Eve library

Phase 179: Concordia Cooperative Mediation & Negotiation Intelligence — In-House Mediator.ai-Class Bargaining Engine, Multi-Party Preference Modeling, Nash / Pareto / Coalition Search, ODR Guardrails, Procurement Agents, Restorative Justice, Governance Dispute Resolution, Smart-Contract Settlement, and Ecosystem-Wide Integration

Tracker

Revision 1

On this page

Historical task source. The Eve task board holds current task state.

Sovereign implementation of an AI mediation and cooperative-negotiation stack that matches and surpasses the April-2026 frontier represented by Mediator.ai, AI-powered online dispute resolution products, autonomous procurement negotiators, and current multi-agent negotiation research.

The core thesis: any Oshun workflow where multiple parties, agents, communities, companies, creators, suppliers, users, or autonomous systems must reach acceptable terms should use a shared bargaining substrate instead of ad hoc chat, simple voting, fixed policy, or unilateral recommendation. The system models private preferences, proposes candidate agreements, searches the Pareto frontier, explains tradeoffs, detects coercion and strategic manipulation, and routes consequential outcomes through appropriate human, legal, governance, or smart-contract review.

Primary owning domains: Nous, Iris, Themis, Maat, Kuanyin, Aje, Contracts, OpenAPI, Proto, Sophia, Oshun, Shared. Deep integration targets: every domain listed in DOMAINS/README.md, with productized first-class paths for Maat, Themis, Kuanyin, Iris, Yemaya, Aje, Maya, Lilith, Sophia, Isis, Freya, Cybele, Brigid, Asase, Metis, Euterpe, Calliope, Kalika, Veritas, Psyche, and Neith.

Tech split: TypeScript (~45%) for services, SDKs, contracts, workbench, and domain integrations; Python (~30%) for preference learning, evaluation, agent simulation, and ML experimentation; Rust (~20%) for deterministic optimization, sandboxing, cryptographic audit ledgers, and low-latency search kernels; Solidity / Cairo / Move (~5%) for escrow, settlement, DAO execution, and agreement-hash anchoring.

Library layout: libs/contracts/concordia/, libs/proto/src/concordia/, libs/openapi/src/specs/concordia/concordia-api.yaml, libs/nous/cooperative-bargaining/, libs/nous/preference-inference/, libs/nous/agreement-search/, libs/iris/concordia-assistant/, libs/themis/dispute-resolution/, libs/maat/negotiation-intelligence/, libs/kuanyin/concordia-restorative/, libs/aje/settlement-escrow/, libs/sophia/concordia-knowledge/, libs/oshun/concordia-integration/, apps/oshun/web/src/app/studio/concordia-workbench/, apps/maat/negotiation-console/, services/concordia/orchestrator/, testing/concordia/.

Research snapshot checked 2026-04-21:

  • Mediator.ai presents a private-intake, agreement-generation, repeated scoring, and survivor-search loop over candidate agreements, grounded in Nash bargaining and LLM-inferred utilities.
  • Mediator.ai's technical post describes pairwise LLM comparisons to infer each party's utility, then genetic search using crossover and Lua mutators, while naming open risks: unstable inferred preferences, incomplete option-space search, specification gaming, BATNA manipulation, enforceability, and leverage-seeking counterparties.
  • LLMediator shows GPT-4-style ODR support for message reformulation, mediator response drafting, and possible autonomous engagement in high-volume, low-intensity disputes.
  • "Robots in the Middle" reports that LLMs can choose intervention types and generate mediation messages competitively in a blind evaluation, but this is still a small controlled dataset rather than production assurance.
  • ProMediate adds a 2026 framework for proactive multi-party mediator agents, measuring consensus change, intervention timing, effectiveness, and socio-cognitive intelligence.
  • ANAC remains the long-running automated negotiation benchmark family, now with automated, supply-chain, and human-agent negotiation tracks.
  • Pactum demonstrates production autonomous procurement negotiation: specialized agents for sourcing, payment terms, discounts, rebates, and post-sourcing negotiations, embedded in enterprise workflows with audit/security claims.
  • ZODR, Bot Mediation, MediationAI, Disputell, and related ODR tools cover synchronous legal mediation, AI co-mediation, pre-mediation intake, escrow, settlement drafting, transcript analysis, and mediator-assist workflows.
  • EU AI Act treatment of justice-adjacent AI, UNCITRAL ODR principles, and NIST AI RMF push the design toward human oversight, logging, traceability, security, transparency, risk management, and non-replacement of qualified professionals in legal or high-stakes contexts.

Strategic differentiation target:

Oshun should not clone a single legal mediation app. It should build a general cooperative decision and agreement engine: legal ODR when configured with qualified review, procurement negotiation for Maat and Brigid, DAO proposal settlement for Themis and Aje, restorative circles for Kuanyin, creative royalties for Yemaya / Euterpe / Calliope, multiplayer guild and creator-market agreements for Maya, agent-to-agent contracting for Iris / Nous, and everyday human conflict support for Iris / Lilith / Tara. The frontier advantage comes from combining LLM socio-cognitive mediation with formal optimization, uncertainty-aware preference learning, privacy-preserving party isolation, domain contracts, audit-grade governance, smart-contract execution, and continuous evaluation.


