Fact
Qualitative codes are research judgments.
A code used in analysis makes a claim about what a passage means for a study.
Trust & provenance
Every AI coding suggestion requires an explicit human decision.
OpenVerbatim separates machine suggestions from reviewed research decisions, then records each adjudication in an append-only audit trail.
Review rule
Facts, consequences, and product behavior stay separate.
The distinction matters because a method claim and a product behavior claim should be easy to evaluate on their own.
Consequence
Software should keep judgment visible.
If a tool hides whether a code was proposed, accepted, or rejected, reviewers have less to inspect when a finding is challenged.
Spec
OpenVerbatim uses suggested-until-confirmed review.
AI coding suggestions stay in a suggested state until a human reviewer confirms or rejects them.
Decision record
How a decision is recorded.
The review rule is designed around explicit state changes instead of silent promotion from AI output to research evidence.
AI output starts as suggested
An AI coding suggestion is stored in a suggested state. It is not final evidence, a confirmed code, or a completed analytic decision.
A human reviewer makes the decision
The reviewer must explicitly confirm or reject the suggestion before it can leave the suggested state.
The adjudication writes an event
Each decision writes an append-only audit event. Database-layer triggers enforce that audit events cannot be changed or deleted.
Downstream work reads the reviewed state
Later coding, sharing, and evidence review can distinguish suggested material from reviewed decisions.
Trust pages
Review the concrete trust workflows.
Each page covers one public question researchers, supervisors, and reviewers ask before private interview material leaves the workspace.
deletion receipts
Deletion receipts for participant withdrawal
Interview-level erasure removes the withdrawn source, dependent transcript and coding records, local audio storage, and affected share items while keeping a deletion audit event.
methods disclosureAI disclosure statement generator
Researchers choose which disclosure fields to include, generate neutral methods text, and copy it into a thesis or paper methods section.
redacted sharingRedacted sharing for interview quotes
Public links expose only reviewed excerpts that pass row-level masking, exclusion, or explicit consent confirmation.
Withdrawal, disclosure, and sharing need separate controls
A participant withdrawal is not the same operation as a public share, and neither one is the same as a methods disclosure statement. OpenVerbatim keeps those workflows separate so each one can expose only the information it needs.
Withdrawal is source-level erasure with a deletion audit event. Disclosure is an author-controlled draft for methods text. Sharing is a row-by-row public release workflow for confirmed excerpts.
Sharing still requires row-by-row de-identification preview
Before material reaches a public view, OpenVerbatim asks the reviewer to inspect each selected excerpt. The reviewer chooses one of three outcomes for the item.
Mask
The reviewer can mask sensitive text before a row is allowed into a shareable view.
Remove
The reviewer can remove a row from the shared material when it should not leave the private workspace.
Confirm consent
The reviewer can mark that consent allows the item to be shared. Public endpoints still expose only whitelisted fields.
Inspectable source
Open source changes the review surface.
OpenVerbatim is Apache-2.0 open source. Researchers and institutions can self-host the full feature set and inspect the code paths that handle review state, audit events, deletion, and sharing.
OpenVerbatim entity
What OpenVerbatim is.
OpenVerbatim is an open-source (Apache-2.0) qualitative data analysis platform for coding and analyzing interview transcripts. AI-suggested codes stay marked as suggestions until a human reviewer confirms or rejects them, and every decision is kept in an audit trail. The full feature set is available when self-hosted; there is no paid feature wall.
FAQ
Trust and provenance questions
Can I trust AI qualitative coding?
Only if the workflow keeps AI output reviewable. In OpenVerbatim, every AI coding suggestion requires an explicit human decision before it can become reviewed material.
How does AI provenance work in qualitative research software?
OpenVerbatim records each adjudication as an append-only audit event and keeps suggested material separate from confirmed or rejected decisions.
What happens when a participant withdraws consent?
OpenVerbatim deletes the full interview record through cascade deletion, including the audio file. The deletion action itself remains traceable.
What can be shared publicly from a qualitative coding project?
Shared material must pass row-by-row de-identification preview. A reviewer chooses masking, removal, or consent confirmation for each item, and public endpoints use a field whitelist.
Is OpenVerbatim open source and self-hosted?
Yes. OpenVerbatim is Apache-2.0 open source, and the full feature set is available when self-hosted with no paid feature wall.
Try the evidence loop
Inspect the review loop before using private research material.
The public sandbox uses generated demonstration material, while the repository shows the code behind review state, audit events, deletion, and sharing.