TL;DR
- 01Mediators operate with a small set of cognitive moves: separate position from interest, distinguish claim from commitment, surface BATNAs and red lines, track narrative drift, and keep the dispute legible to both parties without dissolving it.
- 02Most of these moves map cleanly to ACO primitives. The kernel was designed downstream of this practice, not upstream.
- 03The hardest move for a software system is the one most mediators do unconsciously: hold contradicting accounts side by side without choosing.
- 04TACITUS does not automate mediation. It legibility-supports it. CONCORDIA listens during a session and structures the transcript; the mediator decides.
Mediators have a small set of cognitive moves: separate position from interest, distinguish claim from commitment, surface BATNAs and red lines, track narrative drift. Most of these moves map cleanly to ACO primitives. The kernel was designed downstream of this practice, not upstream.
The cognitive moves, mapped
- ▸Position vs Interest →
Claim(type=Position)linked viaHOLDS_INTERESTtoInterest. The Interest carries aderivationchain (cannot be left bare). - ▸Claim vs Commitment → distinct primitives at the type level. A Commitment has a
statuswith bi-temporal transitions; a Claim has amodality(Asserted / Denied / Reported / Withdrawn). - ▸BATNA / Red line →
Claim(type=BATNA|RedLine)with a unique holder; the validator rejects shared red lines across parties at overlapping valid-times. - ▸Narrative drift →
Narrativeprimitive with an ordereddrift_chain; reframing is data, not noise. - ▸Contestation →
CONTRADICTSedges as first-class objects; the system does not pick a winner.
What experienced mediators do that LLMs do badly
- ▸Hold two accounts of the same event without averaging them. (LLM failure mode: smoothing.)
- ▸Notice that a party's position is drifting while their interest is stable — or vice versa. (LLM failure mode: collapsing to most recent.)
- ▸Surface the implicit commitment buried in an indirect statement. (LLM failure mode: literal reading.)
- ▸Track when a "soft" ask becomes a "hard" red line. (LLM failure mode: time-flattening.)
What CONCORDIA actually does in the room
CONCORDIA listens during a live or async session. It writes a Case-scoped subgraph: actors, claims, commitments, interests, contradictions, narrative drift. It surfaces structure to the mediator without intervening. On session close, the subgraph promotes to Analytical with the mediator's edits committed. The system never says "X was right." That is not its job.
SOURCES
