Concordia Research — Source Matrix

Matrix date: 2026-04-23

Scope: Phase 179 Concordia task 179.1.1.2.

This matrix is the per-source backing layer for the competitive audit at mediation-negotiation-2026.md and for the bibliography work in 179.1.1.5. It separates product claims (vendor marketing, self-published architecture posts, self-reported metrics) from evidence (peer-reviewed work, preprints that have been independently evaluated, standards texts, regulatory instruments, benchmark harnesses, and documented deployments).

Every source below carries a primary category tag from the taxonomy required by Phase 179.1.1.2:

• commercial — vendor site, blog, press release, self-published whitepaper
• academic — peer-reviewed paper or widely cited preprint
• standards — multilateral or professional-body text
• regulatory — government or regulator text
• benchmark — public evaluation harness or competition
• case-study — documented deployment, third-party analysis, or long-form journalism that functions as a case study

A secondary claim-type column distinguishes how the content should be treated when building Concordia:

• marketing — aspirational, unverified, vendor-framed
• self-reported — quantified vendor claim with no independent verification
• self-architecture — vendor-published technical description (treat as serious input but not as proof)
• independent — peer-reviewed, independently measured, or third-party analyzed
• primary — authoritative source for the text of a rule, law, standard, or protocol
• secondary — journalism or analyst coverage interpreting a primary source

An evidence-strength heuristic summarizes how much weight a reader should give the source when making product decisions:

• A — peer-reviewed, standards, regulatory primary text, large-n benchmark, or a long-running high-volume deployment
• B — well-sourced press, preprint with methodology, documented third-party deployment, certified compliance posture (SOC 2 Type II, ISO 27001)
• C — vendor-self-published architecture or outcome claims with named customers
• D — marketing copy or unverifiable self-reported metrics

Column legend#

A. AI mediation, ODR, and cooperative-negotiation products#

B. Commercial negotiation agents (procurement, commerce)#

C. Decentralized dispute resolution#

D. ODR platform and history reference set#

E. Institutional ADR providers#

F. Human mediator-assist tooling#

G. Academic and research corpus (from Phase 179 intro + bibliography)#

These are the peer-reviewed or widely cited preprint sources that anchor Concordia's technical claims. They are tracked in this matrix so future Concordia code and product decisions cite load-bearing research rather than vendor marketing.

H. Standards, regulatory, and governance#

I. Case studies and documented deployments#

Claim-vs-evidence partition#

Product claims to treat as marketing unless or until independent evidence appears:

• Mediator.ai, MediationAI, TheMediator.AI self-reported metrics and "court-enforceable" framing.
• ODR.com "1.1B disputes / 60M per year" aggregate — inherits eBay's figure, not a deployed-platform metric.
• Any "SOC 2-aligned" language that is not backed by a SOC 2 Type II report.

Product claims that reach self-architecture (treat as serious input but not as proof):

• Mediator.ai's Nash-bargaining / GA / LLM-ranker blog post.
• ZODR's overlay-on-Zoom mediator-assist description.
• Pactum and Nibble's agentic-negotiation descriptions with disclosed compliance certifications.

Evidence-grade sources for Concordia decisions:

• Peer-reviewed work (LLMediator, Robots in the Middle, ProMediate, Advancing AI Negotiations, Price of Thought, LLM Rationalis?, Rule & Del Duca, foundational Nash / Kalai-Smorodinsky / Rubinstein).
• Primary standards and regulatory text (UNCITRAL, EU AI Act, NIST AI RMF, JAMS AI Rules, ABA Model Rules, Colorado AI Act).
• Documented deployments at scale (eBay / PayPal, BC CRT, Utah courts, Matterhorn, Walmart-via-Pactum independent reporting).

Refresh policy#

Per 179.1.1.6, this matrix is re-run before every Phase 179 phase transition and before every pilot. Specifically:

1. Re-fetch every row with a f:YYYY-MM-DD date; update if content shifted.
2. Re-check (403) and (404) rows from a non-blocked fetcher.
3. Pull the latest arXiv versions of rows under section G; add any new Concordia-relevant preprints that appeared since the last refresh.
4. Re-check EU AI Act implementing acts, NIST AI RMF updates, and state-level AI regulation (Colorado, California, New York) under section H.
5. Archive a dated copy of each product homepage under docs/research/archive/YYYY-MM-DD/<source>.html when a pilot is about to launch; the Phase 179 refresh gate blocks the pilot if archives are missing.