179.1 Research, Product Boundary, and Risk Classification#

179.1.1 Field Research and Competitive Audit#

  • 179.1.1.1 Build docs/research/mediation-negotiation-2026.md with a dated competitive audit of Mediator.ai, MediationAI, Bot Mediation, ZODR, Disputell, Dyspute.ai, TheMediator.AI, Pactum, Nibble, Kleros, ODR.com, Modria-class ODR history, eBay / PayPal ODR precedent, AAA mediator search, JAMS AI Rules, and human mediator-assist tooling.
  • 179.1.1.2 Create a source matrix separating product claims from peer-reviewed or preprint evidence; tag each source as commercial, academic, standards, regulatory, benchmark, or case-study.
  • 179.1.1.3 Write a gap analysis against Mediator.ai: multi-party support, formal agreement DSL, uncertainty-calibrated preference inference, multimodal evidence, provable privacy isolation, domain-specific templates, smart-contract execution, governance appeal paths, and cross-domain Oshun integrations.
  • 179.1.1.4 Write a gap analysis against Pactum / Nibble: procurement and commerce negotiation workflows, ERP / marketplace integration, supplier audit trails, objective savings metrics, supplier satisfaction metrics, campaign management, and guardrailed autonomous execution.
  • 179.1.1.5 Build a research bibliography covering Nash bargaining, Kalai-Smorodinsky bargaining, Nash social welfare, Rubinstein bargaining, mechanism design, ANAC, GeniusWeb, NegMAS, LLMediator, Robots in the Middle, ProMediate, large-scale autonomous negotiation competitions, multilingual reasoning-cost negotiation studies, bargaining-capability failure analyses, online dispute resolution, restorative justice, and mediation ethics.
  • 179.1.1.6 Add a research refresh gate before Phase B and every pilot: rerun the source matrix, archive dated copies of product claims, recheck ODR / procurement / AI-governance regulatory assumptions, and block pilot launch when a major source or legal posture changed without review.
  • 179.1.2.1 Define ConcordiaUseCaseClass with values: low_stakes_personal, community_restorative, creative_collaboration, commerce_negotiation, supplier_procurement, dao_governance, platform_dispute, employment_sensitive, family_sensitive, legal_claim, regulated_financial, medical_sensitive, agent_to_agent_contract, and simulation_only.
  • 179.1.2.2 Define hard boundaries for AI-only output: no binding legal claim resolution, no custody / domestic-violence / coercive-control decisions, no employment adverse-action settlement, no medical or disability-rights settlement, no financial product suitability settlement, and no high-stakes rights waiver without qualified human review.
  • 179.1.2.3 Implement jurisdiction and risk banners that state whether a case is brainstorming, facilitated mediation support, procurement negotiation, governance process, arbitration support, or legal-review required drafting.
  • 179.1.2.4 Require explicit informed consent before collecting private party statements, evidence, BATNA estimates, reservation points, confidential concessions, emotional context, or settlement authority.
  • 179.1.2.5 Create escalation policies mapping use-case risk to human mediator, counsel, compliance, safety, or governance reviewer queues.

179.1.3 Standards and Compliance Baseline#

  • 179.1.3.1 Map UNCITRAL ODR principles into product requirements: fairness, transparency, due process, accountability, accessibility, neutrality, and party control.
  • 179.1.3.2 Map EU AI Act high-risk requirements for justice and democratic-process adjacent uses into Maat / Themis / Kuanyin review gates: risk management, data quality, logging, technical documentation, deployer information, human oversight, robustness, cybersecurity, and accuracy.
  • 179.1.3.3 Map NIST AI RMF functions into engineering controls: govern, map, measure, and manage for mediation-specific harms.
  • 179.1.3.4 Add model-card and system-card templates for every mediation model, evaluator, domain policy bundle, and optimization strategy.
  • 179.1.3.5 Define audit retention, privilege/confidentiality flags, data residency, deletion, legal hold, and export controls for each use-case class.

179.2 Cross-Domain Contracts and Data Model#

179.2.1 Canonical Concordia Contracts — libs/contracts/concordia#

  • 179.2.1.1 Create libs/contracts/concordia/ package exporting Zod schemas for ConcordiaCase, Party, PartyRole, Issue, Interest, Need, Constraint, Redline, BATNA, ReservationPoint, EvidenceItem, PrivateStatement, SharedStatement, AgreementCandidate, AgreementTerm, UtilityModel, PreferenceQuery, PairwisePreference, FairnessMetric, SearchRun, SettlementDraft, ReviewDecision, and ExecutionArtifact.
  • 179.2.1.2 Add privacy-scoped fields: visibility, legalSensitivity, safetySensitivity, retentionPolicy, and privilegeClaim. visibility must support private_to_party, mediator_only, shared_with_all, reviewer_only, and sealed_until_consent.
  • 179.2.1.3 Add authority fields: canNegotiate, canAccept, canBindOrganization, requiresHumanApproval, requiresCounselReview, requiresGuardianOrRepresentative, and settlementLimit.
  • 179.2.1.4 Add domain-specific extension points for Maat procurement, Themis governance, Aje escrow, Kuanyin restorative circles, Yemaya royalties, Euterpe licensing, Maya guilds, Lilith creator disputes, and Iris personal coordination.
  • 179.2.1.5 Add contract tests ensuring no private party fields can be serialized into shared views or opposing-party prompts.

