How This Register Was Developed

1. Analytical Stance. Methodological anchor: qualitative, interpretive and reflexive thematic reading. Seven starting categories guide attention, while additional codes and counterexamples challenge them. 
2. Reading & Extraction. All supplied sources were read through their supplied English renderings, with original-language reading where no English exists and targeted Chinese checks of consequential passages, corrections and limitations. This is not a line-by-line translation audit. Extended extracts retain the surrounding conversation rather than silently selecting only supportive sentences.
3. Verbatim Source Text. Original Chinese and any supplied English are rendered directly from source paragraphs. There is no polishing, filler removal, invented pause, repaired wording or internal ellipsis. The English is a supplied translation, not a second verbatim record of original speech. 
4. Traceability. P means the one-based Word body-paragraph number, including blank paragraphs in the count. Blank separators are not displayed. Each visible block identifies its original and supplied-English paragraph locations. A range is a continuous span of original paragraphs; bilingual blocks beginning in that span are shown in full, so a translation can sit immediately beyond the stated original endpoint. A7 uses its original chat paragraphs.
5. Speaker Attribution. A7 preserves explicit source labels. Student speaker labels inherited from the earlier register remain tentative inferences, including unresolved overlaps. Most tutor and administrator paragraphs lack authenticated labels and are marked Source speaker unassigned. Some paragraphs contain both voices. Such passages are exchanges, not automatically participant-only quotations; consult the recording before publishing a speaker-specific quote.
6. Permission Boundary. T5 P2218 records an explicit deletion request. The immediate surrounding exchange, original P2155–P2272 and its supplied translations, is withheld from both extract data and the source view. The exact intended extent of deletion is not established; the marker remains for researcher review. No conclusion relies on the withheld wording. Preliminary and post-pause material is flagged separately, rather than silently treated as ordinary formal testimony.
7. Translation Gaps. A7 has no supplied English. A2 P49 and A6 P346 lack separately paired English. A6’s supplied English at P341 appears misaligned with the original at P340 and may correspond to P346. A4 P1663–P1666 also needs checking for a logical inconsistency. These are preserved and flagged, not silently repaired. 
8. Coverage. The register aims at comprehensive extraction across all files and categories, with 498 extended passages. It does not claim that an interpretive first pass can guarantee every possible mention or latent meaning. Searchable source views support checking passages outside the register. Counts measure extract organisation, not individual mentions, endorsement, prevalence or thematic importance.
9. Evidence Status. Recollections, opinions, organisational self-descriptions, hearsay, hypotheticals and interviewer examples are distinguished in record-level notes. Admissions, finances, qualifications, employment, allegations and policy statements have  been externally verified. 
10. Comparison Limits. These are comparisons between interview and chat accounts only. A recalled workshop remains an interview account, not independently analysed workshop material. No separately supplied field notes, observations, artefacts or university admissions criteria have been triangulated.