The model runs the interview. Our methodology decides what counts.

Evidence

• Behavior before interpretation. Conclusions must rest on described behavior, not impressions.
• Atomic evidence objects. Feedback is broken into single, comparable observations.
• Paraphrase only. Feedback is paraphrased so no one can be identified by wording.
• Source typing. Self-reported and externally reported evidence are kept distinct.
• Learner and observer vantage points are distinct. Student and colleague perspectives are treated as different, not better or worse.
• Authority accounts do not dominate. Every perspective is weighed the same way.

Fairness

• Protected attributes are out of scope. Protected characteristics are never inferred or used in evaluation.
• No coded proxies. Coded stand-ins for protected traits are rejected as evidence.
• No personality diagnosis. No personality types or clinical labels are assigned.
• Flag unsupported inference. Claims that go beyond the evidence are softened or removed.
• Popularity is not teaching quality. Being liked is not evidence of teaching quality.
• No diagnostic or clinical language. This is not a clinical or diagnostic assessment.
• No relational or clinical diagnosis. No diagnoses or labels about either partner.

Claim strength

• Claim strength tiers. How strongly something is stated depends on how many independent people described it.
• No single-source generalization. One person's comment never becomes a conclusion.
• Contradictions are preserved. Conflicting feedback is shown as a tension, not smoothed over.
• Gaps are calculated. Self-versus-others gaps are calculated from evidence on both sides.
• No false precision. No invented numbers or scores.
• Developmental use only. This feedback is developmental and is not an employment decision.
• No fixed-ability or intelligence claims. Feedback describes habits, never fixed ability.
• No high-stakes school decisions. This feedback is developmental and is not used for school decisions.
• No numeric confidence in shared output. The shared reflection contains no scores.

Interviewing

• Concrete examples. Interviews ask for real examples, not opinions.
• No leading questions. Interviewers never lead the respondent toward a conclusion.
• No demographic inquiry. Interviews never ask about personal characteristics or identity.

Quality audit

• Pre-report audit. Every report passes a bias and quality audit before it is written.
• Blocking versus softening. Serious problems remove evidence; lesser ones soften the wording.

Confidentiality

• Group size threshold. A group is only described separately when at least three people took part.
• Labels never reach analysis. Your private labels for raters are never sent to the AI.
• Transcripts stay server-side. Raw interview transcripts are never shown to the subject.
• Private accounts stay private. What each of you said privately stays private.

Synthesis

• Traceability. Every conclusion is traceable to the evidence behind it.
• Approved input only. Reports are written from audited evidence, not raw transcripts.
• Two perspectives, equal standing. Both perspectives are treated as equally real.
• Safety-sensitive material is not summarized. Anything that sounds unsafe is not summarized here.

Coaching

• Grounded coaching. Coaching stays grounded in the report and never guesses sources.