179.2.2 OpenAPI / Proto / Event Surface#

  • 179.2.2.1 Create libs/openapi/src/specs/concordia/concordia-api.yaml with endpoints: POST /concordia/cases, POST /concordia/cases/{id}/parties, POST /concordia/cases/{id}/intake, POST /concordia/cases/{id}/issues, POST /concordia/cases/{id}/preference-queries, POST /concordia/cases/{id}/search-runs, GET /concordia/cases/{id}/pareto-frontier, POST /concordia/cases/{id}/drafts, POST /concordia/cases/{id}/reviews, POST /concordia/cases/{id}/execute, and GET /concordia/audit/{id}.
  • 179.2.2.2 Create libs/proto/src/concordia/ with streaming RPCs for live session transcription, real-time co-mediator suggestions, offer/counter exchange, and search-progress updates.
  • 179.2.2.3 Add domain events: concordia.case.created, concordia.party.joined, concordia.intake.completed, concordia.issue.identified, concordia.preference.updated, concordia.offer.generated, concordia.offer.compared, concordia.search.completed, concordia.draft.reviewed, concordia.settlement.accepted, concordia.execution.completed, and concordia.escalation.required.
  • 179.2.2.4 Add generated client checks to contracts:check so Concordia schemas, OpenAPI docs, protobuf artifacts, and TypeScript clients cannot drift.

179.2.3 Agreement DSL and Clause Model#

  • 179.2.3.1 Build an AgreementDSL that represents terms as typed clauses with variables, constraints, dependencies, temporal conditions, signatories, enforcement paths, and review requirements.
  • 179.2.3.2 Support common term families: money, equity, royalties, deadlines, deliverables, service levels, licensing, exclusivity, territory, data rights, governance rights, voting thresholds, apology, confidentiality, non-disparagement, restitution, restorative actions, access revocation, buy-sell clauses, escrow release, and monitoring.
  • 179.2.3.3 Implement clause-level provenance: which party interest, evidence item, legal/policy template, or optimizer move produced the term.
  • 179.2.3.4 Add static validation for impossible, contradictory, unlawful, unenforceable, unsafe, or policy-blocked clauses before candidates enter optimization.
  • 179.2.3.5 Build jurisdiction and domain template registries with mandatory human-review flags for legal, employment, family, regulated finance, healthcare, and minors-related clauses.

179.2.4 Repo Integration and Generated Artifact Registration#

  • 179.2.4.1 Scaffold Concordia packages with repo-standard package.json, project.json, tsconfig, Vitest config, exports, tags, and ownership metadata so Nx affected builds and project graph discovery include every Concordia library.
  • 179.2.4.2 Register Concordia OpenAPI specs under libs/openapi/src/specs/concordia/, generated TypeScript clients under libs/openapi/src/generated/, docs under libs/openapi/docs/concordia/, and @oshun/codegen checks used by root pnpm contracts:check.
  • 179.2.4.3 Register Concordia protobuf under libs/proto/src/concordia/ with buf lint/generate coverage, generated descriptor updates, package exports, and compatibility checks for streaming RPC changes.
  • 179.2.4.4 Mount the Oshun workbench inside the existing apps/oshun/web App Router at /studio/concordia-workbench, with route metadata, navigation hooks, Playwright fixtures, and API mock boundaries.
  • 179.2.4.5 Define whether services/concordia/orchestrator/ is a new Fastify service or a Python service before implementation; align Docker, project targets, health checks, OpenAPI export, and service discovery with the chosen repo convention.

179.2.5 Identity, Authority, and Access Control#

  • 179.2.5.1 Add party identity verification, representative delegation, organization authority proof, guardian / counsel linkage, and signatory verification states before any acceptance or execution workflow.
  • 179.2.5.2 Implement RBAC / ABAC policies for party, mediator, reviewer, counsel, procurement officer, DAO steward, safety reviewer, and auditor roles with tenant and case-level isolation.
  • 179.2.5.3 Add conflict-of-interest disclosures and assignment rules for mediators, reviewers, counsel, procurement approvers, and DAO stewards.
  • 179.2.5.4 Add negative authorization tests proving users cannot infer, list, search, export, log, or subscribe to another party's private fields, another tenant's cases, or reviewer-only materials.
  • 179.2.5.5 Add rate limits, abuse throttles, session controls, and suspicious-access alerts for case invitations, evidence downloads, private-intake reads, and agreement exports.

179.3 Private Intake, Preference Learning, and Utility Inference#

179.3.1 Party-Isolated Intake Agents — @iris/concordia-assistant#

  • 179.3.1.1 Build private intake conversation flows that elicit facts, interests, needs, acceptable outcomes, redlines, emotions, relationship context, objective evidence, BATNA, reservation points, and authority to settle.
  • 179.3.1.2 Add trauma-aware and coercion-aware language for sensitive contexts, routing to Kuanyin / human review when intimidation, abuse, power imbalance, self-harm, or immediate danger appears.
  • 179.3.1.3 Add multilingual intake with language-preserving summaries and explicit detection of internal-reasoning language drift for multilingual negotiation.
  • 179.3.1.4 Add intake completeness scoring: unresolved facts, missing evidence, ambiguous terms, unverified authority, hidden stakeholders, and unmodeled constraints.
  • 179.3.1.5 Add party review screens where each party can correct their inferred priorities before any search run begins.

179.3.2 Pairwise Preference Engine — @nous/preference-inference#

  • 179.3.2.1 Implement pairwise comparison prompts: "Given this party's private statement and priorities, would they prefer agreement A or B, and why?" with strict JSON output and citation to private intake snippets.
  • 179.3.2.2 Train and calibrate Bradley-Terry, Thurstone-Mosteller, Plackett-Luce, Gaussian-process preference, and neural utility-ranking models from pairwise comparisons.
  • 179.3.2.3 Add active-learning selection of comparison pairs that maximally reduce uncertainty near the Pareto frontier and party redlines.
  • 179.3.2.4 Add preference-stability tests: paraphrase robustness, prompt-template robustness, model-version robustness, order-effect robustness, and adversarial option framing.
  • 179.3.2.5 Add uncertainty output to every utility score: posterior mean, credible interval, comparison count, nearest known comparison, and instability warnings.
  • 179.3.2.6 Add abstention and incomparability states for low-confidence, unstable, refused, redline-blocked, or morally / legally non-tradeable comparisons; prevent those utilities from entering optimizer selection.

