{"path":"competitive-landscape.md","content":"# Deliberus Competitive Landscape\n\n## Existing Platforms\n\n*The wider frame for this landscape: civilization has working epistemic infrastructure for conclusions, quality control, agreement-shape, and procedure, and none for reasoning itself. The taxonomy lives in [The Missing Layer](the-missing-layer.md); the platforms below are the nearest neighbors in adjacent slots.*\n\n### Historical Note: Delibr / DelibrAI\n\n**Delibr** (now DelibrAI) was co-founded with Nils Janse during the 2012-2013 startup phase. Nils pivoted it toward a commercial SaaS product; Fredrik kept the deeper vision. They are different products with different goals. The non-compete expired ~2014. No overlap with what Deliberus is becoming — Delibr went commercial/pragmatic, Deliberus goes epistemological/societal.\n\n### Argument Mapping / Structured Debate\n\n| Platform | Status | What it does | What Deliberus does differently |\n|----------|--------|-------------|--------------------------------|\n| **Kialo** | Active | Pro/con debate trees | Kialo is flat pro/con. Deliberus has multi-axis scoring, argument bundles, evidence linking, LLM-assisted extraction |\n| **DebateGraph** | Active (niche) | Visual argument mapping | Academic/niche. Deliberus targets broader adoption with LLM friction reduction |\n| **Argdown** | Active | Argument markup language | Developer tool, not a platform. Could be a useful interchange format |\n| **Argunet** | Academic | Argument analysis software | Research tool, not consumer-facing |\n\n### Collective Intelligence / Democracy\n\n| Platform | Status | What it does | What Deliberus does differently |\n|----------|--------|-------------|--------------------------------|\n| **Pol.is** | Active | Opinion clustering via matrix factorization | Pol.is finds consensus clusters but doesn't map argument structure. No logical validation |\n| **Decidim** | Active | Participatory democracy (Barcelona-origin) | Process-focused (proposals, voting, budgets). Not argument-structure-focused |\n| **Loomio** | Active | Collaborative decision-making | Meeting/proposal workflow. Not argument mapping |\n| **ConsiderIt** | Unclear | Deliberative polling interface | Simpler pro/con weighing, no deep structure |\n| **Remesh** | Active (enterprise) | AI-moderated group conversations | Enterprise survey tool, not open argumentation |\n\n### Knowledge Graphs / Personal Knowledge\n\n**Reframed Aug 2026, and it changes which row matters.** This table was built around argument-mapping platforms as the competitive set. Working out what the typed graph actually buys ([structure-versus-scale.md](research/structure-versus-scale.md)) identified a different primary rival: **not unstructured LLM reasoning, and not Kialo, but a well-kept prose wiki with a language model over it.** Cheap structure — good writing, stable permalinks, a model that reads the whole corpus — already delivers persistence, addressability and contestability, which are the properties the project's own positioning had been leaning on. So the row flagged below is the one to argue against, and the argument has to be made on what prose *cannot* do: enumerate what a summary dropped, and hold what remains unsettled. Everything else in this table is a weaker competitor than that row.\n\n| Platform | Status | What it does | What Deliberus does differently |\n|----------|--------|-------------|--------------------------------|\n| **Roam Research** | Active | Networked thought / personal knowledge | Personal, not collaborative argumentation. No logical structure |\n| **Obsidian** | Active | Personal knowledge management | Same — personal, not social/deliberative |\n| **Karpathy LLM Wiki** ⚠ **primary rival** | Described 2026 | LLM compiles raw sources into plain-Markdown wiki; no RAG; LLM reads its own index; health checks for gaps/contradictions | Deliberus uses typed graph (not prose); claims are extracted (not compiled); shared graph (not personal); LLM drafts, human confirms (not LLM writes authoritatively). Both reject RAG/embeddings as primary storage. See [research/wiki-that-writes-itself-and-productive-friction.md](research/wiki-that-writes-itself-and-productive-friction.md) |\n| **NotebookLM** (Google) | Active | Uploads sources, generates podcast overviews, mind maps, flashcards | Static by default; lives in Google's cloud; no argument structure, no community checking, no bridging signal. Extended Brain critique: \"excels at effortless consumption but may short-circuit genuine understanding\" |\n\n### Academic / Research Systems\n\n| System | Status | What it does | Relevance |\n|--------|--------|-------------|-----------|\n| **MIT Deliberatorium** | Inactive? | Large-scale deliberation research | Validated the scaling approach. Learn from what failed (adoption) |\n| **Argument Web** (Chris Reed) | Active research | Largest public argumentation corpora | Dataset source, interoperability