A record that helps a mediator ask better questions

Start with the process, not the transcript

A mediator may need to establish who can speak for a party, what can be shared, and which questions the participants are willing to discuss before any useful document analysis begins. The UN Guidance for Effective Mediation places preparedness, consent, impartiality, and inclusivity among the fundamentals of a mediation process.[1] These are responsibilities of people and institutions, not properties conferred by recording a meeting.

This note asks a narrower question: what kind of record could help a mediator revisit a conversation accurately? The proposal is analytical support for preparation and review. A transcript may omit tone, an interpreter’s qualification, an off-record understanding, or people who were never invited. Its apparent completeness can be misleading.

Keep the condition attached to the offer

Consider a fictional workplace dispute. A manager says, “I could support a different rota if the team covers Friday.” An employee replies, “I need predictable hours because I arrange care.” The first statement is a conditional indication of support. The second states a concern. Neither establishes an agreed rota, a binding promise, or a complete account of either person’s priorities.

A useful record would help the mediator ask who can authorize a change, what Friday coverage means, and whether other arrangements meet the employee’s stated concern. It would preserve the question rather than fill the gap with a plausible answer. Harvard’s account of principled negotiation explains why exploring interests can broaden discussion beyond stated positions.[2] It does not license a system to infer a private motive as fact.

A disagreement deserves an accurate record

Two parties may describe the same conversation differently. The assistant should preserve attribution, source wording, and each account’s scope. It should not manufacture a compromise version. Equally, an apparent contradiction may disappear when the mediator learns that the statements concerned different dates, people, or obligations. “Contradiction” is a candidate analytical label that needs review.

For defensive narrative work, this distinction is especially important. A changing account may reflect correction, strategic framing, misunderstanding, or additional information. The record should make the change inspectable without converting it into a claim of deception. The mediator retains responsibility for what to ask next and whether a question should be asked at all.

Design around permission and review

The UN DPPA and HD digital mediation toolkit discusses confidentiality, access, exclusion, and the risks introduced by digital communication.[3] These concerns shape our proposal. A private caucus note and a jointly approved summary cannot be treated as interchangeable inputs. Capture, retention, access, and reuse require an explicit agreement appropriate to the process; consent to participate is not automatically consent to every downstream use.

CONCORDIA is an exploratory research direction, not a deployed listening service. The wider TACITUS design places reviewed knowledge at the DIALECTICA boundary and uses capsules in the PRAXIS agentic workspace. A future mediation workflow would need permission-aware source selection and human approval of any shared output. Describing that requirement is not evidence that the necessary controls have been built or certified.

Evaluate usefulness without claiming to measure mediation

An initial evaluation could use fictional or suitably authorized transcripts annotated by practitioners. Tests should count speaker swaps, lost conditions, unsupported commitments, and invented interests. Reviewers should also assess whether summaries preserve disagreements and whether suggested questions are grounded in the available record. Agreement between annotators should be reported, including where the task itself remains ambiguous.

A separate access test should attempt to draw material from a restricted caucus into a shared summary. Such a failure would stop the workflow, regardless of how fluent the summary is. Time saved during preparation can be measured, but it should not be confused with a better settlement, a fairer process, or consent from the parties. Those outcomes require different evidence and a different study.

Ask for acceptance before treating the proposal as shared

Sam’s “we are agreed” message asserts an assignment; it cannot establish Alex’s acceptance at that moment. The later “sounds good” response is relevant but may leave scope and timing ambiguous, and the later denial remains attributed. Return a question about authority, explicit acceptance and the intended obligation. Interests remain tentative interpretations to test with the people concerned; private material requires consent-aware handling. [4]