Sessions & coding
Mirrors AI_Coding_Protocol steps 2–3. Log the session, paste the MacWhisper .txt (timestamps on) after QC and the data manager's first anonymisation pass, run the second pass, then let the AI code it into Evidence_Log rows marked “To review”.
Session ID
Session_Log
Evidence log
Every AI row starts as “To review”. Changing a field before approving makes it “Approved (edited)”. Quote check compares the original-language quote with the clean transcript, as column Z does.
Triangulation
v5 rule. High: ≥ 3 entries, ≥ 2 methods, ≥ 2 sessions and ≥ 1 first-hand rights-holder account. Medium: ≥ 2 entries, ≥ 2 sessions or ≥ 2 methods, and ≥ 1 rights-holder source. Otherwise Low.
Saturation by segment
Saturation reached when a segment has ≥ 3 sessions and no new issue code in its last 2 sessions. Sessions in date order.
Impact register
AI drafts impact statements and proposed ratings with rationale from approved evidence. Two analysts set the final ratings; severity, priority, salience, action axis and timeline use the workbook formulas. Only confirmed impacts reach unit saliency.
Unit saliency
Highest priority per issue and unit from confirmed impacts. Green = Low · yellow = Medium · orange = High · red = Critical. The last column combines External – TR and External – RO.
Export & data
Columns follow the v5 workbook: Session_Log A–U, Evidence_Log A–Z (AA is a formula), Impact_Register A–AJ. Paste input columns only so grey formula columns stay intact.
Where data lives
Sessions, anonymised transcripts, evidence rows, impacts and the audit log are stored in Minerva's Supabase database (EU – Frankfurt). Each user sees only the projects they are a member of; every change and every export is recorded in the audit log. Raw transcripts never enter the app. Per the workbook's precondition, no real field data is processed until Minerva's legal counsel approves AI processing under KVKK and GDPR.
Admin
Invite users, set roles, manage projects and membership, and read the audit log. Access is removed by deactivating a user; users are never deleted, so the audit trail stays complete.