179.3.3 Multi-Attribute Utility and Fairness Models#

  • 179.3.3.1 Infer issue weights, diminishing returns, time-discounting, risk aversion, inequity aversion, relationship preservation, procedural fairness, and non-monetary preferences.
  • 179.3.3.2 Support hard constraints and soft preferences separately so optimizer search cannot launder redline violations into high utility.
  • 179.3.3.3 Model BATNA and walkaway utility with a confidence interval and strategic-manipulation warning instead of treating self-reported BATNA as ground truth.
  • 179.3.3.4 Implement fairness metrics: Nash product, utilitarian sum, max-min, egalitarian welfare, Kalai-Smorodinsky proportional gains, envy-freeness, regret, inequality, burden symmetry, and procedural dignity.
  • 179.3.3.5 Add cultural, organizational, and domain-specific fairness profiles without letting demographic inference drive protected-class disparate treatment.

179.4.1 Candidate Generation#

  • 179.4.1.1 Generate initial candidates from party proposals, mediator templates, domain playbooks, precedent agreements, policy constraints, and LLM brainstormed tradeoff bundles.
  • 179.4.1.2 Add structured proposal decomposition so one candidate can borrow deadlines from another, payment schedules from another, governance rights from another, and restorative obligations from another.
  • 179.4.1.3 Implement clause mutators in a sandboxed DSL instead of raw prompt text: numeric perturbation, deadline shift, proportional split, installment schedule, equity vesting, royalty waterfall, scope narrowing, escalation ladder, audit right, and reversible trial period.
  • 179.4.1.4 Add domain candidate generators for procurement, creative royalties, DAO governance, platform moderation appeals, marketplace disputes, cofounder equity, supplier service levels, collaborative production, multiplayer guild governance, and agent-to-agent contracts.
  • 179.4.1.5 Filter generated candidates through safety, legality, policy, feasibility, and enforceability gates before utility scoring.
  • 179.4.2.1 Implement Nash-product genetic search matching the Mediator.ai public architecture, then extend it with uncertainty-aware selection and formal clause validation.
  • 179.4.2.2 Implement multi-objective evolutionary search (NSGA-II / MAP-Elites) over the Pareto frontier with diversity preservation so users see genuinely different agreement families.
  • 179.4.2.3 Implement MCTS / LATS agreement search using @nous/tree-search-reasoning from Phase 178, where nodes are candidate agreements and actions are clause edits.
  • 179.4.2.4 Implement CP-SAT / MILP search for agreements with discrete constraints: schedules, budgets, ownership percentages, milestones, roster assignments, routing, shipment terms, and resource allocation.
  • 179.4.2.5 Implement Bayesian optimization for expensive scoring cases where each utility evaluation requires a frontier model, legal reviewer, simulation, or external pricing calculation.
  • 179.4.2.6 Implement PSRO / opponent-model search for agent-to-agent and competitive commercial negotiations where parties may strategically adapt.
  • 179.4.2.7 Implement multi-party coalition stability: core membership, blocking coalition detection, Shapley-value attribution, bargaining set, and side-payment generation.

179.4.3 Explanation and Choice Architecture#

  • 179.4.3.1 Render Pareto-frontier explanations showing why no party can be improved without worsening another under the current utility models.
  • 179.4.3.2 Provide party-specific private explanations: "why this may be acceptable to you," "where it asks you to concede," "what you gain over BATNA," and "what remains uncertain."
  • 179.4.3.3 Provide shared neutral explanations that do not reveal private concessions, reservation points, or confidential strategy.
  • 179.4.3.4 Add "show me more like this," "protect this term," "do not show this style again," and "human mediator review" controls.
  • 179.4.3.5 Add coercive-choice safeguards: no dark patterns, countdown-pressure, hidden settlement authority, manipulated anchors, or asymmetric disclosure.

179.5 Safety, Security, Privacy, and Anti-Gaming#

179.5.1 Confidentiality and Party Isolation#

  • 179.5.1.1 Implement physically separate prompt contexts per party; no shared model call may include opposing-party private fields.
  • 179.5.1.2 Add sealed-memory stores with envelope encryption, per-party data keys, HSM/KMS integration, and audit logs for every read.
  • 179.5.1.3 Add zero-retention local-model mode through Nous for privileged, sensitive, or air-gapped mediation.
  • 179.5.1.4 Explore secure enclave / confidential-compute execution for utility scoring where the platform should not see raw party inputs.
  • 179.5.1.5 Add privacy red-team tests for prompt injection, tool exfiltration, accidental summary leakage, side-channel inference, and shared-explanation leakage.

179.5.2 Manipulation, Deception, and Strategic Misrepresentation#

  • 179.5.2.1 Add BATNA plausibility checks against objective evidence, market data, historical settlements, internal business data, or domain-specific simulators where available.
  • 179.5.2.2 Detect inconsistent statements across private intake, evidence uploads, live negotiation, and final acceptance.
  • 179.5.2.3 Add adversarial candidate tests for specification gaming: agreements that score well but contain brittle, exploitative, vague, or one-sided clauses.
  • 179.5.2.4 Detect collusion and coercion in multi-party cases: identical language, external intimidation cues, unusual concession patterns, hidden side payments, and forced acceptance.
  • 179.5.2.5 Add "cooling-off" and delayed confirmation for sensitive or high-value settlements.

