Qualitative Codebook Reliability and Reflexivity Protocol

Develop or audit a qualitative codebook for a coding-reliability or structured codebook approach while preserving source context, interpretive plurality and researcher accountability. This protocol is not a substitute for reflexive thematic analysis methods that reject agreement as a quality objective.

## Inputs

Research question and qualitative method:
{{research_question_and_method}}

Sanitized sample material and draft codebook:
{{sample_material_and_draft_codebook}}

Coding team and reflexivity context:
{{coding_team_and_reflexivity_context}}

Ethics, privacy and project constraints:
{{ethics_privacy_and_project_constraints}}

## Working rules

- Treat excerpts as supplied evidence; code meanings and themes are interpretations. Label assumptions, rival interpretations, evidence conflicts, missing context and unresolved uncertainty.
- Preserve enough excerpt and document context to explain a code without exposing unnecessary participant identity or sensitive detail.
- Do not invent quotations, participant characteristics, field context, code frequencies, coder decisions or agreement statistics.
- Use this protocol only where a codebook and a coding-reliability, negotiated-agreement or structured comparison process fits the supplied method. If the project uses reflexive thematic analysis without coder-agreement or adjudication goals, stop and return a method-fit warning rather than imposing this protocol.
- The AI reviews supplied human coding records and excerpts. It does not act as an independent coder, and its suggestions are provisional analytic prompts. Never report them as coder agreement, researcher consensus or participant meaning.
- Keep codebook changes versioned. The research team controls code adoption, theme construction and interpretive claims.

## Protocol

1. Establish the methodological contract: epistemic stance, analytic unit, coding approach, role of prior theory, desired consistency and appropriate treatment of disagreement.
2. Build an evidence inventory identifying excerpt IDs, source type, context available, consent or use restrictions, redactions and sampling limitations.
3. For every proposed code, define name, concise meaning, inclusion criteria, exclusion criteria, close alternatives, positive and negative examples, contextual cautions and parent/child relationships.
4. Review a deliberately varied calibration sample and the coding records produced independently by the human coders. If actual coding records are unavailable, specify the sample and recording sheet but mark the exercise `Not run`; do not generate a substitute AI coding round.
5. Compare coding at the level appropriate to the method. Record exact agreement, partial overlap, divergent interpretation, missed context and codebook ambiguity without reducing disagreement to a single score.
6. Generate rival interpretations for consequential excerpts. For each, identify supporting evidence, disconfirming evidence, positionality considerations and a discriminating question.
7. Conduct reflexivity review: researcher roles, relationships to the setting, theoretical commitments, power, language and decisions that may shape attention or interpretation. Do not infer identity or bias that was not supplied.
8. Adjudicate only through a documented researcher process. Record retain, revise, split, merge, retire or leave plural, with rationale and dissent preserved.
9. Design the next calibration round and stopping condition. Avoid claiming saturation or reliability solely because the codebook is detailed.

## Output contract: Codebook and Reflexivity Pack

Return:

1. **Method fit statement**: intended analytic approach, appropriate consistency goal and prohibited overclaims.
2. **Source and privacy ledger**: excerpt ID, context, consent/use boundary, redaction and analytical limitation.
3. **Versioned codebook**: code ID, label, definition, include/exclude rules, examples, near-neighbour distinctions and version note.
4. **Disagreement matrix**: excerpt, coder labels when supplied, disagreement type, rival readings, evidence and adjudication status.
5. **Reflexivity register**: standpoint or process factor, supplied basis, possible analytic effect, counter-practice and unresolved question.
6. **Adjudication log**: decision, participants, evidence considered, dissent, codebook change and effective version.
7. **Calibration plan**: sample logic, independent coding steps, comparison method, review roles and repeat/stop rules.
8. **Status note**: distinguish drafted, calibrated, revised, agreed, plural and untested elements.

## Completion conditions

The pack is complete only when definitions distinguish neighbouring codes, examples trace to authorized excerpts, disagreements and dissent remain visible, the chosen comparison method fits the qualitative tradition, and adopted changes identify the responsible researchers. If source context, consent boundaries or independent coding evidence are missing, mark the affected parts incomplete and request them.

For each completion condition, record the expected evidence, the actual supplied evidence, `Met`, `Not met`, or `Blocked` status, and any unresolved researcher action.

Do not certify thematic saturation, intercoder reliability, validity, consensus or ethical compliance unless the named procedure was actually completed and evidenced.