat automatic transcription confidence as verified fidelity. Without permitted audio comparison, distinguish text-level consistency checks from unverified recording-to-text accuracy.
4. For translation, retain traceable original/translated segment pairs and version relationships where permitted. Check meaning, polarity, local terms, register and context through an appropriate language-competent comparison. Flag alternative readings and culturally specific terms for review rather than silently normalizing them. Back-translation may be an optional diagnostic, not proof of fidelity. Preserve original-language text and mark any gloss as an interpretation. Do not invent a translation for inaccessible or unintelligible source material, or treat a polished translation as an exact participant quotation.
5. Agree explicit notation for unintelligible speech, uncertain terms, overlaps, source gaps, speaker uncertainty and redactions. Describe the notation in words or use conventions supplied by the user. Keep unresolved text visibly uncertain; do not fill gaps with plausible dialogue. Plan de-identification checks for direct and indirect identifiers while retaining analytically necessary meaning. Use consistent safe labels and restricted source links rather than exposing a re-identification key. Do not promise anonymity from removing names. Redaction and translation choices must retain a documented rationale without copying sensitive content into an unrestricted log.
6. Deliver a QA checklist and scoped review plan, discrepancy/adjudication log template, notation conventions and version/status rules. Log source/segment reference, issue, evidence available, proposed correction, reviewer/decision status, unresolved alternatives and affected versions. Preserve originals and apply confirmed corrections in a traceable derived version. If safe excerpts are supplied, findings must be limited to the actual checks performed on those excerpts; name absent source layers and unreviewed material. Label planned, checked-with-open-issues and completed scope distinctly, with completion criteria proposed unless supplied. Transcript QA does not validate participant claims, establish study findings or grant public-quote clearance.