179.5.3 Governance and Human Oversight#

  • 179.5.3.1 Build reviewer queues for human mediators, counsel, compliance, DAO stewards, procurement officers, safety teams, and domain experts.
  • 179.5.3.2 Add human override, pause, reopen, appeal, and nullification workflows with complete audit trail.
  • 179.5.3.3 Add model and optimizer version pinning so every accepted agreement can be reproduced exactly.
  • 179.5.3.4 Add incident reporting for unsafe recommendations, privacy leaks, biased outcomes, unauthorized acceptance, or execution mismatch.
  • 179.5.3.5 Integrate with Maat governance and Themis audit reports for periodic board-level oversight of mediation automation.
  • 179.5.4.1 Build a consent ledger for collection, processing, sharing, model scoring, reviewer access, meeting transcription, settlement export, training-data reuse, and smart-contract execution, with per-purpose timestamps and revocation state.
  • 179.5.4.2 Implement revocation and withdrawal workflows: stop future processing, quarantine derived utility models, notify affected reviewers, preserve legally required audit records, and explain what cannot be deleted because of legal hold or already-shared settlement artifacts.
  • 179.5.4.3 Add evidence chain-of-custody: malware scanning, file hash, uploader identity, source assertion, timestamp, redaction state, authenticity confidence, privilege flag, retention class, and tamper detection.
  • 179.5.4.4 Add redaction and minimization pipelines before evidence or private statements enter summaries, prompts, vector stores, logs, candidate clauses, shared explanations, or benchmark datasets.
  • 179.5.4.5 Add data-subject export, deletion, legal-hold, residency, and retention workflows for raw statements, evidence, model-derived utility artifacts, search traces, audit events, and settlement documents.

179.5.5 Operational Oversight and Settlement Lifecycle#

  • 179.5.5.1 Define reviewer capacity controls: queue SLA, assignment policy, dual-control for high-risk cases, calibration review, audit sampling, escalation when no qualified reviewer is available, and blocked launch status for unstaffed review gates.
  • 179.5.5.2 Add off-chain settlement lifecycle support: versioned redlines, counsel comments, e-signature / countersignature, document export, acceptance ceremony, amendment, rescission, default notice, remedy tracking, and obligation completion monitoring.
  • 179.5.5.3 Add kill-switch and feature-flag controls for mediation model routing, autonomous acceptance, smart-contract execution, search kernels, and domain adapters.
  • 179.5.5.4 Add financial and procurement controls for sanctions / KYC, payment rails, invoice reconciliation, tax treatment, currency exposure, approval thresholds, and segregation of duties.
  • 179.5.5.5 Add training-data governance: de-identification review, re-identification risk scoring, sensitive-case exclusion, consent propagation, dataset lineage, removal / unlearning policy, and model-card linkage for any trace reused in Phases 85-96.

179.6 Product Surfaces and APIs#

179.6.1 Oshun Concordia Workbench#

  • 179.6.1.1 Build apps/oshun/web/src/app/studio/concordia-workbench/ for case creation, party invitation, private intake, issue map, evidence room, preference review, candidate frontier, settlement drafting, review routing, and audit export.
  • 179.6.1.2 Add mediator mode: live session transcript, agreement-point tracker, talk-balance metrics, suggested reframes, caucus notes, offer ladder, and private mediator-only preparation report.
  • 179.6.1.3 Add party mode: private statement editor, correction of inferred interests, redline controls, candidate comparison, questions for mediator, and acceptance confirmation.
  • 179.6.1.4 Add organization mode for procurement and governance: settlement authority, approval chain, budget constraints, policy bundles, signatory status, and execution integration.
  • 179.6.1.5 Add accessibility, localization, and low-bandwidth support so mediation is not limited to high-end devices or English speakers.

179.6.2 Iris Concordia Assistant#

  • 179.6.2.1 Add an Iris "Concordia" agent archetype for low-stakes conflict brainstorming, shared-plan negotiation, calendar coordination, household agreements, team task allocation, and creator collaboration.
  • 179.6.2.2 Add two-device / multi-device private intake where each participant can talk to Iris separately and consent to shared synthesis.
  • 179.6.2.3 Add "turn this conversation into a structured agreement" commands that produce an AgreementDSL draft with review flags.
  • 179.6.2.4 Add meeting co-mediator mode for Teams / Zoom / WebRTC / Oshun calls, with explicit consent, transcription disclosure, and participant controls.
  • 179.6.2.5 Add "not appropriate for AI mediation" detections that route to human support or professional resources.

179.6.3 Service Orchestrator#

  • 179.6.3.1 Create services/concordia/orchestrator/ with Fastify or Python FastAPI gateway matching the repo's service conventions.
  • 179.6.3.2 Add background workers for intake summarization, preference comparison batches, search runs, candidate validation, legal/policy template matching, and settlement drafting.
  • 179.6.3.3 Add event-bus integration with @oshun/contracts/events and queue backpressure so search runs cannot starve user-facing services.
  • 179.6.3.4 Add observability: traces per case, scoring cost, model latency, candidate diversity, preference uncertainty, agreement rate, escalation rate, and privacy-gate failures.
  • 179.6.3.5 Add tenant and domain configuration: legal disclaimers, eligible use cases, model routing, review thresholds, data residency, execution adapters, and retention policy.