standards |\n| **DeliData** | Dataset (2023) | 500 group dialogues, 14k utterances | Validation: 64% of groups found better solutions than individuals |\n\n### AI-Augmented / Modern Entrants (2024-2026)\n\n| Platform | Status | What it does | Relevance |\n|----------|--------|-------------|-----------|\n| **Claimify** (MS Research) | ACL 2025 paper | LLM-based atomic claim extraction | Core technology for Deliberus ingestion pipeline |\n| **ClaimsMCP** | Open source (13 stars) | MCP server for Claimify methodology | Directly pluggable into existing setup |\n| **DELIBERATION.IO** | Active (Stanford+MIT) | Opinion polling with AI-mediated Socratic questioning + steelman synthesis. Firebase + FlutterFlow + GPT-4. NOT an argumentation platform — no claims, evidence, logical relationships, or argument structure. Closer to Polis than to Kialo. | Deployed in D.C. Jul 2025. Open-source ([itchytoo/deliberation.io](https://github.com/itchytoo/deliberation.io)). Useful: Socratic pushiness levels (1-3), steelman synthesis prompt, placebo/control groups. Confirms market gap: even Stanford+MIT didn't attempt formal argumentation |\n| **Delibera.ai** | Commercial | Multi-perspective AI analysis (legal/finance) | Different market (enterprise), but shows AI+deliberation demand |\n| **ARGSBASE** | Research (EACL 2026) | Multi-agent structured human-AI deliberation | Architecturally very relevant — moderator agent + argumentation theory |\n| **ARGUS** | Open source (Feb 2026) | Multi-agent claim verification via LLM debate | Bayesian reasoning, adversarial evidence, audit trails |\n| **Rebut** | Commercial (2025) | Structured video debates for Gen-Z | Different medium, but AI-powered matching of opposing views |\n| **AI4Deliberation** | EU Horizon project | Government deliberation tools | ARG-tech-led. Significant funding. The institutional play |\n\n### Related Projects Found in Archives\n\n| Project | Location | What it is |\n|---------|----------|-----------|\n| **flowback** | Local archive | Django platform implementing predictive liquid democracy + deliberative decision-making. 422MB. Potentially reusable concepts |\n\n### Historical Predecessors (from Safari bookmarks, 2010-2011)\n\nConcurrent with founding the FB group, Fredrik systematically researched:\n- **Junto** (emergent by design) — Benjamin Franklin-inspired structured discourse project. Open letter to Tim Berners-Lee for help building it\n- **Climate Collaboratorium** (MIT) — deliberation platform for climate policy\n- **Austhink / bCisive** — argument mapping commercial tools (Tim van Gelder's company)\n- **TruthMapping.com** — argument truth-tracking\n- **Hypothes.is** — internet peer review (Kickstarter-funded)\n- **betrireykjavik.is** — Icelandic participatory democracy (found by Tomas Kronvall in the FB group)\n- **Human Advancement Program** (Christer Nylander, Lund, 2015) — a Swedish book-length proposal to \"raise global collective wisdom\" with Sweden as pilot country, *Why and how to upgrade human collective wisdom* (ISBN 978-91-982240-0-9), endorsed by Peter Russell (*The Global Brain* — the pre-Wright noosphere lineage) and Nicholas Maxwell (UCL, Friends of Wisdom). An independent Swedish convergence on the collective-wisdom project from the inner-development side — the same complementarity Björkman's *Nordic Secret* has to Deliberus's structural side. The author corresponded with the project in its 2013-14 era; both his domains are now offline, so the book stands as the citable precedent\n\n## Why Previous Platforms Failed at Scale\n\nThe history of argumentation platforms (Kialo, Debategraph, MIT Deliberatorium, ConsiderIt) reveals consistent failure patterns:\n\n1. **Structured argumentation is effortful** — people find it cognitively expensive\n2. **Low-friction platforms win** (Reddit, Twitter, even Wikipedia) despite structural poverty\n3. **Network effects** — 250 users is insufficient; need critical mass\n4. **Chicken-and-egg** — platform needs content to attract users, needs users to create content\n\n## Deliberus's Differentiated Bet\n\nThe core bet is that **LLMs change the friction equation**:\n- System does the structuring, users just talk\n- Automatic claim extraction from pasted text or conversation\n- Automatic duplicate detection and linking\n- Automatic argument scheme classification\n- The system becomes an \"AI facilitator\" that structures deliberation in real-time\n\n**Additionally**:\n- Multi-dimensional scoring (not just pro/con)\n- Argument bundles as first-class UI objects\n- Explicit fact/value separation\n- Social dynamics awareness\n- Graph-based navigation, not linear threads\n\n## Academic Field: Computational Argumentation\n\nThe academic field closest to Deliberus is **Computational Argumentation**:\n- Conference: COMMA (Computational Models of Argument)\n- Key researchers: Chris Reed (Dundee), Doug Walton (argumentation schemes), Phan Minh Dung (abstract frameworks)\n- Frameworks: Dung's AF, ASPIC+, ADF\n- Datasets: Argument Web, DeliData, Discussion Tracker Corpus\n\n### Key Academic References\n\n- **Dung (1995)**: Abstract Argumentation Frameworks — the foundation\n- **Walton et al.