179.7 Ecosystem Integration Map#

179.7.1 Maat and Industrial / Organizational Negotiation#

  • 179.7.1.1 Integrate with @maat/strategy, @maat/finance, @maat/supply-chain, and @maat/agents for supplier pricing, payment terms, rebates, service levels, joint business plans, capital allocation, shared services, hiring offers, board resolutions, and cross-company resource conflicts.
  • 179.7.1.2 Build Pactum-class procurement agents for Asase, Freya, Cybele, Brigid, Saraswati, Aje, and Maat portfolio companies.
  • 179.7.1.3 Add working-capital optimization: DPO extension, early payment discounts, volume commitments, quality guarantees, warranty terms, and supplier relationship scoring.
  • 179.7.1.4 Add executive approval gates for thresholds by value, supplier criticality, jurisdiction, sanctions/KYC risk, and reputation impact.
  • 179.7.1.5 Benchmark procurement outcomes against manual historical negotiations: savings, cycle time, supplier satisfaction, dispute rate, and realized value.

179.7.2 Themis, Aje, and Governance / Settlement Execution#

  • 179.7.2.1 Integrate with Themis proposal lifecycle so contentious DAO, corporate, community, and policy proposals can enter mediated amendment before adversarial voting.
  • 179.7.2.2 Add governance compromise search: budget reallocations, phased rollouts, quorum changes, amendment bundles, delegation terms, sunset clauses, audit rights, and minority protections.
  • 179.7.2.3 Add Themis dispute modules for election challenges, delegate misconduct, IP originality appeals, moderation appeals, creator-marketplace conflicts, and inter-DAO agreements.
  • 179.7.2.4 Integrate Aje smart-contract escrow for deposits, milestone payouts, royalties, refunds, arbitration bonds, agreement-hash anchoring, and on-chain execution after required review.
  • 179.7.2.5 Add Kleros / optimistic challenge style adapters as optional arbitration backstops without making them the default decision-maker.

179.7.3 Kuanyin, Lilith, Tara, and Restorative Safety#

  • 179.7.3.1 Integrate with Kuanyin restorative circles for community harm repair, apology, restitution, reintegration, moderation appeals, and performer / creator protection.
  • 179.7.3.2 Add safety-specific terms: no-contact boundaries, content takedown timelines, community participation limits, apology review, educational completion, monitoring windows, and escalation triggers.
  • 179.7.3.3 Add Lilith and Aphrodite boundary negotiation support with strict consent, adult-safety, exploitation, age-verification, and coercion controls.
  • 179.7.3.4 Add Tara wellness conflict support for low-stakes relational reflection while routing therapy, abuse, crisis, or legal claims out of automated mediation.
  • 179.7.3.5 Measure restorative outcomes: recurrence, participant sense of fairness, safety incidents, completion of agreed obligations, and community health.

179.7.4 Creative Production and IP Domains#

  • 179.7.4.1 Integrate Yemaya production workflows for credit, scheduling, budget, royalty, revision scope, approval authority, producer conflicts, remote crew disputes, and delivery acceptance.
  • 179.7.4.2 Integrate Euterpe / Calliope royalty splits, songwriting credits, sample-clearance negotiations, performance agreements, label / artist terms, and AI-generated music originality disputes.
  • 179.7.4.3 Integrate Isis generation workflows for model/provider disputes, asset rights, prompt provenance conflicts, regeneration credits, dataset licensing, and post-production acceptance.
  • 179.7.4.4 Integrate Aja, Bellona, Neith, Maya, and Hathor for motion capture acceptance, game-asset marketplace disputes, mod governance, guild/resource conflicts, narrative canon disputes, and virtual-world social contracts.
  • 179.7.4.5 Integrate Themis Universal Originality Shield so mediated IP settlements cite originality evidence, source overlaps, license lineage, and confidence bands.

179.7.5 Commerce, Marketplace, and Consumer Domains#

  • 179.7.5.1 Integrate Aglaea and Freya for returns, custom-order disputes, fit/alteration compromises, pricing offers, resale authenticity, creator collabs, and luxury-service recovery.
  • 179.7.5.2 Integrate Asase, Demeter, Hestia, and Annapurna for supplier quality disputes, crop delivery issues, restaurant vendor terms, food safety remediation, and contract farming agreements.
  • 179.7.5.3 Integrate Cybele, Seshat, Athena, Brigid, and Saraswati for construction change orders, workshop bids, machine downtime claims, warranty negotiations, R&D partnership terms, and service-level credits.
  • 179.7.5.4 Integrate Lakshmi and Aje for debt restructuring, subscription refunds, payment plans, escrow disputes, token-sale support, and investment-club governance with regulated-finance review gates.
  • 179.7.5.5 Integrate Nyx, Metis, Mnemosyne, Nisaba, Kalika, and Veritas for collaborative research authorship, citation disputes, educational group work, publication corrections, editorial conflicts, and evidence quality disagreements.

179.7.6 AI-Agent and Machine-to-Machine Negotiation#

  • 179.7.6.1 Add agent-to-agent negotiation protocol for Iris / Nous / Maat agents to contract over tasks, budgets, deadlines, tool access, compute allocation, data-sharing, and expected quality.
  • 179.7.6.2 Add capability negotiation for MCP servers, plugins, Neith engine services, Maya live-service shards, edge devices, and Oya drone swarms.
  • 179.7.6.3 Add policy-constrained autonomous acceptance where agents can accept only within explicit authority bounds and must log all offers.
  • 179.7.6.4 Add adversarial negotiation tests for prompt injection, hidden-tool access, budget escalation, side-channel data leakage, and collusive agent behavior.
  • 179.7.6.5 Add economic-value scoring for agent negotiations: utility captured, relationship impact, compute cost, latency, safety violations, and human intervention avoided.

179.8 Evaluation, Benchmarks, and Red Teaming#

179.8.1 Benchmark Harness#

  • 179.8.1.1 Create testing/concordia/ with synthetic two-party, multi-party, procurement, legal-low-stakes, DAO, creative, restorative, and agent-to-agent benchmark suites.
  • 179.8.1.2 Reproduce public examples from Mediator.ai-style scenarios: cofounder equity, roommate chores, shared parenting-plan brainstorming, contractor payment, supplier terms, creative royalties, and marketplace refunds.
  • 179.8.1.3 Implement ANAC / GeniusWeb / NegMAS adapters so Oshun agents can compete against classic automated negotiation baselines.
  • 179.8.1.4 Implement LLMediator / Robots-in-the-Middle-style mediator intervention evaluation: intervention type, wording quality, impartiality, understanding, contextualization, and appropriateness.
  • 179.8.1.5 Implement ProMediate-style proactive mediator evaluation: consensus change, intervention latency, targetedness, participant satisfaction, topic coverage, and socio-cognitive intelligence.

179.8.2 Outcome Metrics#

  • 179.8.2.1 Track agreement rate, Pareto efficiency, Nash product, Kalai-Smorodinsky distance, welfare sum, max-min utility, envy, regret, candidate diversity, and stability under preference uncertainty.
  • 179.8.2.2 Track subjective value: perceived fairness, dignity, procedural transparency, relationship preservation, voice, control, and willingness to use again.
  • 179.8.2.3 Track domain metrics: procurement savings, DPO change, supplier satisfaction, moderation recurrence, DAO proposal pass rate, creative delivery acceptance, marketplace refund leakage, and settlement execution completion.
  • 179.8.2.4 Track cost: model spend, search iterations, human review minutes, latency, GPU time, and acceptance per dollar.
  • 179.8.2.5 Track safety: privacy leaks, redline violations, legal-review misses, coercion flags, biased outcomes, hallucinated law/policy, and unenforceable clause rate.

179.8.3 Red Team and Release Gates#

  • 179.8.3.1 Build adversarial party simulations: bluffing BATNA, false evidence, emotional manipulation, prompt injection, private-info fishing, hidden side deals, collusion, delay tactics, and bad-faith leverage seeking.
  • 179.8.3.2 Build sensitive-case refusal tests for domestic abuse, custody, employment rights, medical care, criminal allegations, regulated finance, immigration, and minors.
  • 179.8.3.3 Build clause exploit tests: ambiguous deadlines, unconstrained waiver, impossible deliverable, unconscionable penalty, hidden fee, one-sided confidentiality, assignment trap, and perpetual license overreach.
  • 179.8.3.4 Require release gates: no private-info leakage, calibrated uncertainty, stable pairwise preferences, safe refusal, human-review routing, audit reproducibility, and passing fairness thresholds.
  • 179.8.3.5 Publish internal leaderboards comparing model backbones, prompts, preference learners, search kernels, and domain adapters.

179.9 Implementation Sequence#

179.9.1 Phase A — Foundation#

  • 179.9.1.1 Land contracts, OpenAPI, protobuf, event schemas, privacy views, and generated clients.
  • 179.9.1.2 Land private intake, issue map, agreement DSL, and basic candidate generation.
  • 179.9.1.3 Land pairwise preference inference with uncertainty and redline separation.
  • 179.9.1.4 Land Nash-product genetic search baseline plus Pareto frontier visualization.
  • 179.9.1.5 Land benchmark harness and privacy red-team tests before any domain rollout.
  • 179.9.1.6 Land repo-standard scaffolding, generated-artifact registration, identity / authority checks, RBAC / ABAC policies, consent ledger, evidence custody, and abstention states before any workbench flow can accept real party data.

179.9.2 Phase B — Surpass Mediator.ai#

  • 179.9.2.1 Add multi-party, coalition, CP-SAT, MCTS, Bayesian, and multi-objective search kernels.
  • 179.9.2.2 Add formal clause validation, executable agreement DSL, and shared/private explanation boundaries.
  • 179.9.2.3 Add uncertainty-aware active learning and stability tests for every inferred utility model.
  • 179.9.2.4 Add human mediator workbench, live co-mediator mode, and domain review queues.
  • 179.9.2.5 Add procurement and DAO first-class integrations to prove value beyond consumer/legal mediation.
  • 179.9.2.6 Add reviewer capacity operations, off-chain settlement lifecycle, kill-switch controls, and negative authorization automation before any autonomous execution or smart-contract settlement path.

179.9.3 Phase C — Ecosystem Rollout#

  • 179.9.3.1 Roll out to Maat procurement, Themis governance, Kuanyin restorative moderation, Iris personal coordination, Aje escrow, and Yemaya creative collaboration as the first production pilots.
  • 179.9.3.2 Roll out to Euterpe / Calliope royalties, Maya marketplace and guild governance, Lilith / Aphrodite safety-limited boundaries, Aglaea / Freya commerce, Cybele / Brigid industrial contracts, and Metis / Kalika / Veritas evidence disputes.
  • 179.9.3.3 Add domain adapters for every remaining domain in DOMAINS/README.md, even if initially template-only.
  • 179.9.3.4 Add agent-to-agent negotiation for compute, tools, budget, and service-level agreements across Iris / Nous / Neith / Maya.
  • 179.9.3.5 Add internal training data flywheel from de-identified, consented, outcome-scored mediation traces into Phases 85-96.