**: Argumentation schemes and critical questions\n- **Mercier & Sperber**: Argumentative theory of reasoning — \"reasoning is argumentative, solitary reasoning is biased, group reasoning works\"\n- **Chris Reed et al. (2017)**: The Argument Web ecosystem\n- **DeliData (2023)**: 64% of groups found better solutions collaboratively\n- **Davies & Gangadharan (2009)**: \"Online Deliberation: Design, Research, and Practice\" (`Deliberus Lecture Notes.pdf`, local archive)\n  - Ch. 25: Online environment for democratic deliberation (design principles)\n  - Ch. 28: Decision Structure (Pingree) — new approach to three problems in deliberation\n  - Ch. 29: Design requirements for argument mapping software (Easterday, Kanarek, Harrell)\n\n---\n\n**See also**: [Vision](vision.md) · [Technical Direction](technical-direction.md) · [History](history.md)\n\n\n## Nanoarguments (Knowledge Pixels, NLnet-funded, April 2026) — the nanopub side colonizes the argument layer\n\nFounder-relayed lead (Aug 2026): [Nanoarguments](https://nlnet.nl/project/Nanoarguments/) is Knowledge Pixels' NLnet-funded (NGI0 Commons Fund) extension of nanopublications into a \"global, federated graph of scientific claims as LinkedData\" — claims, evidence chains and discussion as cryptographically signed nanopubs, with a browsing/editing interface, inline-annotation integration and an ActivityPub bridge. This is the nanopublications lineage ([science-reasoning-infrastructure.md](research/science-reasoning-infrastructure.md) already covers it as one of the input-friction-stalled attempts) now moving one layer up, toward discourse.\n\n**Differentiation, and why the missing-layer claim needs one precision rather than a retreat**: Nanoarguments is scientific-discourse-scoped, RDF/LinkedData-formal (the Semantic Web lineage whose documented killer is maintenance + input friction — and it does NOT dissolve input friction: researchers structure claims themselves), with no LLM extraction, no strength semantics, no value/residue layer, no cross-register ambition. The relayer's own framing: *\"precis lagom långt ifrån deliberus för att bägge ska få plats\"* (just far enough from Deliberus that both fit). The precision the empty-slot claim now carries: the reasoning layer is being approached at its scientific-empirical corner; the general layer (values, weighing, cross-register, strength, honesty instruments) remains unclaimed.\n\n**Two uses beyond differentiation**: it is direct PRECEDENT that NLnet's fund pays for argumentation-graph infrastructure (load-bearing for the September–November NLnet application window), and nanopub export is a candidate interchange format for the fact layer — the \"facts ship first\" leg could interoperate with their network rather than compete with it.\n\n\n## Poliscope / Polity Nova (Suzanne Tempelhof, alpha Aug 2026) — the fast layer's kindred claims instinct\n\nFounder-relayed (Aug 2026). Tempelhof (the Bitnation founder) launched **Poliscope: Cipher** ([politynova.com](https://politynova.com)): a \"narrative intelligence cockpit\" for the live news cycle — structured multi-source feed, a **verification layer where claims and rumours become trackable objects** (\"receipts, denials, confirmations, corrections and unresolved gaps stay attached to the record… it does not decide truth for you; it keeps the evidence trail intact\"), narrative/framing-divergence tracking across outlet blocs, and amplification-network mapping with coordinated-behaviour detection. Motivating case: a distorted UN figure that front-paged worldwide while the correction three days later went unseen.\n\n**Kinship is real and specific**: claims-as-objects with attached evidence trails is the addressability instinct arriving at the minutes-timescale, and her correction-invisibility story is precisely the persistent-record argument (a correction that ATTACHES to the claim's record instead of scrolling past it). **The layer difference is equally specific**: Poliscope works the fast rung — sentiment, framing velocity, amplification forensics, per-source credibility scores (cr: 95) and recommended actions — with no argument structure between claims, no decomposition, no strength semantics, no value layer, and its credibility numbers are unexplained scalars of exactly the class this corpus removed from its own surface in August. Notably it confronts strategy-class adversaries (coordinated amplification) head-on, the encounter our instruments have not yet had. **Complementarity shape**: their narrative tracker detects live disputes → candidate feeder for debate-cluster ingestion; our slow persistent layer holds what their news-cycle layer cannot (the six-month-later record). Ally-scouting context and the person: `.private/` register.\n"}