179.9.4 Phase D — Bleeding-Edge SOTA#

  • 179.9.4.1 Implement confidential-compute preference scoring and privacy-preserving utility aggregation.
  • 179.9.4.2 Implement learned mediator policies trained from simulated negotiation, human mediator review, and outcome feedback.
  • 179.9.4.3 Implement causal mediation models distinguishing agreements that merely look fair from agreements likely to remain durable.
  • 179.9.4.4 Implement cross-cultural, multilingual, and accessibility calibrated negotiation models with fairness audits.
  • 179.9.4.5 Implement self-play and automated negotiation leagues inside Oshun, with continuous adversarial improvement of strategies and safeguards.

179.10 Verification and Completion Criteria#

  • 179.10.1 pnpm contracts:check passes with Concordia schemas, generated OpenAPI clients, protobuf artifacts, and event schemas.
  • 179.10.2 Unit tests cover preference inference, agreement DSL validation, search kernels, privacy views, and domain adapters.
  • 179.10.3 Integration tests cover private intake, pairwise scoring, search run, draft generation, review routing, acceptance, execution, and audit export.
  • 179.10.4 Playwright coverage verifies workbench flows for mediator, party, reviewer, procurement officer, DAO steward, and safety reviewer.
  • 179.10.5 Security tests prove private party data cannot leak through shared summaries, explanations, logs, traces, generated agreements, or domain events.
  • 179.10.6 Benchmark gates show improvement over baseline LLM mediator, simple Nash genetic search, static templates, and human-authored seed-offer-only workflows on at least six domain suites.
  • 179.10.7 Governance review approves use-case boundaries, high-risk routing, human oversight, audit logging, data retention, model cards, and incident response.
  • 179.10.8 Pilot deployments in Maat, Themis, Kuanyin, Iris, Aje, and Yemaya show measurable value without unresolved P0/P1 safety or privacy incidents. Local readiness is complete: the six-domain fail-closed evidence gate is implemented in libs/contracts/concordia/src/pilot-evaluation/, all six policy-bundle cards are populated, and the dedicated incident runbook is at docs/runbooks/concordia-incident-response.md; evidence collection and exit criteria are defined in docs/runbooks/concordia-pilot-evidence.md. The checkbox remains open until signed production-pilot deployment, outcome, and incident evidence for every named domain passes evaluatePilotProgram; fixtures and repository tests are intentionally not treated as deployments. Tagged 2026-09-18 from the body above rather than from a fresh judgement: the local half is complete and named, and what remains is six signed production pilots in other domains. Nothing in this checkout can produce that evidence — the body already refuses to let fixtures stand in for it. blocked:release
  • 179.10.9 Repo integration checks prove Concordia packages appear in the Nx graph, pnpm contracts:check covers the actual libs/openapi/src/specs/concordia/ and libs/proto/src/concordia/ paths, generated artifacts are stable, and no stale top-level spec paths are referenced by codegen.
  • 179.10.10 Authorization automation covers RBAC / ABAC denial, tenant-boundary denial, reviewer-only denial, cross-party prompt denial, event-subscription denial, and audit-export denial for private fields.
  • 179.10.11 Consent and evidence tests cover collection consent, sharing consent, consent revocation, withdrawal, legal hold, redaction, evidence chain-of-custody, malware rejection, tamper detection, and data export / deletion workflows.
  • 179.10.12 Accessibility automation covers axe / WCAG checks, keyboard-only flows, screen-reader labels, focus order, localization, low-bandwidth mode, and mobile viewport behavior for every workbench role.
  • 179.10.13 Settlement lifecycle tests cover versioned drafts, countersignature, authority verification, off-chain document export, smart-contract anchoring, amendment, rescission, default / remedy tracking, and obligation completion.
  • 179.10.14 Training-data gates prove reused traces are consented, de-identified, lineage-tracked, sensitive-case filtered, revocation-aware, and documented in model cards before entering Phases 85-96.

Source Pointers#

  • Mediator.ai, "Cooperative negotiation is a solvable problem": https://mediator.ai/
  • Mediator.ai, "Using Nash Bargaining and LLMs to Systematize Fairness": https://mediator.ai/blog/ai-negotiation-nash-bargaining/
  • Mediator.ai bakery agreement example: https://mediator.ai/examples/bakery-partnership-agreement/
  • LLMediator: GPT-4 Assisted Online Dispute Resolution: https://arxiv.org/abs/2307.16732
  • Robots in the Middle: Evaluating LLMs in Dispute Resolution: https://arxiv.org/abs/2410.07053
  • ProMediate: socio-cognitive framework for proactive multi-party negotiation: https://arxiv.org/abs/2510.25224
  • Advancing AI Negotiations: large-scale autonomous negotiation competition: https://arxiv.org/abs/2503.06416
  • The Price of Thought: reasoning, performance, and cost of negotiation in LLMs: https://arxiv.org/abs/2510.08098
  • LLM Rationalis? Measuring Bargaining Capabilities of AI Negotiators: https://arxiv.org/abs/2512.13063
  • ANAC Automated Negotiating Agents Competition: https://scml.cs.brown.edu/
  • Pactum autonomous procurement negotiation: https://pactum.com/
  • Pactum procurement agents: https://pactum.com/procurement-agents
  • MediationAI: https://www.mediationai.app/
  • Bot Mediation: https://botmediation.com/
  • ZODR AI co-mediator: https://www.zodr.ai/
  • Disputell pre-mediation preparation: https://www.disputell.com/
  • UNCITRAL Online Dispute Resolution: https://uncitral.un.org/en/texts/onlinedispute
  • EU AI Act overview: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
  • NIST AI Risk Management Framework: https://www.nist.gov/itl/ai-risk-management-framework